Matt Shumer sparked panic around AI and the labor market, but the data do not confirm a wave of layoffs

Matt Schumer's Viral Post Triggered AI Panic, But the Data Doesn't Show a Wave of Layoffs

Matt Schumer's viral post claiming AI's impact on the job market would be "bigger than Covid" garnered tens of millions of views and sparked a wave of anxiety. But if you set aside the dramatic tone and look at the data, the picture is far more complex: there's currently a huge gap between what AI theoretically can do and what companies have actually implemented.

Why the Post Worked

Schumer hit the market's perfect pain point: the fear that models are already capable of replacing most office workers today. His framing was simple and alarming—"something massive is happening," "most people won't know about it until it's too late." Over 85 million people saw the post, and for many, that was enough to accept the emotional thesis as an analytical conclusion. The problem is that the post contained almost no data on employment, implementation rates, or actual layoffs.

Later, once the wave had already spread, the author's tone became noticeably softer. In a CNBC interview, he admitted he didn't want to scare anyone and suggested that some of his wording should have been rewritten. This is crucial for the market: panic spreads quickly, while corrections are read by few. This is precisely why viral posts about AI are dangerous for leaders. They easily prompt layoffs, hiring freezes, or rushed "transformations" without understanding where models actually deliver results and where they merely look good in demos.

"If I had known how viral this would become, I would have rewritten some parts."

Where Automation Breaks Down

The main counterargument to "AI already replaces everyone" is visible in Anthropic's data. Theoretical task coverage by models is indeed very high: for some professions it approaches 90% or higher. But actual usage is notably lower. In IT, for example, Claude covers only about a third of tasks in practice, despite much higher potential on paper. This is the key gap: capability is not equal to implementation, and implementation is not equal to full human replacement.

The reason is simple: work in companies consists not just of the task itself, but everything surrounding it. Models hit a wall not in text or code generation, but in context, dependencies, and organizational constraints. Even if a system can write an email, analyze a document, or suggest code, that doesn't mean it understands when to launch the process, who to ask for approval, and what to do when exceptions arise. This is precisely why automation stalls where everything looks straightforward from outside.

• internal sign-offs that are nowhere formally documented
• legacy systems and non-standard workarounds
• regulatory constraints and compliance checks
• team's implicit knowledge about clients, risks, and priorities
• constantly changing rules within the organization itself

By this logic, AI often turns out to be even weaker than a new employee. A human at least gradually integrates into the environment, learns from conversations, notices exceptions, and picks up unwritten rules. A model starts almost from scratch each time and depends on how carefully context was handed to it. And context in a living company changes faster than it gets formalized. Therefore, mass personnel replacement based on AI's theoretical capabilities today is not a strategy but an expensive experiment with a high risk of rollback.

Where Demand Is Shifting

The most useful part of this discussion is not the debate about how many professions disappear, but the question of which tasks finally become economically solvable. The author suggests looking at a simple demand curve. At its "head" live mass products like CRM, office software, and design services. They've long covered standard processes. But in the "tail"—thousands of small, painful, and very specific tasks that went unautomated for years simply because development cost more than the outcome. This could be a report in a unique format for one client, a non-standard compliance process, a private dashboard for a CFO, or a local bureaucratic procedure that derails the entire business timeline.

AI makes such tasks addressable: they can now be solved by a small team or even a single person with deep domain expertise. Hence the paradox of 2026: AI doesn't necessarily shrink a company, but it almost always speeds it up. In January 2026, Citadel Securities reported developer vacancy growth of 11% year-over-year, and forecasts for many professions affected by AI still point to growth, not collapse.

What This Means

Right now the main competitive advantage is not to panic over viral theses, but to understand where in your business the gap lies between "AI could do this" and "AI actually does this." Those who cut people too early risk having to hastily rebuild expertise later. Those who first choose the problem and then the tool will likely win from the current automation wave.