At the AI firm Valon, a fresh policy mandates that most newly hired employees learn their roles through traditional methods, eschewing AI tools in the process.
CEO and cofounder Andrew Wang explained that this initiative was implemented last month after a comprehensive analysis of how employees were utilizing AI, which had been widely accessible. He observed that team members frequently resorted to the most advanced, and costly, AI models for straightforward tasks. Wang expressed concern that this practice not only incurred unnecessary expenses but also hindered the development of fundamental skills among new employees.
Engaging with Business Insider, Wang, who previously worked as an analyst at Goldman Sachs, emphasized the importance of foundational work for cultivating a deeper understanding of job responsibilities. "Performing the basic tasks, instead of leaning on AI, helps in building comprehension," he remarked.
Headquartered in New York, Valon employs approximately 320 individuals and specializes in mortgage servicing software powered by AI agents. In a recent blog post outlining the new guideline, Wang recognized the contradiction inherent in such a decision for a company leading the way in AI innovation.
He noted the necessity of this policy, particularly in light of the responses he received when querying staff on their reliance on powerful AI models for basic assignments. Many indicated that they believed the AI was almost invariably correct, leading Wang to worry that employees were becoming less discerning about the AI's conclusions and less capable of identifying errors.
The policy is aimed at newly hired personnel across nearly all departments, including senior-level recruits. They are restricted from employing AI until their manager is assured they possess the ability to recognize when AI outputs may be flawed. Engineers are exempt from this rule, as all code is subject to peer review prior to deployment. However, roles in finance and human resources do not enjoy similar safeguards.
Valon was valued at $1.75 billion in 2024 and has secured $275 million in venture funding from prominent investors such as Andreessen Horowitz and WestCap. Wang reported that, thus far, there has been no opposition to the new directive; in fact, many experienced employees have embraced the change, having previously dealt with the inaccuracies generated by newer team members' AI work.
This challenge extends beyond Valon. A survey conducted in September 2025 by BetterUp, in collaboration with Stanford’s Social Media Lab, revealed that 40% of 1,150 full-time U.S. desk workers had received AI-generated work from a colleague within the previous month, with each incident requiring nearly two hours on average to address.
The issue has become so widespread that tech insiders have coined the term “meat proxies” to describe individuals who circulate unchecked AI outputs. This phrase gained traction following a blog post by German software developer Niklas Gruhn on August 3.
According to Wang, Valon’s policy is already yielding results. He anticipates a significant reduction in annual token spending, projecting it to fall to around $4 million to $5 million this year, down from approximately $15 million to $20 million. Additionally, new hires are increasingly seeking assistance from more seasoned colleagues due to the restriction on AI reliance, thereby fostering a better understanding of their roles.
While some AI enthusiasts outside of Valon have criticized this policy, Wang remains open to feedback. Responding to detractors on LinkedIn, he called for alternative suggestions, saying, "If you have a much better idea on how to facilitate learning, please share."
