Three innovative startups are revolutionizing the way travelers handle airfare fluctuations, particularly when prices drop after a ticket has already been purchased, as highlighted in a recent Reuters article published on October 30.

By engaging with the services offered by JetBack, Junova, and Refare, travelers can leverage artificial intelligence to track fare changes and potentially secure refunds or flight credits stemming from price reductions.

JetBack operates on an annual fee of $100 for their service. In contrast, Junova and Refare charge a percentage of the savings they achieve for their clients, with rates set at 20% and 25%, respectively.

Junova revealed that from its monitoring of flights between September 2025 and July, approximately 22% experienced at least one price drop. Meanwhile, Refare provided data indicating that the negotiability of fares varied across five airlines, with 13% to 29% of flights being eligible for price negotiation. Average savings through these negotiations typically ranged from $34 to $396 per flight.

Since its inception in early 2025, JetBack has successfully recovered $2.4 million for travelers, showcasing the practical benefits of their service.

The viability of these business models hinges on airlines maintaining the ability to adjust fares without imposing substantial change fees, as mentioned in the report. In a bid to boost travel in the aftermath of the pandemic, many airlines had waived change fees for certain ticket types.

It's important to note that policies regarding change fees differ between airlines and are subject to change at any time. JetBack noted in its launch announcement in July 2025 that fare volatility has increased significantly with the implementation of AI and real-time pricing strategies among airlines.

According to an August blog post from Junova, numerous flights often undergo multiple price changes, with fluctuations potentially occurring right up until departure. The company's service minimizes the time travelers would otherwise spend tracking prices, making calls to airlines, and attempting to negotiate refunds.

Refare, in its blog post dated October 19, acknowledged that while no outcomes can be guaranteed, their historical data reflects a success rate exceeding 90% in recovering fares from eligible price drops.

Previously, PYMNTS reported in August that airlines are increasingly employing artificial intelligence to swiftly analyze various factors and recommend fare changes. Additionally, as of July, passengers may encounter challenges in discovering affordable seats as airlines enhance their pricing strategies with AI technology.

AI Startups Reclaim Traveler Refunds When Airline Prices Decrease

Generate a cinematic news thumbnail for title:AI Startups Claw Back Traveler Refunds When Airline Fares Drop. No text in image.

AI Startups Claw Back Traveler Refunds When Airline Fares Drop - PYMNTS.com

- Startups JetBack, Junova, and Refare assist travelers in securing refunds after airfare drops.  
- JetBack charges $100 annually, while Junova and Refare take a percentage of savings.  
- Historical data shows significant price drop eligibility, with average savings ranging from $34 to $396.  

Three startups are helping travelers save on airfare when there is a price drop after they’ve already purchased a ticket, Reuters reported Friday (Oct. 30).

By completing this form, you agree to receive marketing communications from PYMNTS and to the sharing of your information with our sponsor, if applicable, in accordance with our Privacy Policy and Terms and Conditions .

Complete the form to unlock this article and enjoy unlimited free access to all PYMNTS content — no additional logins required.

JetBack, Junova and Refare use artificial intelligence to monitor fares and seek refunds or flight credits when airlines cut prices, according to the report.

JetBack charges $100 a year for its service, while Junova and Refare take 20% and 25%, respectively, of the savings when they secure a refund, the report said.

Junova said in the report that among the flights it monitored from September 2025 through July, about 22% had at least one price drop.

Refare shared data that showed that among five airlines, the percentage of their flights that were eligible for negotiation ranged from 13% to 29%, and the average savings for each negotiation ranged from $34 to $396.

JetBack said in the report that it has recovered $2.4 million since early 2025.

We’d love to be your preferred source for news. Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks! Add as Preferred Source

The business model for these services depends on airlines continuing to allow fare drops without charging a large change fee, according to the report.

Many airlines eliminated change fees on select tickets shortly after the pandemic in an effort to encourage people to travel, the report said.

Change fee policies vary by airline and can be altered at any time, per the report.

When JetBack announced its launch in July 2025, the company said in a press release that fares have become more volatile than ever since airlines began using AI and real-time pricing algorithms.

Junova said in an August blog post that many flights experience more than one price drop and that the prices can continue moving right up until the flight departs. The company said its service saves travelers the time they would have spent monitoring prices, calling the airline and trying to negotiate a refund.

Refare said in an Oct. 19 blog post that while it cannot guarantee results, its historical performance includes a better than 90% success rate on recovering fares on eligible, identified price drops.

PYMNTS reported in August that airlines are using artificial intelligence to analyze variables and recommend changes to fares far faster.

It was reported in July that travelers could face more trouble finding low-cost seats as airlines adopt AI-powered pricing practices.

AI Startups Claw Back Traveler Refunds When Airline Fares Drop

Current AI models, whether created by OpenAI, Anthropic, or built on open-source frameworks, face a significant challenge: they cannot learn from their mistakes without additional training. When businesses implement these models in specialized sectors like law and finance, they often resort to fine-tuning general models. This typically necessitates either building internal research teams or enlisting costly consultants, referred to as forward-deployed engineers (FDEs).

Trajectory, a startup launched about a year ago by three Stanford alumni with backgrounds at Google DeepMind and Apple, is tackling this issue head-on. Their goal is to automate the process that enables AI models to learn from user feedback and improve continuously for particular tasks.

Recently, Trajectory announced its success in securing $45 million in a Series A funding round, achieving a valuation of $300 million. This round was spearheaded by Sequoia, with additional investments from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures.

Co-founder Ronak Malde likened foundational models to an exceptionally bright PhD student perpetually starting new roles. "But what the economy values is someone who has been in the field for 30 years," he stated in an interview with TechCrunch.

Trajectory plans to address this limitation through a methodology called continual learning. Organizations can begin with an open-source AI model and integrate the Trajectory platform into their systems, enabling the model to gather feedback on its performance. This data is then utilized to retrain the model, promoting its improvement over time in specialized applications.

This functionality allows professionals in law or medicine to enhance the AI models almost autonomously, which Malde believes could significantly expedite and reduce the costs of implementing specialized AI solutions.

Companies such as Harvey and Rogo have already adopted Trajectory's platform to enhance open-source models, helping them evolve to function more like seasoned lawyers or financial analysts capable of drawing insights from past tasks, such as analyzing legal contracts or utilizing specific methodologies for company evaluations. Other clients using Trajectory include Decagon, Mercor, and Nvidia, and the company is now making its post-training platform broadly accessible.

Users of Trajectory's system have the autonomy to define "what good looks like" for their organization, translating this vision into tangible improvements for their AI models, Malde explained.

While Trajectory's methodology of continual learning shares some concepts with recursive self-improvement (RSI), co-founder Michael Elabd emphasizes the importance of retaining human involvement in the training process. This precaution aims to minimize the risks associated with wholly autonomous self-improvement, which could inadvertently lead to negative consequences.

Trajectory is not alone in its quest to simplify post-training processes; it faces competition from various startups, including Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, as well as Applied Compute, which boasts a recent valuation of $3.25 billion, and Prime Intellect, which assists companies in creating and training their AI agents.

Malde asserts that the greatest strength of Trajectory lies in its team. He, along with Elabd, are among the few researchers whom Google DeepMind acquired from Windsurf, while co-founder Arjun Karanam has expertise in generative foundation models from his time at Apple. The startup has also successfully recruited talent from OpenAI, Stripe, and Netflix.

The significance of design is paramount for a company specializing in AI training, according to Karanam. The team has developed an intuitive interface aimed at making post-training user-friendly for clients who may not have technical backgrounds.

Trajectory secures $45M to enable nontechnical experts to access AI model training.

Generate a cinematic news thumbnail for title:Trajectory raises $45M to make AI model training accessible to nontechnical experts. No text in image.

Trajectory raises $45M to make AI model training accessible to nontechnical experts - TechCrunch

- Trajectory, a startup, raised $45 million for its AI continual learning platform.  
- The platform allows AI models to learn and improve from user feedback automatically.  
- Trajectory aims to make AI deployment faster and cheaper for specialized fields like law.  

Whether developed by OpenAI or Anthropic or built on open source alternatives, current AI models share a critical limitation: They don’t learn from their mistakes without further training.

Businesses adapting these models for specialized fields, such as law and finance, typically get around this by fine-tuning general models. However, that often requires hiring internal research teams or relying on expensive consultants, commonly known as forward-deployed engineers (FDEs).

Trajectory, a startup founded about a year ago by three Stanford classmates with experience at Google DeepMind and Apple, wants to automate that process, allowing AI models to learn from feedback and continuously improve at specialized tasks.

The 18-person company just announced that it raised $45 million in a Series A round at a $300 million valuation, led by Sequoia, with participation from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures.

Trajectory co-founder Ronak Malde compared foundational models to a very smart PhD student who’s always starting a new job. “But what the economy values is someone who worked on the job for 30 years,” he told TechCrunch.

The startup aims to solve this problem through an approach known as continual learning. Businesses using Trajectory can start with an open source AI model and connect the platform to their applications, allowing it to collect feedback on how the model performs. Trajectory then uses that information to retrain the model, helping it improve at specialized tasks over time.

With Trajectory, lawyers, doctors, and other specialists can improve the models near automatically themselves, Malde said. That capability could make deploying specialized AI faster and cheaper.

Companies like Harvey and Rogo are already using Trajectory’s platform to post-train open source models, helping them perform more like experienced lawyers or financial analysts by learning from previous tasks, such as reviewing a particular type of legal contract or valuing a company using a specific methodology. Trajectory’s other customers include Decagon, Mercor, and Nvidia. The startup is now making its post-training platform publicly available.

The people using Trajectory decide “what good looks like for their organization and can turn that into improvements for their model,” Malde said.

Trajectory’s approach to continual learning shares some similarities with recursive self-improvement (RSI), the idea that AI systems can improve their own capabilities. But co-founder Michael Elabd argues that keeping humans involved in the training process makes Trajectory’s approach safer, since autonomous self-improvement could lead to unintended or even dangerous outcomes.

Trajectory isn’t the only startup trying to automate post-training. It considers Thinking Machines Lab, founded by former OpenAI CTO Mira Murati and the company behind the fine-tuning platform Tinker, a direct competitor, along with Applied Compute, which was valued at $3.25 billion last month, and Prime Intellect, which helps companies build and train their own AI agents.

Malde believes the startup’s biggest advantage is its team. He was among the handful of researchers Google DeepMind acqui-hired from Windsurf. Elabd also worked at DeepMind, while Trajectory’s third co-founder, Arjun Karanam, previously worked on generative foundation models at Apple. The startup has also recruited researchers and product designers from OpenAI, Stripe, and Netflix.

Why does design matter so much to a company focused on AI training? Karanam says the startup has developed an interface that makes post-training more accessible to its nontechnical customers.

Trajectory raises $45M to make AI model training accessible to nontechnical experts

The idea of machine unlearning, which involves eliminating specific data from an AI system after it has been trained, was primarily theoretical when Ben Luria, an Israeli entrepreneur, co-founded his startup. However, everything changed last year when China's DeepSeek introduced its first open-weight large language model, prompting Luria to recognize a significant business opportunity in adjusting it to align with Western principles.

Several Chinese open-weight models, including those from DeepSeek, Moonshot, Z.AI, and Alibaba, have gained traction among users, including American firms and startups, due to their cost-effectiveness, robust features, and flexibility. However, concerns about these models being too closely associated with the Chinese Communist Party and their safety compared to closed systems persist.

According to Ben Luria, co-founder of the AI safety startup Hirundo, “We’re not making the model anti-Chinese. We are bringing it up to speed on safety and to the standard of mainstream Western AI.” In an interview with Rest of World, he explained that their objective is to minimize risks for businesses adopting open-weight models. The company employs precision techniques akin to neurosurgery to adjust the models, ensuring they reflect the values, geopolitical perceptions, censorship tendencies, and biases that Western companies expect.

Hirundo's team modifies the model weights to enhance security against cyber threats, promote fairness, and diminish access to hazardous information like bio-weapons data. A report from cybersecurity firm CrowdStrike noted that prompts touching on topics deemed sensitive by the Chinese government were up to 50% more likely to produce code with serious security vulnerabilities. Additionally, another study indicated that code generated by Chinese models was less secure when users identified as U.S. government officials.

Hirundo’s evaluations of Alibaba’s Qwen models revealed their lack of acknowledgment regarding sensitive political topics in China, such as the repression during the Tiananmen Square protests in 1989 and events in Hong Kong in 2019, along with the existence of forced labor camps in Xinjiang. Discussions around the character Winnie the Pooh are also avoided, likely due to its use as an allegory for Premier Xi Jinping among dissidents.

Luria emphasized that Hirundo's adjustments aim to provide the balanced viewpoints similar to those offered by OpenAI's ChatGPT and Anthropic's Claude. He reiterated, “We’re not making the model anti-Chinese. We are bringing it up to speed on safety and to the standard of mainstream Western AI.”

Chinese models are catching up to leading closed systems from U.S. tech labs, trailing by a mere 3% in benchmark scores and attracting considerable download activity on the Hugging Face platform. Notably, companies like Airbnb, Uber, and the AI firm Harvey are generating proprietary models based on these Chinese innovations, while Europe is working on its OpenEuroLLM sovereign AI model using Qwen technology.

Despite their advantages, apprehensions about security remain. The Trump administration recently prohibited civilian government agencies from utilizing DeepSeek products or services, citing national security risks.

Graham Webster, editor-in-chief of the DigiChina Project at Stanford University, echoed concerns about using Chinese models in sensitive contexts. However, he noted that for low-risk applications, such as developing websites for small businesses, the potential issues may be minimal.

In Boston, Lazarus AI is providing on-premise AI services for the U.S. government and other regulated sectors, successfully realigning Alibaba’s Qwen models by removing enforced censorship. The company identified sensitive topics outlined in leaked censorship directives and generated relevant prompts, utilizing responses from Google’s Gemini Flash to recalibrate the models.

Post-intervention, these models began providing information previously suppressed, such as the names of pro-democracy activists, indicating they had initially acquired this knowledge but were restricted from sharing it, according to Eric Hartford, former chief scientist at Lazarus AI.

“In the United States, there is a prevalent anxiety about Chinese models inheriting dangerous behaviors,” he said. He aims to present these models to businesses and heavily regulated sectors in a way that makes them more acceptable.

Lazarus AI’s ReAligned models, promoted as “open-source frontier models free from propaganda,” have been adopted by around a dozen U.S. companies. However, many clients, particularly government entities, are wary of embracing Chinese models despite their superiority over other open-weight versions from OpenAI and Thinking Machines Lab.

Despite claims of effectiveness from companies like Hirundo and Lazarus, Oriane Peter, a Ph.D. researcher in AI at King’s College London, argues that aligning a model to Western values or crafting it to be universally safe is a “near-impossible standard.” She describes alignment as the process of guiding a model’s output to meet specific desired behaviors within a defined cultural context.

Thus, those employing Chinese models beyond China must adapt them, not solely because Western ideologies are inherently safer or more ethical, but due to the necessity of tailoring models to the sociocultural environments in which they will be used.

As the adoption of Chinese open-weight models grows, tools designed to uncensor them and eliminate their censorship mechanisms—referred to as abliteration—are becoming more common. However, many of these tools fall short, with models often still retaining elements of censorship and misinformation, Luria noted.

He pointed out that achieving these alignments is complex and deeply ingrained in the models, making it challenging to eradicate them.

Globally, there is a mounting backlash against Western biases in AI systems. Researchers are observing that even some models from China and Russia default to American-led narratives about historical events. Countries across Asia, Latin America, Africa, and the Middle East are developing their own sovereign AI models designed to be more reflective of their unique languages and cultures.

The centralized control of model alignment in Silicon Valley can risk creating a “massive homogenization effect” in AI systems, as noted by Peter, which has been linked to cognitive uniformity and intellectual decline among users.

Most users of Chinese models seem unconcerned about these alignments, according to Graham Webster. He explained that while caution might be warranted in sensitive regions like Taiwan, low-risk applications like those aimed at small businesses are generally not affected by the underlying issues present in the training data.

Developer Stu Clott, based in San Diego, utilizes DeepSeek for various tasks. He stated that he has yet to encounter situations where the model's censorship posed a problem for him.

Startups adjust Chinese AI for Western markets

Generate a cinematic news thumbnail for title:Startups recalibrate Chinese AI for the West. No text in image.

Startups recalibrate Chinese AI for the West - Rest of World

- Machine unlearning aims to realign AI models to Western values and safety standards.  
- Concerns persist over Chinese AI models potentially harboring risks and state-aligned biases.  
- Models must be adapted to local contexts, as alignment to Western values remains challenging.

The concept of machine unlearning — or removing data from an artificial intelligence system after it has been trained — was largely an academic pursuit when Ben Luria, an Israeli entrepreneur, co-founded his startup. Then, China’s DeepSeek launched its first open-weight large language model last year, and Luria immediately saw the business potential of realigning it to Western values.

A slew of Chinese open-weight models from DeepSeek, Moonshot, Z.AI, Alibabai Alibaba Alibaba, founded in 1999 by Chinese entrepreneur Jack Ma, is one of the most prominent global e-commerce companies that operates platforms like AliExpress, Taobao, and Tmall.READ MORE , and Tencenti Tencent Best known for its super-app WeChat, Tencent is a Chinese technology conglomerate and a major player in the video gaming industry.READ MORE have won over users — including American businesses and startups — with their lower costs, strong capabilities, and customization options. Others fear the models are too aligned with the Chinese Communist Party, and that they are not as safe as closed systems.

“We’re not making the model anti-Chinese. We are bringing it up to speed on safety and to the standard of mainstream Western AI.” Ben Luria, co-founder of AI safety startup Hirundo

“What we’re trying to do is find a way to de-risk the adoption of open-weight models,” Luria, co-founder and chief executive of AI safety startup Hirundo, told Rest of World. “What we do is like neurosurgery, an intervention that’s surgical and very specific. It is to realign the values, geopolitical perceptions, censorship, and state-aligned bias and propaganda to what American and Western companies expect from models they deploy in their workflows.”

Luria’s team edits the weights of open models to harden them and make them more secure against cyberattacks, increase their fairness, and reduce their dangerous knowledge of, for example, bio weapons, he said.

Cybersecurity firm CrowdStrike’s analysis of DeepSeek’s R1 model last year showed that prompts containing topics the Chinese government considers politically sensitive were up to 50% more likely to receive code with “severe security vulnerabilities.” Another study found that Chinese models produced less secure code when the user identified as a member of the U.S. government.

Hirundo’s own analysis of Alibaba’s Qwen models showed they do not acknowledge politically sensitive topics in China such as the violent crackdown on protests at Tiananmen Square in 1989 and Hong Kong in 2019, or the existence of forced labor camps in Xinjiang. They also don’t engage on prompts about Winnie the Pooh, possibly because dissidents have used the cartoon character to refer to Premier Xi Jinping.

Hirundo’s realignment ensures Chinese open models give “the balanced perspective” that OpenAI’s ChatGPT or Anthropic’s Claude give, Luria said.

“We’re not making the model anti-Chinese,” he said. “We are bringing it up to speed on safety and to the standard of mainstream Western AI.”

“More palatable”

The top Chinese models lag behind the leading closed systems of U.S. frontier labs by just 3% on benchmark scores, and account for the biggest share of downloads on hosting platform Hugging Face as more developers use them. Companies including Airbnb, Uber, and legal AI firm Harvey have also built proprietary models on top of Chinese models, while Europe is developing its OpenEuroLLM sovereign AI model on Qwen.

Yet concerns remain. In October, the Trump administration barred civilian government agencies from using products or services from DeepSeek, citing a potential threat to national security.

If I were running a product in Taiwan, I would be more concerned. … But if I’m making a product that helps people make a website for their coffee shop, it doesn’t matter.” Graham Webster, editor-in-chief of DigiChina Project at Stanford University

Boston-based Lazarus AI, which deploys on-premise AI services for the U.S. government and highly regulated industries, realigned Alibaba’s Qwen models by removing their mandatory censorship. It identified the banned topics from leaked censorship guidelines, then generated prompts related to those topics, and used answers produced by Google’s Gemini Flash to realign the models.



After the intervention, the models were able to give information that was not in the training materials, such as names of pro-democracy activists, suggesting the models had originally learned such knowledge and were suppressed from expressing it, Eric Hartford, former chief scientist at Lazarus AI, told Rest of World.



“In the United States, we have this fear of Chinese models potentially having dangerous behaviors baked into them,” he said. “My goal was to … present these models to enterprises and to highly regulated industries and to have [the models] be palatable for them.”



The company’s ReAligned models, billed as “open-source frontier models without the propaganda,” have been deployed at about a dozen companies in the U.S., Hartford said. But other clients, including government agencies, are reluctant to adopt Chinese models even though they are more capable than the open-weight models from OpenAI and Thinking Machines Lab, which they prefer, he said.

“An impossible standard”

While companies such as Hirundo and Lazarus say their method is effective, aligning a model to so-called Western values or making it universally safe “is an impossible standard,” Oriane Peter, a Ph.D. researcher in AI at King’s College London, told Rest of World.



Alignment is about “steering a model’s output toward a specific desired behavior within a well-defined cultural context and application,” she said.



So anyone using a Chinese model outside China must realign it, “not because Western ideology is inherently safer, or more ethical, but because any model must be adapted to the specific social context and environment in which it is meant to operate,” Peter said.



As Chinese open-weight models are widely adopted, there are more tools that claim to uncensor them and remove their refusal mechanisms, a process called abliteration. But many of these tools are not effective, and the models can still retain some of the censorship and misinformation, Luria said.



“It shows that the alignments are deeply embedded, and it’s not that simple to get rid of them.”



Worldwide, there is growing pushback against Western bias in AI systems, with even some Chinese and Russian models defaulting to American geopolitical narratives on historical events, according to researchers. Countries across Asia, Latin America, Africa, and the Middle East are developing sovereign AI models to better reflect their languages and culture.



Silicon Valley’s centralized control over model alignment can lead to “a massive homogenization effect” in AI systems, said Peter. This has been linked to cognitive homogenization and intellectual decline among users.



Most users of Chinese models do not care about their alignment, Graham Webster, editor-in-chief of the DigiChina Project at Stanford University, said at a recent Rest of World event.



“If I were running a product in Taiwan, I would be more concerned because it would be throughput that is potentially relevant to the stuff that the Chinese government has enforced strict control over in the training data and the training,” he said. “But if I’m making a product that helps people make a website for their coffee shop, it doesn’t matter.”



Stu Clott, a developer in San Diego, uses DeepSeek for writing, research, and coding tasks. He does not see the need for a realigned model, he told Rest of World.



“I have not had a single situation where censorship got in my way,” he said.

Startups recalibrate Chinese AI for the West

During its Halloween episode, "Saturday Night Live" took a comedic jab at the tech startup scene and artificial intelligence, rolling out several sketches that highlighted the quirks of entrepreneurial culture. One standout sketch featured Ben Marshall portraying a sinister version of Pennywise, the iconic Stephen King character, who unveils a mock startup named WeAllFloat.ai.

"Our mission? To replace the classic clown tactic of trapping one child with a balloon," Pennywise declared. "Imagine instead deploying hundreds of AI agents, each equipped with their own digital balloons!"

Before this spooky character made his entrance, "Weekend Update" anchor Michael Che also delivered a punchline about the struggling stock performance of Alphabet, Google's parent company. "Today’s Google Doodle hung itself," Che quipped, referencing the playful animations featured on Google’s homepage.

In another humorous skit set in a Latin telenovela, a Sprite representative acknowledged that their new flavor didn’t taste great because "we used AI to develop it," sarcastically claiming, "For the first time ever, AI messes up."

SNL has a history of parodying AI and the buzz surrounding startup culture, and this episode focused notably on Texas, painting it as a hotspot for profit-driven entrepreneurs. Since 2020, over 200 companies have moved their headquarters to the state, attracted by its lenient regulations and business-friendly environment.

However, relocating to Texas doesn't inevitably lead to success. In the skit, even Pennywise faces challenges with his AI venture. "I’ve gathered a crew of cool, like-minded, sober, predatory tech types," Marshall jokes. "I’m currently living in a loft in Austin with three other ambitious young Christians."

'SNL' targets AI, startup culture, and 'exploitative tech executives'

Generate a cinematic news thumbnail for title:'SNL' goes after AI, startup culture, and 'predatory tech guys'. No text in image.

'SNL' Goes After AI, Startup Culture, and 'Predatory Tech Guys' - Business Insider

- "SNL" parodied startup culture and AI with a sketch featuring Pennywise as a startup founder.  
- Michael Che joked about Alphabet's stock decline, humorously implying Google's Doodle "hung itself."  
- A Sprite ad spoof highlighted AI’s failure in product development through sarcastic remarks.  

"Saturday Night Live" took aim at startup culture and AI across multiple sketches in its Halloween broadcast, including a spoof of the stereotypical big-talking startup dude, channeled through Stephen King's Pennywise.

"We started a company — it's called WeAllFloat.ai," comedian Ben Marshall, playing the malevolent clown, said in the segment. "Instead of trapping one kid with one balloon, what if you had hundreds of AI agents, each with their own digital balloons?"

Before Pennywise came on, "Weekend Update" host Michael Che got a joke in about the floundering stock price of Google's parent company, Alphabet.

"Today's Google Doodle hung itself," Che said, referring to the animations Google sometimes uses on its homepage.

And in another spoof of advertising execs trying to plug a new Sprite flavor on a Latin telenovela, a Sprite rep admits the drink tastes bad because "we used AI to make it."

Related video How AI takes a toll on the freelancers who train, test, and refine it

"For the first time in history, AI screwed up," she says sarcastically.

Though "SNL" hasn't shied away from lampooning AI and the buzzy startup culture associated with it before, Saturday's broadcast honed in on the Lone Star State as a place where soulless, profit-driven techies are flocking to start new companies.

Indeed, since 2020, more than 200 companies have relocated their headquarters to Texas. Many cite comparatively lax regulatory laws and a business-friendly atmosphere.

Of course, though, moving down south doesn't guarantee profitability. Pennywise, for instance, might have some trouble getting his AI startup off the ground.

"Got a little crew together of cool, like-minded, sober, predatory tech guys," Marshall says in the segment. "I live in a loft in Austin with three other motivated young Christians."

'SNL' goes after AI, startup culture, and 'predatory tech guys'

"The role of human salespeople in the sales process is evolving, thanks to Coldi," asserts Or Goldshmidt, co-founder and CEO of the managed voice AI startup. 

Established in 2025 alongside Ilia Ronin, Coldi emerged from a social-impact initiative that followed the events of October 7, which aimed to harness technology for substantial societal change. This initiative led to the development of AI-driven solutions and multilingual conversational bots, inspiring its founders to adapt these technologies for the business sector.

Coldi has successfully raised $500,000 in funding and is addressing a prevalent issue within the sales industry: the extensive time sales professionals dedicate to repetitive tasks prior to finalizing a deal.

“Coldi reflects a significant shift in Voice AI,” Goldshmidt explains. “The goal has advanced beyond simply creating an agent capable of natural dialogue. For effective operation within a business context, agents must integrate seamlessly with existing workflows, databases, calendars, telephony systems, and human teams.”

Operating under the name Coldi Labs Ltd, the company operates in the FinTech and AI sectors.

Coldi provides a comprehensive managed voice AI platform tailored for high-volume calling industries such as brokerage, fintech, insurance, and debt collection. Rather than offering a basic toolkit for clients to build their systems, Coldi supplies a complete package that includes AI voice agents proficient in over 30 languages, CRM and calendar integrations, campaign management, and quality monitoring, all managed by the proprietary Campaign Brain system.

The AI agents simplify various tasks: they qualify and reconnect with leads, follow up on onboarding drop-offs, schedule meetings, and transfer promising prospects to human sales personnel. The service begins with a paid pilot, after which clients are billed per minute of usage along with a support fee. Coldi is also ISO 27001 and GDPR certified, acting as a data processor in compliance with a published Data Processing Agreement (DPA). The company currently collaborates with clients across Europe, the Middle East, Latin America, and South Africa, and recently launched Hebrew capabilities in September 2026 for its initial Israeli customers.

The origins of Coldi lie in the founders’ collective experience within the fintech arena. Or Goldshmidt has a rich background, having spent seven years at Panda Trading Systems and previously working at Colmex Pro in Cyprus. As the CEO, he oversees Coldi's strategic direction, development, and client engagement. Co-founder Ilia Ronin has also spent considerable time in sales, having established a trading academy and prop firm in Minsk and previously worked extensively in the BPO industry supporting numerous call centers.

The concept for Coldi was conceived in the aftermath of October 7 as part of a larger social-impact drive using technology for positive change. This endeavor gave rise to AI solutions and conversational bots with multiple language capabilities, leading Goldshmidt and Ronin to consider their application in the commercial domain. They realized that sales teams frequently struggle with repetitive tasks before engaging with prospects, which can waste valuable time. Coldi was designed to automate these initial interactions, enabling human salespeople to concentrate on more complex, relationship-building conversations.

The necessity for Coldi's product is clear: businesses pour significant resources into lead generation, yet sales teams often can't engage swiftly and consistently with all potential clients. Many leads either get delayed responses, remain untouched, or languish in databases for extended periods. Repetitive tasks such as prospecting, follow-ups, and qualifying interest consume considerable time that could be better spent on actual sales discussions. Coldi’s managed Voice AI agents facilitate initial outreach and qualification, enabling human reps to step in at crucial moments where their skills are most impactful.

Coldi’s innovative model signifies a transformation in the sales market, altering when and how human salespeople participate in the sales cycle. By automating traditionally labor-intensive tasks—like prospect outreach, follow-ups, and appointment scheduling—more focus can shift towards the nuanced aspects of sales that require human insight and interaction. Goldshmidt emphasizes that the future of sales does not view AI as a competitor to human talent, but rather as a tool that augments productivity within the existing sales framework.

The market potential for Coldi is expansive, rooted in the growing demand for Voice AI, AI agents, and contact-center automation. According to Mordor Intelligence, the global market for AI within call-center applications is projected to expand from approximately $5.08 billion in 2026 to $13.15 billion by 2031, achieving a compound annual growth rate (CAGR) of around 21%. The overall AI agents market is anticipated to grow even more rapidly, with forecasts suggesting tens of billions in market size by decade's end.

Coldi's primary customers include organizations with substantial volumes of sales leads or customer inquiries, particularly where the bottleneck is human resource capacity. Key sectors encompass performance marketing, financial services, customer onboarding, and international businesses that operate across diverse languages and regulatory frameworks.

Currently, Coldi’s technology is live and functioning effectively in real-world scenarios. In a recent U.S. initiative, the system processed over 10,000 leads and attempted nearly 27,000 calls in just over six weeks, resulting in more than 500 transfers to sales representatives and over 570 meetings scheduled. Similarly, in another deployment for an international payments entity, Coldi engaged with numerous clients across 28 countries. The company's ability to launch its Hebrew language application successfully demonstrates its capacity to scale and adapt.

The Voice AI sector is becoming increasingly competitive, with players like Vapi, Bland AI, Synthflow, Retell AI, and PolyAI entering the fray alongside established companies that integrate AI agents into their communication platforms. Coldi distinguishes itself not only through its voice technology but also by its focus on the operational infrastructure necessary for successful deployment in business contexts. It combines technological proficiency with dedicated service, ensuring clients achieve their desired outcomes through effective AI integration.

Goldschmidt and Ronin offer a wealth of relevant experience that enhances Coldi's value proposition. Their shared history in the fintech realm equips them with a deep understanding of the challenges sales organizations face, ensuring the company's solutions are well-tailored to address practical business needs. Their vision prioritizes leveraging AI to streamline high-volume interactions while reserving the human element for more impactful sales engagements.

In terms of future growth, Coldi is aiming to raise $2 million in a seed funding round to solidify its presence in the Voice AI space for sales and customer engagement. Recognizing the urgency to capitalize on current market dynamics, the funding will be allocated to three key areas: first, enhancing technology and product development by recruiting skilled engineers and AI experts; second, expanding marketing efforts to establish a strong brand presence; and third, strengthening the sales team to manage increasing demand and ensure swift transitions from pilot projects to long-term implementations.

Israeli company Coldi is implementing AI agents to handle cold calls.

Generate a cinematic news thumbnail for title:Israeli startup Coldi is deploying AI agents to take over the cold call. No text in image.

Israeli startup Coldi is deploying AI agents to take over the cold call - CTech

- Coldi's managed voice AI platform automates repetitive sales tasks, enhancing human sales efficiency.  
- The startup has raised $500,000 and is expanding its technology and marketing efforts.  
- Coldi targets high-volume calling industries, offering over 30 languages and complete system integration.  

“The change Coldi represents is in where the human salesperson enters the sales process,” says Or Goldshmidt, CEO and co-founder of managed voice AI startup Coldi.

Founded in 2025 by Goldshmidt and his Co-founder Ilia Ronin, Coldi was inspired by a post-October 7 social-impact initiative that sought to use technology to create meaningful impact. During that time, AI-based solutions and multilingual conversational bots were developed, leading the founders to explore how those same capabilities could be applied to the business world.

1 View gallery Coldi Co-Founders Or Goldshmidt & Ilia Ronin. ( Photo: Coldi )

Having raised $500,000 in funding so far, Coldi now seeks to answer a recurring problem in the sales world: the significant time salespeople sink into repetitive work that happens before the actual sale.

“Coldi's approach also reflects a broader evolution in Voice AI,” Goldshmidt continues. “The challenge is no longer simply building an agent that can hold a natural conversation. To work in a real business environment, the agent must operate as part of a complete system that connects to company workflows, databases, calendars, telephony and human teams.”

Company Name: Coldi Labs ltd

Sector: FinTech, AI

Product/Service description:

Coldi is a managed voice AI platform that runs outbound and inbound phone campaigns for regulated, high-volume calling industries: brokers, fintech, insurance and debt collection. Instead of handing clients a toolkit and asking them to build around it, Coldi delivers the full stack: AI voice agents in over 30 languages, telephony, CRM and calendar integrations, campaign management and quality monitoring, all run by our proprietary orchestration layer, Campaign Brain.

Our agents qualify and re-engage leads, call back onboarding drop-offs, book meetings and live-transfer hot prospects to human sales teams. Clients start with a paid pilot and then pay per minute of usage plus a support fee. Coldi is ISO 27001 and GDPR certified and acts as a data processor under a published DPA. We are live with clients across Europe, the Middle East, LATAM and South Africa, and in September 2026 we launched Hebrew in production with our first Israeli customers.

Founder Bios:

Or Goldshmidt, CEO and Co-founder, is a veteran of the fintech industry. He spent seven years at Panda Trading Systems and before that worked at Colmex Pro in Cyprus. He leads Coldi's strategy, development and client delivery.

Ilia Ronin, Co-founder and commercial, comes from Colmex Pro (Limassol), where he and Or worked together in the brokerage industry. Later on he established the first trading academy and prop firm in Minsk, Belarus. Before joining Coldi he worked for six years in the BPO industry serving hundreds of call centers.

Year of Founding: 2025 Last Investment Round: $500,000 Last Investment Stage: 01/01/2026 Date of Last Investment: $500,000 Total investment to date: $500,000 Investors: Gil Group, David Gal (Angel) Current number of employees: 15 Open positions: 2 Website: www.Coldi.ai Social Media: LinkedIn

How was the idea born?

The story of Coldi began after October 7, as part of a social-impact initiative that sought to use technology to create meaningful impact. During that activity, AI-based solutions and multilingual conversational bots were developed.

The experience gained through that initiative led founders Or Goldschmidt and Ilia Ronin to explore how the same capabilities could be applied to the business world.

The two had previously worked together in fintech and were familiar with sales-intensive environments. They identified a recurring problem: salespeople spend a significant part of their time on repetitive work that happens before the actual sale, including reaching prospects, following up with leads, reactivating old databases, qualifying interest and scheduling meetings.

Coldi was created to automate that layer, allowing AI to handle volume and repetitive first-touch interactions while human salespeople focus on conversations that require understanding, persuasion, judgment and closing.

What is the need for the product?

Companies invest significant resources in generating leads and attracting potential customers, but sales teams often do not have the capacity to contact every lead quickly and consistently.

As a result, some leads receive a late response, remain untouched altogether or sit in company databases for months or years. At the same time, skilled salespeople spend a substantial part of their working day trying to reach prospects, following up and filtering leads instead of focusing on actual sales conversations.

Coldi addresses this gap with managed Voice AI agents that can contact new or existing leads, conduct an initial conversation, follow up, identify interest, schedule a meeting or transfer a relevant prospect to a human representative in real time.

The voice conversation itself is only one part of the product. Behind it is an operational system that manages campaign logic, telephony, retry attempts, lead status, CRM and calendar integrations, regulatory calling restrictions, phone-number reputation and the handoff between AI and human teams.

How is it changing the market?

The change Coldi represents is in where the human salesperson enters the sales process.

Tasks that traditionally required significant human capacity, such as reaching prospects, repeated follow-ups, initial qualification, database reactivation and appointment scheduling, can increasingly be handled by AI at scale.

This allows human salespeople to enter the process when the conversation requires skills that remain distinctly human: understanding the customer's needs, handling objections, persuasion, judgment and closing.

Coldi's approach also reflects a broader evolution in Voice AI. The challenge is no longer simply building an agent that can hold a natural conversation. To work in a real business environment, the agent must operate as part of a complete system that connects to company workflows, databases, calendars, telephony and human teams.

Coldi therefore does not position AI as a replacement for strong salespeople. Its premise is that AI will increasingly take over the repetitive layer around sales and change what salespeople spend their time doing.

How big is the market for the product and who are its main customers?

Coldi operates at the intersection of Voice AI, AI agents and contact-center automation, so the size of the opportunity varies depending on how the category is defined.

Mordor Intelligence estimates the global market for AI in call-center applications at approximately $5.08 billion in 2026, with projected growth to $13.15 billion by 2031, representing a CAGR of approximately 21%.

The broader AI-agents market is growing even faster, with estimates projecting tens of billions of dollars in market size by the end of the decade.

Coldi's main target customers are companies with significant volumes of sales leads, customer inquiries or onboarding activity, particularly where human capacity creates a bottleneck. These include performance-marketing companies, sales organizations, fintech and financial-services companies, customer-onboarding operations, call centers, companies with large historical lead databases and international businesses operating across multiple languages, time zones and regulatory environments.

Does the product exist already?

Yes, Coldi is already operating in live production environments.

In one U.S. implementation, the system handled 10,967 leads and 26,857 dialing attempts over six and a half weeks. The activity resulted in 521 live transfers to sales specialists and 577 meetings booked directly into calendars, creating 1,098 opportunities for continued human sales conversations.

In another deployment, Coldi is operating for an international payments platform serving customers across 28 countries.

Coldi has also deployed its technology in Hebrew in Israel. In a seven-day campaign, the system dialed 2,923 unique leads, conducted 1,992 live conversations and returned 172 leads to the client for further human follow-up.

The company is therefore beyond the prototype or proof-of-concept stage and is already deploying and operating its system with customers in real-world environments.

Who are the main competitors in this sector and how big are they?

The Voice AI market is fast-growing and increasingly competitive. Relevant AI-native companies include Vapi, Bland AI, Synthflow, Retell AI and PolyAI, alongside larger contact-center and communications companies that are integrating AI agents into their platforms.

Coldi's differentiation is not based solely on the voice model itself. The company focuses on the implementation and operational layer required to make Voice AI work in production, including campaign logic, telephony, lead management, integrations, compliance, number health and human handoff.

It combines the technology with a managed service, taking responsibility for implementing and operating the system around the business outcome the customer wants to achieve.

What is the added value that the founders bring to the company and the product?

Goldschmidt and Ronin bring a combination of experience in fintech, trading, sales and business environments that are directly relevant to the problem Coldi is addressing.

The founders first worked together in the fintech industry.

Ilia comes from a fintech, business and trading background and studied business administration and trade in Vienna. His experience gave him familiarity with commercial and sales-driven financial environments.

Or later spent several years at a company that developed technology infrastructure for brokers and investors, including CRM systems, trading-related technology and other tools used by financial businesses. His work there included sales, marketing and project management. He subsequently spent approximately three years working at a bank.

Together, their backgrounds gave them first-hand exposure to the operational reality of sales-driven businesses and to the amount of time and resources spent on the stages that precede the actual sale.

That experience helped shape Coldi around a practical business problem rather than around Voice AI technology alone: how to use AI to handle volume and repetitive customer interactions while ensuring that human salespeople enter the process where their skills create the greatest value.

What will the money coming in from the round be used for?

Coldi is raising a $2 million seed round to turn its early traction into a leading position in Voice AI for sales and customer engagement. Speed is critical in this market. Companies are choosing their Voice AI providers now, and those that move slowly will find it much harder to win customers once the market matures. The funds will be invested in three main areas:

1. Technology and product: Coldi will recruit experienced engineers and AI specialists to accelerate development of its platform. The focus will be on conversation quality across languages, deeper CRM, telephony and calendar integrations, and the operational layer that lets companies run AI agents at scale.

2. Marketing: The Voice AI category is still taking shape, and the window to establish a leading brand is limited. Coldi plans to substantially expand its marketing, including content, industry events, digital campaigns and partnerships, to reach as many customers as possible in its core verticals and regions.

3. Sales: Coldi will grow its sales team to meet rising demand and shorten sales cycles. It will also build out customer success, so that customers move quickly from pilot to full, long-term deployment.

Israeli startup Coldi is deploying AI agents to take over the cold call

This narrative is derived from a chat with Rachel Woods, the CEO and co-founder of AMP, and has been condensed for clarity and brevity.

Rachel's journey into the world of artificial intelligence began with her internship at Facebook, where she transitioned into a full-time role in 2018 focused on AI research and development for the company’s advertising system. Despite lacking a traditional engineering background—with a degree in management information systems from The University of Texas at Austin—her hands-on experience with personal projects that involved training AI models and conducting data analysis was pivotal in securing her position.

In 2020, Rachel ventured out from Facebook to launch a business in the wine sector, which she later sold. Following that success, she established an AI-focused enterprise. Rachel emphasizes the need for effective AI playbooks, stating that without them, her team would have had to double in size.

She notes a common misconception: many overlook how to design AI systems that can operate autonomously without causing sleepless nights for their operators. 

Business Insider aims to connect with founders of small AI-driven companies with fewer than ten employees and those collaborating with AI agents, seeking insights into what the era of "Tiny Teams" looks like, including triumphs, challenges, and the essential human skills that emerge in this landscape.

Since the age of 15, Rachel has embraced entrepreneurship, starting with a photography business, despite being too young to drive herself to client meetings, which often required rides from others—something her parents disapproved of.

Her interest in technology blossomed during college, where she witnessed peers engaged in data analytics and coding at internships. This exposure prompted her to seek online courses and realize that breaking down business challenges could enable AI to address components such as prediction and sorting.

While applying for Facebook, she utilized AI and data analysis to compile a report on coffee shops in Austin optimal for her study sessions, a project that likely contributed to her recruitment.

Post-Facebook, Rachel founded Vinebase, a wine business that deployed AI to assist winemakers in selling their products online while managing the digital aspects of their operations. At that time, AI was not mainstream, so they marketed their product as an "auto magic" system to avoid alienating potential clients. Vinebase was acquired by Corksy in fall 2022.

Currently, Rachel co-leads AMP—initially named The AI Exchange—alongside her sister. The company was founded before the debut of ChatGPT in November 2022, as Rachel recognized the rapid advancement in AI technologies during her time at Vinebase. After ChatGPT's launch, interest in their platform surged.

Presently, AMP employs four full-time staff and four part-time contractors, but Rachel asserts they would need to hire twice as many personnel without their reliance on AI. With their current structure, they would require additional roles including a sales manager, customer support specialist, executive assistant, and operations coordinator.

AMP's operational approach hinges on what they term "playbooking." Each department within the organization maintains its own system comprised of playbooks that can function independently or in tandem. To develop an AI playbook, they focus on a repetitive task, outlining its initiation, execution, and expected outcomes before detailing it as if training a new employee. This approach facilitates an ongoing refinement of the playbook until it becomes reliable enough to function autonomously.

The company has integrated an AI agent to manage their sales playbook, streamlining the review of customer interactions, thus freeing the sales manager from manual evaluations. Team members now consult the playbook for operational inquiries, which automatically retrieves information from internal guides.

AMP incurs two main types of AI costs: subscriptions to AI services like Claude and ChatGPT for each team member and operational expenses associated with their background-running AI playbooks. For example, in September, they spent $8,510 on Claude Code to manage these playbooks efficiently.

Their mantra, "own the playbook, rent the tech," underscores the importance of developing robust playbooks that are adaptable to various technological platforms—be it Claude Code, ChatGPT, or others—allowing flexibility despite rapid tool changes. 

To mitigate the risks of uncontrolled AI, they prioritize thorough documentation of processes. Human oversight remains essential to design and dictate the execution of work, with AI serving to enhance productivity rather than replace human roles.

Rachel recommends initiating AI workflows with manageable tasks. Once teams experience their first positive outcome from automating mundane tasks, the momentum can lead to broader implementations, creating an encouraging cycle of innovation and productivity.

I manage an 8-person startup utilizing AI agents, and we have developed a structured playbook system instead of relying on intuition.

Generate a cinematic news thumbnail for title:I run an 8-person startup that uses AI agents. We don't let our agents run on vibes — here's the playbook system we created. No text in image.

I run an 8-person startup that uses AI agents. We don't let our agents run on vibes — here's the playbook system we created - Business Insider

- Rachel Woods, CEO of AMP, transitioned from Facebook to founding AI-focused businesses.  
- AMP uses AI playbooks to streamline operations and reduce staffing needs significantly.  
- Woods advocates for creating detailed playbooks to ensure effective AI operation and oversight.  

This as-told-to essay is based on a conversation with Rachel Woods, CEO and cofounder of AMP. The following has been edited for length and clarity.

My first official role in AI was at Facebook. I started as an intern and then joined full-time in 2018. I worked on AI research and development for the ads system that's powered by AI.

I don't have an engineering degree or background; I studied management information systems at The University of Texas at Austin. Part of why I think I got my role at Facebook was because I spent time building my own side projects and teaching myself how to train AI models and do data analysis.

I left Facebook in 2020 to start my own business in the wine industry, and after that company sold, I founded an AI company. I would need to double the number of people on my team if we didn't use AI playbooks.

If you're being thoughtful about setting up your AI operations, it's no different than building a team that you can trust.

Related video The CEO of a startup helping non-techies code still sees a lot of value in engineering degrees

That's something a lot of people are missing: how to actually make these systems things you can take your eyes off and not keep yourself up at night in the process.

Business Insider wants to talk to founders of AI-powered companies with fewer than 10 workers, and employees working alongside AI agents and bots, to understand what the "Tiny Teams" era really feels like — the wins, the fears, and the human skills that stand out. Share your story by filling out this quick form.

I've been an entrepreneur since the age of 15

My first business was a photography business, and I was too young to even drive to meet my clients. I would always catch rides with people, and my parents hated this.

I got into tech because in college, I was exposed to colleagues doing data analytics and working with coding and AI at summer internships. That led me to look online and try to take related classes, and I started to see that if you break a business problem into small enough pieces, you can get AI to handle parts of it: predicting, sorting, generating.

Later, when I was applying to Facebook, I used AI and data analysis to build a report of coffee shops in Austin that were less likely to be too busy for me to study at. I think that project helped me land my role.

After I left Facebook, I started a wine business called Vinebase in 2020; we built a tech platform that used AI to help winemakers sell their wine online and manage the digital side of their business.

Because this was before AI became more mainstream, we didn't market it as AI. We called it our "auto magic" system. I think that if we had said it was AI, it could've been a turnoff for our customers. Vinebase was acquired by Corksy in fall 2022.

Business Insider has Games! Want more Business Insider in your news feed? Add BI in Google so our reporting is easier to find when you’re searching for what matters. Add to Preferred Sources

I use my playbooks and AI agents to replace additional staff

My sister and I started AMP, originally called The AI Exchange, before ChatGPT was released in November 2022, because I was using AI in Vinebase and saw how the technology was progressing. ChatGPT came out about a month later and started really accelerating the AI timeline, so we went all in.

We have four full-time employees and four part-time contractors right now, but we'd need double if we weren't running as much as we are on AI. Given the company's current size, we'd need at least four to five additional people, including a sales and marketing manager, customer support specialist, executive assistant, and operations coordinator.

The way we structure these systems is based on something we call playbooking. Every department in the company has its own system. Each system is a collection of playbooks with AI agents that can run independently or together.

How we build our AI playbooks

To build an AI playbook, pick one piece of work you do over and over, and write down what starts it, what goes into the task, and what should come out at the end. Next, write the steps the way you'd train a new hire: each step, the decisions in it, and what "good" looks like.

Start running it with AI, and while you watch, compare the result to how you would've done it. Then, fix the playbook, not the output, until you trust it to run without you.

I always think through how I want a playbook to work and outline it by hand. Whenever you're making an AI playbook, you want to own the thinking and the logic. Otherwise, you can fall into the trap of including something that you don't really understand, and then you're not going to be able to trust that system anyway.

We have an AI agent running our sales playbook to review customer calls and ensure they meet our standards, rather than having a sales manager review them manually. Also, instead of sales team members needing to ask their manager questions, they just ask the playbook, which then pulls the answers from our guide for that team.

There are two types of AI costs we have in the business. The first is the subscriptions to Claude and ChatGPT for every person on the team. The other is because the majority of our operations are these AI playbooks that are running in the background without us being the ones starting or triggering it. In September, we spent $8,510 on Claude Code to run our playbooks in the background, which is billed by usage and is separate from the team's monthly Claude and ChatGPT subscription costs.

Own the playbook, rent the tech

We say "own the playbook, rent the tech," which means that if you are creating really good playbooks for your AI to follow, it actually doesn't matter which tool you're running it in.

Right now, we have a setup that's running with Claude Code in the background. But I could run all of these playbooks in ChatGPT, Grok, Gemini, or Microsoft Copilot.

We try to avoid a lot of the risks that people are concerned about with AI going rogue by creating these detailed outlines. We don't just want the agents to run on vibes. There still needs to be a person to decide and design how the work gets done; AI just speeds up everything else. That's the job security we have in this AI era.

Start small, and get everyone involved

A few of us on the team who are more advanced with AI own the bigger, team-wide playbooks that the business depends on, but everyone playbooks their own personal workflow, and then we use that as the testing ground to graduate things into becoming more part of official systems.

My advice would be to just start with something small and really easy, because once you get your first win and see how AI can do that repetitive work for you, it's kind of like a snowball.

I run an 8-person startup that uses AI agents. We don't let our agents run on vibes — here's the playbook system we created

In 2023, Silicon Valley was in the midst of a significant evolution, driven by major tech firms striving to emulate OpenAI's triumph with ChatGPT and Twitter's rebranding to X under Elon Musk. Meanwhile, nearly 3,000 miles away in New York City, two young filmmakers were grappling with these changes and sought to capture their reflections on the current state of technology—so they decided to make a movie.

"It was deeply personal," shared 25-year-old James Hyland with Business Insider. "We aimed to create something that resonated with our experiences and the world around us. This journey led us to themes of artificial intelligence and the disconnection it breeds."

Together with Myles Flusser, age 26, Hyland co-created "Anna by the Window," a coming-of-age drama that delves into the complexities of grief and the impact technology has on human relationships during an age defined by generative AI. The pair wrote, directed, and produced the film in 2023, collaborating with friends through their production company, Baggysuit.

"Anna by the Window" navigates themes of growing up, emotional loss, and the challenges of connecting in a tech-dominated landscape. Flusser expressed concern about how the leaders of platforms like YouTube and Instagram appear unfazed by the encroachment of AI-generated content. "It fascinates and frightens me at the same time," he admitted.

As digital natives, Hyland and Flusser have witnessed firsthand the effects of smartphones and social media on their generation. Recently, the US Surgeon General recommended that social media platforms carry warning labels due to their mental health consequences. Earlier this year, Meta settled allegations from several states claiming their platforms were designed to addict minors.

This overwhelming digital landscape has prompted some young people to gravitate towards neo-Luddite movements, advocating for a return to genuine, in-person interactions.

Hyland is involved with a neo-Luddite group called The Lamp Club. While "Anna by the Window" wasn't directly influenced by this movement, it reflects similar concerns.

"There’s a shared vision between our film and this larger movement growing across the nation, where individuals are increasingly frustrated with technological systems," Hyland noted.

The film crew started screening "Anna by the Window" in July. It tells the story of Ryan, a teenager who ventures to New York City to reconnect with his AI startup founder brother, James. The project was filmed across New York State and incorporated various innovative techniques, including the use of ElevenLabs AI audio software and segments of a real ChatGPT chatbot that underwent "hallucinations" during production.

Hyland described the AI chatbot in the film as playing a "quirky" role, often veering into nonsensical territory, resulting in it delivering multilingual utterances and errors. Flusser articulated their desire to authentically reflect the unsettling aspects of AI interactions, although they preferred traditional filming methods for the bulk of the project, relying on a Sony camera, scripted outlines, and improvisation.

"The peculiarities and sense of alienation prompted by chatbots and AI were central to what we aimed to convey," Flusser said.

While parts of the narrative are fictional, the directors infused personal experiences into their work. Flusser revealed that his brother's recent passing influenced the film's emotional depth. "There’s a pivotal moment between the two brothers that echoes my own life experiences," he revealed. However, he clarified that the character of James is not a direct representation of his brother, describing him instead as someone struggling with reality and using technology in a misguided attempt to find solace.

The cinematic landscape is increasingly interested in the lives of tech entrepreneurs, with films featuring figures like Elon Musk, Sam Altman, Mark Zuckerberg, and Elizabeth Holmes. "Anna by the Window" now joins these narratives that scrutinize the tech industry.

"When we began filming, we were curious about AI, but by the end, our perspective turned quite cynical," Flusser said.

Throughout the filmmaking process, both Hyland and Flusser experienced the surreal evolution of AI. They originally viewed their film as speculative fiction about artificial intelligence. "Now, it feels like reality has caught up with us, and what was once science fiction seems all too real," Flusser reflected.

The "Anna by the Window" team began showing the film in summer in New York City and the Hudson Valley, with more screenings scheduled in Los Angeles this November. The film will be available for streaming on Dolph starting December 15.

This upcoming indie movie centered around an AI startup founder and his family increasingly resembles reality.

Generate a cinematic news thumbnail for title:This new indie film about an AI startup founder and his family feels less like sci-fi by the day. No text in image.

This new indie film about an AI startup founder and his family feels less like sci-fi by the day - Business Insider

- Independent filmmakers James Hyland and Myles Flusser created "Anna by the Window" in 2023.  
- The film explores grief and human connection amid the rise of generative AI.  
- "Anna by the Window" will stream on Dolph starting December 15, 2023.

From tech behemoths scrambling to replicate OpenAI's success with ChatGPT to Twitter rebranding to X under Elon Musk, Silicon Valley was undergoing an undeniable transformation in 2023.

Nearly 3,000 miles away in New York, two skeptical 20-somethings were trying to make sense of it all. So, they made a film.

"It was very personal," James Hyland, 25, told Business Insider. "We wanted to put something together that spoke to the things we were going through and what the world was like. That ended up being artificial intelligence and technological alienation."

Hyland and Myles Flusser created "Anna by the Window," a coming-of-age film exploring grief and how technology can stifle human connection in the era of generative AI. The independent filmmakers wrote, directed, and produced the film in 2023 alongside their friends under their production company, Baggysuit.

"Anna by the Window" is a coming-of-age film exploring grief and human connection in the digital era. Baggysuit

"Particularly, it feels like the people running these platforms — YouTube, Instagram, et cetera — seem more contented than ever in letting it become a total slop apocalypse full of AI bots and non-human content," Flusser, 26, said. "That intrigues me in a way and scares me, obviously."

Related video We filmed his AI relationship for a year

Zoomers like Flusser and Hyland grew up inundated with smartphones, social media, and all the risks that come along with them. A year after they filmed "Anna by the Window," the US Surgeon General said social media should have a warning label because of its impact on mental health. Earlier this year, Meta settled a lawsuit brought by a coalition of states accusing it of designing platforms to addict children.

The din of constant digital exposure has led some young folks to join the neo-Luddite movement, which takes a stand against overreliance on modern technology by emphasizing IRL human connection.

Hyland is an organizer for a Luddite group called The Lamp Club. While the film was not made because of the Luddite movement, it shares many of its themes.

"I think that it speaks to the synergy of our separate film project and this larger organizing project that's materializing all over the country, where people are fed up with these kinds of technological systems," Hyland said.

When fiction becomes reality

The "Anna by the Window" crew began screening the film in July. Baggysuit

"Anna by the Window" is a glimpse inside the world of Ryan, a teenage boy who travels to New York City to reconnect with his older brother, James, the founder of an AI startup.

The film's crew shot across New York State and used a variety of tools to bring the project to life, including ElevenLabs AI audio software. The pair also pushed a real ChatGPT chatbot toward hallucination and included that audio in the film.

Follow Lauren Edmonds Every time Lauren publishes a story, you'll get an alert straight to your inbox! Stay connected to Lauren and get more of their work as it publishes. Sign up By clicking "Sign up", you agree to receive emails from Business Insider. In addition, you accept Insider's Terms of Service and Privacy Policy. Business Insider has Games! Want more Business Insider in your news feed? Add BI in Google so our reporting is easier to find when you’re searching for what matters. Add to Preferred Sources

"There's an AI chatbot that goes off the rails, you could say. We really cranked up all the dials," Hyland said. "It started saying 'error.' It started speaking different languages."

Flusser said they wanted to portray the eerie nature of AI just right. They're not eager to use AI products in future projects, though. The majority of the film relies on traditional tools like a Sony camera, written outlines, and improv.

"There's a strangeness and alienation that come up when we interact with chatbots and AI. We were trying to capture that feeling as purely as possible, so it felt totally necessary to use AI to do that," Flusser said.

James Hyland and Myles Flusser used Eleven Labs and ChatGPT to create parts of the film. Baggysuit

While aspects of the film are fictional, others pull directly from the directors' lives.

"My brother had recently died. There's a lot of stuff in there that has to do with that," Flusser said. "There is a climactic moment between the two brothers in this movie that is actually surgically removed from my own life and recreated on screen."

However, the way the filmmakers chose to characterize James in the film — delusional, stunted, running from reality — is distinct from Flusser's real brother. In the film, James is portrayed by Hyland, who said his version of an AI startup founder isn't based on a specific person.

"My character was in a state of denial. Using this technology to be the thing that could deliver him closure or be the thing that he could connect with," Hyland said. "It makes him, in a way, kind of a sympathetic character."

'Anna by the Window' on the silver screen

Hollywood is fixated on the real lives of tech founders. Production companies have backed films about SpaceX founder Elon Musk, OpenAI CEO Sam Altman, Meta founder Mark Zuckerberg, and former Theranos CEO Elizabeth Holmes. Even "Toy Story 5" tackled tech by introducing a smart tablet as the story's antagonist.

"Anna by the Window" now joins the list of films putting the tech industry under a harsh microscope.

"When we first started filming this, we were more mystified by AI, but toward the end our relationship with AI felt very cynical," Flusser said.

"Anna by the Window" will become available to stream on Dolph on December 15. Baggysuit

Watching AI evolve while creating a film about the technology was surreal, Hyland and Flusser said.

"It was strange. When we were making it, we saw this movie as a speculative sci-fi about artificial intelligence," Flusser said. "Now, it's like nothing really is sci-fi about it anymore."

The "Anna by the Window" crew began screening the film this summer in New York City and the Hudson Valley. There are upcoming screenings in Los Angeles this November. The film becomes available on Dolpho, a streaming platform, beginning December 15.

Find upcoming "Anna by the Window" screenings here.

This new indie film about an AI startup founder and his family feels less like sci-fi by the day

An AI venture co-founded by Jeff Bezos is reportedly nearing a lease agreement for a building in San Francisco. Amidst a significant downturn in the city's office market, the Sobrato Organization made a bold investment of over $80 million to acquire the former headquarters of Gap, located at the waterfront. Their prediction that the building could cater to a new wave of tech tenants is now on the verge of realization, as Prometheus, Bezos’s AI startup, appears set to occupy the refurbished property at One Harrison. This deal could validate Sobrato's investment and underscore how the burgeoning AI sector is reshaping corporate preferences in office environments.

Real estate sources have informed the Chronicle that Prometheus is finalizing a lease for all six floors of the 180,000-square-foot building, conveniently situated near Google’s offices by the Embarcadero. Previously, the well-funded startup had been leasing a smaller, 30,000-square-foot space in the 101 Mission St. tower, but has silently been searching for a more expansive location in recent months.

While Prometheus representatives did not provide comments on the potential lease or its intended use, it’s evident that the new space would significantly enhance the startup's presence in San Francisco and fulfill Sobrato's aspiration to secure a primary tenant at One Harrison.

On Friday, Sobrato refrained from confirming any lease with Prometheus or disclosing the status of a potential deal.

Founded in 2025 by Bezos and former Google X executive Vik Bajaj, Prometheus specializes in developing AI technologies for engineering and manufacturing, aiming to integrate AI solutions into the tangible world. The startup launched late last year with $6.2 billion in funding and recently achieved a valuation of $41 billion after raising $12 billion. Its prestigious investors include JPMorgan Chase, Goldman Sachs, and BlackRock, alongside Bezos and others.

Prometheus is focused on creating software that automates the design and production of various complex items, ranging from airplanes and computer chips to pharmaceuticals. This summer, the company secured a 100,000-square-foot lease for a warehouse situated in an industrial area of West Oakland.

Historically, One Harrison was owned by Gap for over 20 years, most recently serving as the base for Athleta. The building features a contemporary design, characterized by its glass and concrete facade, and has been recently painted white to stand out amid the nearby brick structures. Notably, a corner office on the fourth floor that was once occupied by late Gap co-founder Don Fisher has been redesigned to enhance the surrounding workspaces.

Gap's departure from One Harrison coincided with a broader trend of companies reassessing their office needs due to shifts in workplace dynamics ushered in by the pandemic. In 2021, Gap announced it would relocate Old Navy employees from Mission Bay to streamline its workforce at the Embarcadero. By 2023, however, the company, while still headquartered locally, sought to sell One Harrison, reducing its presence in San Francisco.

The revamped One Harrison now boasts an expanded lobby featuring a living green wall and a cafe for tenants, along with a new top-floor bar and upgraded office spaces designed to foster collaboration.

During a recent tour of the transformed property, Sobrato’s co-chief real estate officer, Chase Lyman, highlighted the unobstructed views of the Bay from every floor, the newly created roof deck adjacent to the Bay Bridge, and an upgraded HVAC system designed for maximum efficiency. Lyman also emphasized the building’s robust electrical infrastructure, an essential factor for AI companies that require significant power to develop and support hardware-intensive technologies like robotics.

Notably, the renovation did not include a gym, and the design limited hardwall construction to provide flexibility in how the office space could be utilized.

Reflecting on Sobrato's investment, Lyman acknowledged moments of uncertainty regarding whether the market would respond favorably but reiterated the organization's long-term vision for its portfolio.

“We are looking at this over 30, 50 years,” he stated.

Despite the ongoing challenges in San Francisco’s real estate market, characterized by high office vacancy rates, the AI industry has sparked renewed activity. In the third quarter of this year, the city experienced five consecutive quarters of positive absorption in office space, with the vacancy rate decreasing to 27.7%. Companies such as OpenAI, Anthropic, and Databricks have all committed to substantial leases.

With a wealth of available office space, well-funded AI startups are increasingly in search of large, adaptable headquarters that can be customized to their specific requirements, including the security of exclusive occupancy. Moreover, the demand for research and development spaces is rising, as businesses seek to move beyond traditional office configurations to innovate and test products related to the physical world. Even firms well-known for their software solutions are delving into AI applications that necessitate hardware, although suitable properties featuring high ceilings, structurally sound floors, and sufficient electrical capacity remain scarce in San Francisco.

Exclusive: Jeff Bezos' AI company is nearing a significant lease in a former Gap office in San Francisco.

Generate a cinematic news thumbnail for title:Exclusive: Jeff Bezos’ AI startup is closing in on a huge S.F. lease in a former Gap office. No text in image.

Exclusive: Jeff Bezos’ AI startup is closing in on a huge S.F. lease in a former Gap office - San Francisco Chronicle

- Jeff Bezos's AI startup Prometheus reportedly leased all six floors of One Harrison.  
- The property, previously owned by Gap, is undergoing renovations to attract tech tenants.  
- San Francisco's AI boom is increasing demand for larger, flexible office spaces for startups.

An AI startup by Jeff Bezos is said to be close to a lease for a building in San Francisco. Kent Nishimura/AFP via Getty Images

At the depths of San Francisco’s office market downturn, the Sobrato Organization paid more than $80 million for Gap’s one-time corporate office at the waterfront, betting it could be remade for a new generation of tech tenants. Now, the refreshed property at One Harrison is said to be landing Prometheus, Jeff Bezos’s AI startup — a deal that could validate the risky gamble and highlight how the city’s AI boom is changing what companies want from office space.

Multiple real estate insiders have confirmed to the Chronicle that Prometheus has leased all six floors of the 180,000-square-foot building, located near Google’s offices at the Embarcadero. The heavily funded startup already leases a smaller, 30,000-square-foot office inside the nearby 22-story 101 Mission St. tower, but has been quietly on the hunt for new office space in recent months.

Representatives of the startup have not responded to inquiries on the potential One Harrison lease, or how the space would be used. What is clear is that the building would give Prometheus a much larger San Francisco base and deliver the single headquarters tenant that Sobrato told the Chronicle it hoped to attract to One Harrison.

Advertisement Article continues below this ad

On Friday, Sobrato neither confirmed a lease with Prometheus nor said whether a deal was pending.

Co-founded in 2025 by Bezos and former Google X executive Vik bajaj, Prometheus is developing AI tools for engineering and manufacturing — an effort to apply AI to the physical world. It launched late last year, backed by $6.2 billion in funding, and this summer raised $12 billion at a $41 billion valuation. Its primary backers include JPMorgan Chase, Goldman Sachs, and BlackRock, alongside Bezos and others.

On a mission to build software that would automate the design and manufacturing of things like jet planes, computer chips and drug compounds, the startup signed a 100,000 square foot lease at a warehouse in an industrial pocket of West Oakland, this summer.

For more than two decades, the clean, glass and concrete structure of One Harrison was owned by the Gap, most recently using it as the headquarters for Athleta. Its sleek corporate facade has been painted white, and stands apart from the brick and stone buildings nearby. A fourth floor, corner office that was once occupied by the late Gap co-founder Don Fisher has been opened up to the surrounding workspace.

Gap’s exit from the property came amid a broader retreat in the city’s offices in the wake of the pandemic, as remote work reshaped companies’ space needs. The company announced plans in 2021 to move Old Navy employees out of Mission Bay and consolidate its local workforce at the Embarcadero. But, by 2023, Gap, which remains headquartered in the area, was looking to sell One Harrison as it decidedly shrank its San Francisco footprint.

Advertisement Article continues below this ad

Today, One Harrison’s expanded lobby features a living green wall and a cafe for tenants. The renovation also added a top-floor bar and new office and collaboration spaces throughout the building.

During a tour of the property earlier this month following its makeover, Sobrato co-chief real estate officer Chase Lyman touted its unobstructed Bay views from every floor, a roofdeck carved out of what once was top floor office space next to the belly of the Bay Bridge, and a newly electrified HVAC system — but also singled out the building’s access to “good power” as a proud feature.

“The power we have here and the location are all pluses,” he said, alluding to heavy electrical capacity AI companies require, including to develop and operate hardware intensive technology like robotics.

A gym was purposely omitted from the renovation and hardwall construction limited, as Sobrato did not want to dictate how the building’s floors would be used.

There were moments his team questioned whether the market “would agree with us,” Lyman said about Sobrato’s investment in One Harrison, but added that the organization’s approach to its portfolio is holding properties for the long term.

Advertisement Article continues below this ad

“We are looking at this over 30, 50 years,” he said.

While San Francisco’s real estate market is still struggling with significant office vacancy, the AI boom has driven recent momentum — the city recorded five quarters of positive office space absorption in the third quarter of the year while vacancy dropped to 27.7%, with companies like OpenAI, Anthropic and Databricks having committed to major leases.

With so much office space available, well-funded AI startups are increasingly looking for large, flexible headquarters that they can shape around their needs, including the privacy of occupying a building on their own.

The AI boom is also driving demand for research and development space, as companies look beyond conventional offices to build and test products tied to the physical world. Even companies best known for software are exploring AI applications that involve hardware, sources told the Chronicle. But suitable buildings — with high ceilings, floors strong enough to support heavy equipment and ample power — are scarce in San Francisco.

Advertisement Article continues below this ad

Exclusive: Jeff Bezos’ AI startup is closing in on a huge S.F. lease in a former Gap office

Andy Elder and Arjun Mathur, former roommates at Stanford who both pursued computer science, have reunited after a decade-long tenure at Palantir to embark on an entrepreneurial journey. Elder was instrumental in leading product design for an extensive team responsible for developing software for U.S. defense and intelligence agencies, while Mathur held the position of senior vice president within the company’s commercial division. When they decided to launch a startup early last year, partnering together was a natural decision.

The duo was inspired by an alarming trend: the construction of new buildings in the U.S. has diminished over recent decades, largely due to skyrocketing development costs, despite advancements in technology. 

“We’ve evolved from dial-up internet to superintelligent AI, yet our capacity to construct the physical world seems to have dwindled,” Mathur expressed in an interview with TechCrunch.

Believing that artificial intelligence could enhance productivity in the architecture, engineering, and construction (AEC) sectors, they founded Dream, a startup focused on creating models that facilitate automated building design and engineering. Recently, Dream has emerged from stealth mode, securing $17 million in seed funding led by Andreessen Horowitz, with additional investments from SV Angel, XYZ, and Zigg Capital. The goal of Dream is to transform outdated AEC tools such as Autodesk's AutoCAD and Revit.

According to Mathur, many architects and structural engineers are still engaged in traditional design practices, relying on legacy software to meticulously produce each part of their projects. “AI has been largely absent from this scene,” he noted.

While typical large language models like ChatGPT can generate building images, these outputs often lack structural integrity. Dream, however, builds its framework on genuine physics and established construction standards. Just as LLMs are trained on textual data, Dream's world model learns from an extensive dataset exceeding 100 million square feet of building information sourced from permit archives throughout the United States.

“Our model comprehensively grasps the specifics within a structure—what materials are involved, the composition of the beams, and the weight they can bear,” Elder explained.

The startup has coined the term “structural cognition” for its approach, highlighting that its model produces structurally accurate designs that consider essential internal components such as plumbing and HVAC systems.

Dream has initiated sales to AEC professionals and developers, who utilize the model for initial building design concepts and cost estimations. The platform is already in operation on ten projects across Texas and California.

Though Dream faces competition from emerging companies like Motif, Snaptrude, and Arcol—each attempting to modernize AEC software with AI—Elder and Mathur assert that they uniquely offer a world model capable of designing entire buildings instantly.

Kimberly Tan, a partner at Andreessen Horowitz, remarked on the notable progress Dream has made since the firm's initial investment a year ago, when their startup model was merely conceptual. “I have not encountered anyone else adopting this technical approach,” she noted. “They’re creating everything based on their proprietary models.”

Former Palantir executives start Dream with $17M funding from a16z to create AI world models for construction.

Generate a cinematic news thumbnail for title:Ex-Palantir execs launch Dream with $17M from a16z to build AI world models for construction. No text in image.

Ex-Palantir execs launch Dream with $17M from a16z to build AI world models for construction - TechCrunch

- Stanford roommates Andy Elder and Arjun Mathur co-founded Dream to address declining U.S. construction.  
- Dream's AI model automates building design, replacing legacy tools with structural cognition technology.  
- The startup has secured $17 million in funding and is active in Texas and California.  

Andy Elder and Arjun Mathur were Stanford roommates who both studied computer science before spending a decade at Palantir. Elder led product design for a team of 800 people building software for U.S. defense and intelligence agencies, while Mathur served as a senior vice president in the company’s commercial practice. When the friends decided to build a startup early last year, there was no question they would do it together.

Despite having no direct background in architecture, Elder and Mathur were troubled by a growing problem: Fewer buildings are constructed in the U.S. today than a few decades ago, as development costs have surged.

“We’ve gone from dial-up internet to superintelligent AI, but our ability to build the physical world around us has actually decreased,” Mathur told TechCrunch.

Believing AI can boost productivity across the architecture, engineering, and construction (AEC) industry, they set out to build Dream, a startup developing built world models that automate building design and engineering. Dream has emerged from stealth with $17 million in seed funding led by Andreessen Horowitz, alongside other backers, including SV Angel, XYZ, and Zigg Capital. The startup aims to replace cumbersome, outdated AEC tools like Autodesk’s AutoCAD and Revit.

If you sit next to an architect or a structural engineer, you’ll notice that they are still “drawing” every piece of their designs part by part on legacy software, Mathur said. “AI really plays no role in that equation.”

While a standard LLM like ChatGPT can generate an image of a building, the result won’t be structurally sound. Dream, by contrast, is grounded in real physics and construction standards. Just as LLMs learn from text, Dream’s world model is trained on more than 100 million square feet of building data gathered from permit archives across the United States.

“Our model actually understands what’s within those walls. What materials are used, what the beams are made of, and how much weight they can support,” Elder said.

The startup calls its approach “structural cognition” because its model generates structurally precise designs that account for internal building elements like plumbing and HVAC.

Dream has already begun selling its product to AEC professionals and developers, who use the model to conceptualize building designs and estimate overall costs, according to the founders. The platform is already deployed across 10 active projects in Texas and California.

While Dream isn’t the only company attempting to update AEC software with AI — other startups working on this include Motif, Snaptrude, and Arcol — Elder and Mathur claim that they are the only ones that developed a world model that instantly designs whole buildings.

Kimberly Tan, a partner at Andreessen Horowitz, said that while the firm first invested in Dream over a year ago, when the startup model was just an idea, she has been extremely impressed with their progress. “I have not seen anybody take the technical approach,” she said. “They’re doing everything based on their own models.”

Ex-Palantir execs launch Dream with $17M from a16z to build AI world models for construction

Chuyue Sun, a researcher from Stanford, along with her team of computer scientists, has created an AI tool capable of completing an engineering task in under half a day for less than $500. In stark contrast, three highly skilled experts took five months to accomplish the same project manually. This significant decrease in both time and expense highlights early findings related to a lesser-known area in coding and mathematics called formal verification, as detailed in Sun's recent research paper.

For years, software developers have aspired to leverage mathematical proofs to ensure their programs are free of bugs. However, this intricate process has historically been time-consuming, requiring engineers to articulate precise specifications for their software and translate their code into checkable mathematical formulas. Although companies conduct sophisticated tests to uncover a multitude of software flaws, formal verification—which can identify even more subtle issues—remains a rarely utilized practice due to its perceived complexity.

A surge of startups is now positioning generative AI as a game-changer in making formal software verification both feasible and indispensable. One such company is &Truth, which Sun joined in August and officially launched last Thursday. Her research showcased the capability of this technology by using AI to validate a segment of the code responsible for securing conversations in the messaging platform Signal.

Ilya Sergey, chief science officer at &Truth, stated, "Bridges and buildings remain structurally sound because mathematical principles were applied during their design. Why not adopt the same approach for code? The tools for mathematical reasoning were always available, but scaling them was a challenge."

&Truth has successfully secured $70 million in investments and aims to eventually expand its verification capabilities across all programming languages. In competition, Harmonic—co-founded by Robinhood CEO Vlad Tenev—has raised $295 million and recently revealed that its AI agent, Aristotle, can assist in verifying software written in C and hardware designs in Verilog. The startup is currently testing its technology with potential clients like chipmakers Tensordyne and Nvidia, whose venture arm has invested in the company.

Recently, Anthropic and Theorem collaborated to leverage AI for detecting bugs in open-source software that had eluded detection despite extensive testing. They also launched an add-on for Claude, aimed at aiding developers in the verification process, though cost details for this initiative have not been disclosed.

The overarching hope is for a future where software functions more reliably and incurs fewer security vulnerabilities. However, the challenge remains: can AI extend the reach of formal verification beyond critical systems at major organizations like NASA and Apple to widespread adoption?

The Frustrating Math That Might Rescue Vibe Coding From Its Own Issues

Generate a cinematic news thumbnail for title:The ‘Pain in the Ass’ Math That Could Save Vibe Coding From Itself. No text in image.

The ‘Pain in the Ass’ Math That Could Save Vibe Coding From Itself - WIRED

- Stanford researcher Chuyue Sun developed an AI tool completing projects in under half a day.  
- Startups aim to make formal software verification practical using generative AI technology.  
- AI's potential promises more reliable software with fewer security issues, but scalability is uncertain.  

Stanford researcher Chuyue Sun and her fellow computer scientists developed an AI tool that completed an engineering project in less than half a day for under $500, compared to the five months it took three highly skilled experts to do the same work manually. The dramatic reduction in time and cost, which Sun reported in a recent paper, is among the early results that are awakening an obscure field of coding and mathematics known as formal verification.

For decades, developers have envisioned using math to prove that their software is bug-free. This process is long and laborious, requiring engineers to write precise statements about the goals of their software and convert their code into formulas that can be validated. While companies already run sophisticated tests to identify as many flaws in their software as possible, they rarely shell out for formal verification, which can catch even more erratic behavior.

Now, a wave of startups is betting that generative AI will make formal software verification both practical and essential. Among them is &Truth, which Sun joined in August and which publicly launched on Thursday. Sun’s research demonstrated the technology’s potential by using AI to verify part of the code that encrypts conversations in the messaging app Signal.

“Bridges, buildings, they don’t collapse, because people did math before constructing them,” says Ilya Sergey, &Truth’s chief science officer. “Why don’t we do the same for code? The technology to mathematically reason was all there, just difficult to scale.”

&Truth has raised $70 million from investors and aims to eventually verify any programming language. Rival Harmonic, which was cofounded by Robinhood CEO Vlad Tenev and has received $295 million in funding, announced this week that its math-focused AI agent Aristotle supports verifying software written in C and hardware designs in Verilog. It is evaluating the technology with potential users, including chipmakers Tensordyne and Nvidia, whose venture capital arm is one of the startup’s investors.

And just last week, Anthropic and the verification startup Theorem jointly announced they had used AI to assist in finding bugs in open source software that trillions of tests had failed to detect. They also introduced an add-on for Claude to help other developers with verification. Costs of the initial project were not disclosed.

The big promise is a future where software runs more reliably with fewer security headaches. But whether AI can bring formal verification beyond the most critical systems at organizations like NASA and Apple and into widespread use remains an open question.

The ‘Pain in the Ass’ Math That Could Save Vibe Coding From Itself

The surge in artificial intelligence is generating immense wealth and attracting record levels of investment, yet it is also creating challenges for startups trying to benefit from the current excitement.

On Monday, Anthropic made headlines by initiating a groundbreaking IPO that could potentially secure up to $100 billion. That same day, SpaceX and Apollo unveiled a substantial $40 billion financing initiative focused on chips.

While these monumental financial movements may signal a positive outlook for AI, the reality is that rising borrowing costs are leading to a stark divide among companies. Some are easily securing funding, while others face significant risks to their survival.

SpaceX CEO Elon Musk took to X to explain that the company's investment-grade credit rating and diverse revenue sources, such as Starlink, enabled access to loans. However, he warned that many companies lacking substantial revenue beyond AI data centers will struggle. "Recent weeks have seen a near-total shutdown of debt market access for firms without strong non-AI revenue," Musk stated, cautioning about potential "death and consolidation" among less established firms.

With interest rates, inflation, and computing expenses on the rise, the landscape has shifted away from a thriving market for all things AI. According to investors from Wellington Management, we are entering a phase where performance outcomes will vary significantly across businesses.

Matthew Witheiler, head of late-stage growth at Wellington, indicated at a recent event that this isn't necessarily a death knell for AI as a whole; rather, it represents significant challenges for businesses that have relied heavily on debt to finance their operations. "In today's market, we see substantial disparities between the successful and struggling firms," he added.

Losers: Heavily Indebted Infrastructure Firms

The companies at the greatest risk are those in the AI infrastructure sector burdened with significant debt. According to Jeff Courey, a global industry analyst at Wellington, the ability of these businesses to construct additional capacity may be adversely affected by high financing costs. Until financial conditions improve, these firms may need to reassess their budget limitations.

Witheiler acknowledged that some of these infrastructure investors might possess a "capital structure that isn't conducive to meeting demand and may face heightened challenges."

Winners: Leading Labs and Hyperscalers

In contrast, businesses that dominate compute demand find themselves in a much stronger position. Their established size and growing borrowing costs afford them considerable advantages. Courey noted, "It’s tough to envision a new competitor breaking into this space," given the resources required to catch up. Meanwhile, firms like Anthropic are reporting impressive revenue figures, with $16.3 billion in the first half of 2026 compared to just $1.2 billion the previous year.

Even less favorable data on AI adoption, such as a recent study indicating only the "Mag Seven" saw improved profitability from AI, can still be perceived positively for leading enterprises as wider adoption increases overall demand.

Winners: Rapidly Expanding AI Startups

For AI-driven businesses showcasing significant growth, obtaining financing remains feasible. Witheiler pointed out that companies achieving growth rates of 150% to 1000% annually at meaningful revenue scales are well-positioned in this environment and easily attract funding in private markets.

Even if a company is cash flow negative, as long as it invests in R&D or sales personnel expecting future returns, it can still access funding, per Courey's observations. He noted that for many use cases, the primary challenge is less about token costs and more about effectively selling products.

Losers: Unprofitable Firms

Conversely, companies registering slower growth rates, between 10% to 25%, and struggling to break even face dire financing challenges. These businesses compete against rapidly expanding firms for funding, resulting in a highly unfavorable position. "If you're not profitable and growing, your value plummets," Witheiler emphasized.

Losers: Non-Defensible Startups

Importantly, simply incorporating AI into a startup's operations does not guarantee fundraising success. Those demonstrating rapid growth without a solid foundation may find it difficult to secure funding. Companies that merely layer a user-friendly interface over existing major AI models lack the defensibility necessary to attract investors. Witheiler urged caution, stating that it’s critical to evaluate whether such firms represent a sustainable business or merely a passing trend. "Is it just a temporary fad that will fade after a brief spike in popularity, or does it have the potential to evolve into a viable public company over several years?" he posed as essential questions for potential investors.

Wellington Investors' Strategy for Identifying AI Successes and Failures

Generate a cinematic news thumbnail for title:How 2 Wellington Investors Are Picking AI Winners and Losers. No text in image.

As AI costs rise, here's who will win and who will lose - Business Insider

- Anthropic's IPO could raise $100 billion, alongside SpaceX's $40 billion financing deal.  
- Rising borrowing costs are creating a divide between thriving and struggling AI companies.  
- Fast-growing AI startups secure funding easily, while slower firms face significant challenges.

The AI boom is minting billionaires and attracting historic amounts of investment, but it's also becoming harder for startups to cash in on the frenzy.

On Monday, Anthropic filed for a historic IPO, which could reportedly raise up to $100 billion. The very same day, SpaceX and Apollo announced a $40 billion chip financing deal.

These historic capital outlays may seem like a bullish sign for AI, but rising borrowing costs are creating a divide between companies that can easily attract capital and those that may struggle to survive.

SpaceX CEO Elon Musk said in an X post that the company was able to borrow because of its investment-grade credit rating and non-AI revenue streams, such as Starlink. Others won't be so lucky, he said.

"As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn't have substantial revenue independent of AI data centers," Musk said, threatening "guppy-level death and consolidation" for neoclouds.

Interest rates, inflation, and the cost of compute are all rising. It's no longer a bull market for anything and everything that calls itself AI. Two investors at asset manager Wellington Management say this new era will see differentiated results for businesses.

It's not that the high cost of capital is "damning for the AI industry," Matthew Witheiler, head of late-stage growth at Wellington Management, said at an event Tuesday; it may be "damning for a specific type of business in the AI industry that used a lot of debt to fund their product bill."

"In today's world, there are a lot of haves and have-nots," he later added. Here's how it might shake out.

Loser: Infrastructure companies loaded with debt

The most obvious companies that will be hit hard by elevated borrowing costs are infrastructure investors with large debt loads.

Business Insider has Games! Want more Business Insider in your news feed? Add BI in Google so our reporting is easier to find when you’re searching for what matters. Add to Preferred Sources

The businesses that make up the AI "infrastructure stack" will be "ones that will be more directly impacted if the marginal amount of capacity doesn't get built because of financing costs," said Jeff Courey, a global industry analyst at Wellington covering tech firms in the public market.

Jeff Courey, global industry analyst at Wellington Management. Wellington Management

Until borrowing costs go down, these companies may have to "figure out what they can afford," he said.

Witheiler agreed that it "could be true" that some of the neoclouds and other infrastructure investors have "the wrong capital structure to fulfill a lot of the demand, and they may be more challenged."

Winner: Frontier labs and hyperscalers

Those with all of the compute demand on the other side of the deal are in a much better place. Their massive size and an increased cost of borrowing mean they have quite a moat.

"It's hard to imagine a new player showing up," Courey said, noting the amount of money it would take to catch up. At the same time, these companies are seeing their revenue start to pour in, as highlighted by Anthropic's $16.3 billion in revenue in the first half of 2026, up from $1.2 billion a year earlier.

Even data points that seem negative for AI adoption, such as a recent finding from Apollo economist Torsten Slok that only the Mag Seven saw increased profitability from AI, could be seen as bullish for these companies if you look at it the right way.

Matthew Witheiler, head of late-stage growth, Wellington Management Wellington Management

"Of course, the early adopters of technology are technology companies, the Mag Seven," Witheiler said, as the other major S&P companies adopt the tech. All that means is that there's "so much more of the economy to impact," he said, driving more demand for these companies.

Winner: Fast-growing AI startups

If you need to borrow money, having an AI-focused business with prolific growth is key. AI companies that are growing 150% — or even better 1000% — year over year, at "real revenue scales," are the "haves" in this market, Witheiler said.

"Those businesses easily get financed in the private market," Witheiler said.

Even if you're cash flow negative, if you're spending on research and development that should soon pay off, or paying for salespeople who will sell your products, you won't have any worries finding cash, said Courey.

"Those kinds of cash flows are very understandable to the market," he said.

And for the "vast majority" of use cases, the actual factor holding companies back is their ability to sell their products, not the cost of tokens, Courey said. That cost is not going up nearly as fast as tokens could.

Loser: Companies that don't make a profit

If you're growing more slowly, call it "10% to 25% revenue growth," and you're breaking even or losing some money and need to raise more money or debt, "that's a really tough place to be in."

These companies have "no way" to get money because they're "competing" for dollars against companies growing 500% instead of 20%, Witheiler said.

"If you are losing money and not growing, you're worth nothing," Witheiler said.

Losers: "Wrappers" or flash-in-the-pan companies

Just because a startup uses AI doesn't mean they're going to have an easy time fundraising. If you have "rapid, rapid growth," but no defense or durability for that growth, you could have a hard time fundraising.

Companies that merely create a wrapper, or a graphical and operational interface over another company's large language model, are "not super defensible" and therefore not worth leaning into, Witheiler said. There are important questions to ask before believing the hype.

"Is it a flash in the pan and it's going to be really cool for two quarters and tons of people are going to use it and then it's going to fall off because it's a fun experiment that people got burnt out on, or is it something that in four years from now can be a stand-alone public company?" Witheiler said.





AI Startups Reclaim Traveler Refunds When Airline Prices Decrease

More from AI Startups & Investments

Trajectory secures $45M to enable nontechnical experts to access AI model training.

Startups adjust Chinese AI for Western markets

'SNL' targets AI, startup culture, and 'exploitative tech executives'

Israeli company Coldi is implementing AI agents to handle cold calls.