AI startup Valon bans new hires from using AI (and some experienced workers too) in a bid to encourage in-depth understanding and stronger, defensible career growthgettyAn AI start-up just announced something that sounds insane at first.It asked new employees to learn their jobs.Valon removed AI access for new hires (and experienced employees) across the company, except for the engineering team, who must peer-review their work anyway. This measure will last until a line manager agrees that the employee is developed and confident enough to use AI tools responsibly.“Deep understanding comes from hard, grungy, low-level work. You can’t get it from reading a spec,” Andrew Wang, CEO and co-founder, along with CPO Jake Mintz, wrote on a company blog. “You have to live and breathe it through hand debugging a reconciliation error, reconstructing all the steps of a mistaken money movement, or tracing a bug through multiple edge cases.”Learning your job sounds like a ridiculously simple ask, right? Well, not really, when you consider that many professionals use AI blindly and mindlessly without stopping to consider the consequences of poor AI output (also known as workslop or AI slop).Workslop Is Becoming One Of The Biggest Workplace Problems In The U.S.Workslop is indeed becoming a real workplace crisis. About 40% of 1,150 U.S. employees said they received AI slop within the last month, per a BetterUp Labs and Stanford Social Media Lab’s survey.And PwC’s recent catastrophe, where they were caught red-handed passing off AI slop in four thought leadership articles and reports (with the investigation discovering fabricated footnotes and questionable citations), is the latest in a string of high-profile AI slop sagas with big consulting firms, with KPMG, Deloitte, and EY no stranger to AI hallucinations going unchecked, leading to negative relations with their clients, partial refunds being issued, and retracted reports.And this is just a handful of examples. I could share dozens more.“Each workslop incident costs $186/month per affected employee,” BetterUp said in their report. “Scaled across a 10,000-person organization — and accounting for the 40% prevalence rate — that’s $9 million per year in lost productivity.”By contrast, Valon’s AI policy is projected to slash the startup’s annual AI token spend by about 75%, from $15-20 million down to $4-5 million this year. That shows just how expensive AI slop and non-defensible work is (which led some employees to use the most expensive models for simple tasks).Why Did Valon Remove AI Access For New Hires?Valon had to make this change and restrict AI use within its AI-native startup because, when asked about their output, many new hires couldn’t defend their work or explain their thinking process in detail. This is concerning when you consider what could go wrong if AI produced poor-quality work and how that could affect a company's reputation and client relationships.A 2025 Microsoft Research survey examined 319 knowledge workers to understand the relationship between critical thinking and generative AI. They found that workers who reported greater confidence in using AI reported less critical-thinking effort.As AI accelerates output, critical thinking is becoming the most important AI skill, as it protects the quality and credibility of the workgettyThe interesting thing about using Generative AI is that, while it can increase productivity from the outside, it can destroy long-term productivity. Initially, you might get the job done faster. You might produce more output, release polished work, and meet deadlines more easily. However, the long-term impact on career productivity, especially for early-career hires, can be fatal.Professionals who are still in the grounding stages of their career, who do not understand the foundational structures of their work and industry, and do not know the reason why they do what they do, the whole thinking process behind it, and the problem-solving required to navigate complexity, will sooner or later find themselves stuck when approached with a situation that AI cannot resolve.For example, what could happen if:An LLM or an AI vendor tool suddenly crashed?The AI loses creativity and stops generating nuanced, original complexity?You’re confronted with a new situation within an in-person setting?The AI makes a mistake, and you don’t understand why the AI output is wrong or what led to the error?These can have costly outcomes, not only on the business, but on your job success and upward career mobility. Even experienced hires fall prey to this issue, as Valon’s policy demonstrates.Critical Thinking Is A Non-Negotiable Skill In The AI EraIt's easy to relax our cognitive skills and effort simply because we're competing with everyone else to see who can produce the fastest, most polished work.The problem is that when you outsource your thinking too often, you atrophy your critical-thinking muscle. Critical thinking has emerged as one of the most essential skills in the AI and future-of-work era. The World Economic Forum’s Future of Jobs report and Coursera's recent Job Skills report both list critical thinking, systems thinking, and problem-solving skills as non-negotiables that will see strong demand over the next few years.Here are some questions to use as a rule of thumb to ensure you continue to flex your critical thinking skills at work, and, if you’re a leader, to ensure that your employees are not multiplying work slop and injuring productivity, revenue, and reputation across the organization:If you cannot defend your output and take ownership of what was produced, you should not be using AI.If you cannot explain in detail the thinking process behind the output and even examine alternative conclusions creatively, you haven’t built your critical thinking muscle and need to dial back your AI use.Ask yourself: do I fully understand the foundational concepts and nuances of my role and industry before I outsource to AI? AI can help you learn things professionally, but it cannot be your only source.When using AI at work, get into the habit of interrogating it, cross-referencing with other sources and different models/AI tools, and applying your domain expertise to fact-check everything.Don’t just accept it at face value because the output looks convincing and AI told you it’s good.