Chuyện gì đã xảy ra
Business Insider reports that Valon, a New York-based mortgage-servicing software company with about 320 employees, now requires most new hires to work without AI until their managers believe they can recognize when an AI system is wrong. Engineers are exempt because their code is peer-reviewed before release. CEO and cofounder Andrew Wang said the policy began after employees received broad access to AI tools.
Business Insider reports that Valon introduced the policy in July after giving employees broad access to AI systems. Wang said he reviewed usage and found employees selecting the most expensive models even for simple assignments. He told the outlet that employees commonly justified this behavior by saying the AI was almost always right.
Wang interpreted that response as evidence that some workers were becoming less likely to question outputs or develop the judgment needed to identify errors. The policy therefore applies to new hires in nearly every part of the business, including senior recruits, rather than only to junior employees. Engineers are excluded because their code undergoes peer review before release, while Wang said functions such as finance and human resources lack comparable safeguards.
According to Business Insider, new employees may begin using AI only after their manager is confident that they understand their responsibilities and can detect incorrect AI-generated work. The article says the change has pushed new recruits to seek help from more experienced colleagues instead of relying on AI to solve problems. Wang told the outlet that he believes this is helping them develop a deeper understanding of their jobs. The company’s annualized spending is also projected by Wang to fall to roughly $4 million to $5 million this year, from approximately $15 million to $20 million. Those figures are management estimates reported by Business Insider, not independently audited results.
Business Insider also reports that Wang acknowledged the policy’s apparent contradiction with Valon’s business: the company builds mortgage-servicing software powered by AI agents and describes itself as being at the frontier of AI use. Wang said tenured employees had welcomed the measure because they had been correcting what he characterized as low-quality AI-generated work from new hires. The report says Wang encountered some criticism from AI enthusiasts after discussing the policy on LinkedIn, but he said he had not seen internal resistance. The source does not identify the number of affected employees, the specific AI systems involved, the length of the restriction, or a formal review process for lifting it. The policy and its outcomes are not independently confirmed here.
Chi tiết nguồn: businessinsider.com ↗
Tại sao nó quan trọng
The policy reflects a concrete workplace response to concerns that can reduce early-career learning, increase review burdens, and encourage employees to accept plausible outputs without sufficient scrutiny. It also shows that some AI-focused companies are treating human judgment and role-specific knowledge as prerequisites for responsible AI use.
The practical issue is not simply whether AI can produce a correct answer. Business Insider’s account centers on whether employees understand the work well enough to evaluate that answer. New hires who outsource basic tasks immediately may complete assignments faster while missing the underlying procedures, assumptions, and warning signs that experienced workers use to catch mistakes. Valon’s policy represents one company’s attempt to sequence AI adoption: first build task knowledge, then introduce automation under managerial supervision. That approach is especially relevant in roles where work may not receive the same technical review applied to software code.
The report also connects AI use with a less visible organizational cost: review and cleanup. Wang said established employees had been dealing with AI-generated work from new hires, and Business Insider cited a September 2025 BetterUp and Stanford Social Media Lab survey in which 40% of 1,150 full-time U.S. desk workers said they had received AI-generated work from a colleague during the previous month. Respondents said handling each instance took nearly two hours on average. The survey does not measure Valon specifically, and the source does not establish that its results apply to all workplaces, but it provides context for why an employer might view unchecked AI output as a productivity problem rather than a productivity gain. The reported spending estimates add a second dimension. If Valon’s figures are accurate, restricting model access for new hires could sharply reduce consumption, particularly when workers use expensive models for routine tasks.
That does not prove the policy is economically superior: savings could be offset by additional manager time, slower onboarding, or heavier reliance on experienced colleagues. Nor does the report show that lower usage produces better work. The significance is that is being framed as a training and accountability decision, not only as a question of tool availability or model capability.
Cơ chế tương tác: Nó thực sự hoạt động như thế nào
Khám phá công nghệ cơ bản đằng sau sự phát triển này một cách tương tác.
crm_get_transaction(id='4092').Why can ethical evaluation not be reduced to one model score?
Xem gì tiếp theo
The key unanswered questions are whether the policy improves employee performance, how long new hires remain restricted, and whether the reported reduction in AI spending is sustained. Business Insider did not independently verify Valon’s internal usage data, spending projections, or claimed effects on employee understanding.
The first test will be whether Valon can measure a benefit beyond lower spending. Useful indicators would include onboarding time, error rates, rework, manager review hours, employee retention, and performance after AI access is granted. Business Insider reports Wang’s assessment that new recruits are gaining a deeper understanding, but it provides no comparative data, formal experiment, or independent evaluation. Without those measures, it remains unclear whether the policy improves learning or mainly shifts work from software systems to senior employees.
The policy’s boundaries will also matter. The source says most new hires are covered and engineers are exempt, but it does not explain how managers assess readiness, whether contractors or existing employees face similar limits, or whether restrictions apply to every AI task. It is also unknown whether employees may use AI for research, drafting, translation, or administrative work before approval, and whether the policy will remain in place after onboarding. Those details would determine whether Valon has created a durable governance model or a temporary training rule.
A broader question is whether other companies adopt comparable limits as concerns about AI-generated workplace material spread. Business Insider reports that some outside AI enthusiasts criticized Wang’s approach, while Valon’s employees reportedly supported it. That reaction is anecdotal and cannot establish wider sentiment. The most meaningful follow-up would be evidence from Valon or other employers showing when AI access helps new workers learn, when it undermines skill development, and which review controls are sufficient outside software engineering. Until such evidence is available, Valon’s policy is best understood as a notable company-level experiment rather than a general prescription for managing AI at work.