AI layoffs in 2026 are stacking up fast, and the loudest justification for them comes from people who have never done the jobs they're cutting. That's the uncomfortable point Box founder Aaron Levie raised in a recent TechCrunch video, What happens when companies become too AI-pilled?. He has a name for the pattern: "AI psychosis."
The framing matters. When a CEO decides AI can replace your job, Levie's argument goes, that CEO is usually the person furthest from the actual work and least equipped to judge what the work involves. I want to unpack what that means for an engineer, student, or two-person startup here in Sri Lanka, because the takeaway is not "AI is fake." It's almost the opposite.
🧠 What "AI psychosis" actually describes
The phrase sounds like a jab, but it's pointing at a real decision-making failure. The person signing off on the layoff and the person who understands the day-to-day work are rarely the same person.
The people deciding AI can replace your job are often the ones least likely to understand what your job truly involves.
A job title is a thin description of a real role. "Customer support" hides the part where someone catches a billing bug before it becomes a refund storm. "QA" hides the engineer who knows which flaky test actually matters. AI agents are genuinely good at the visible 70% of a role and quietly bad at the invisible 30% that keeps things from breaking. Someone close to the work sees both halves. Someone reading an org chart sees only the first.
That gap is where over-eager cuts come from.
📊 The 2026 layoff numbers are not a rounding error
Two data points from the source put scale behind the rhetoric:
| Signal | What the source reports |
|---|---|
| ClickUp workforce cut | 22% of staff, citing AI agents |
| 2026 tech layoffs so far | Already nearly matching all of 2025 |
I'm not going to pretend those numbers are harmless. A 22% cut is real people losing real income. And matching a full prior year before this one is even over means the pace is accelerating, not cooling off.
But notice what the numbers don't prove: they don't prove the AI replacing those roles actually works yet. A layoff is a bet placed today on a capability that's promised for tomorrow. Some of those bets will pay off. A chunk of them are the "AI psychosis" Levie is describing, and those companies will quietly rehire in twelve months.
🌐 Why this lands differently in Sri Lanka
Most of these cuts are happening at funded US tech firms. If you're building or job-hunting from Sri Lanka, the second-order effects matter more than the headlines.
- Remote roles get more competitive. Laid-off engineers in expensive markets start applying for the remote and contract work that SL freelancers also chase. Your competition just got deeper.
- The "cheap labour" pitch is dead. "Hire me, I'm affordable" now competes directly with "hire an AI agent, it's cheaper still." Price was never a moat. Judgment is.
- Small teams get a real edge. A two-person studio in Colombo that uses AI to do the work of six is the upside of this story. The danger isn't AI doing the work. It's pretending it does the whole job.
If you earn in dollars and spend in rupees, the currency math still works in your favour. If you want to see exactly how a remote contract converts after transfer fees, the free tools at induwara.lk/tools include calculators built for that.
🛠️ The skill that survives the hype cycle
Here's my actual argument. The roles most exposed to bad AI bets are the ones that can be described in a sentence. The roles that survive are the ones where someone owns the messy parts a model can't see.
Key takeaway: AI doesn't replace people who understand the whole system. It replaces people whose work was only ever the easy, visible slice of it. Be the person who owns the hard slice.
Concretely, for a student or early-career engineer:
- Learn to debug, not just to generate. Anyone can prompt an AI to write a function. Far fewer can tell when that function is subtly wrong. That judgment is the moat.
- Build something end to end. Ship a small project that talks to a database, handles a payment, survives bad input. The "invisible 30%" is exactly this glue work.
- Stay close to the real problem. The engineers being cut by mistake are usually the ones who let themselves drift far from what the product actually does for a user.
You don't need a paid course or expensive hardware to start. A browser and a willingness to break things is enough. If you're learning, our free in-browser code compilers let you run Python, Java, C++, and more without installing a toolchain, which is plenty to build real fundamentals.
💡 What this means for you
The 2026 AI layoff wave is real, and some of it is genuine restructuring. But a meaningful slice is "AI psychosis" — confident decisions made by people too far from the work to judge it. That's bad news if your role was only ever the visible, describable part of a job. It's good news if you're willing to own the hard part.
My advice, plainly:
- Don't panic-pivot away from engineering. The demand for people who understand systems is going up, not down.
- Do get uncomfortably good at the messy 30% — debugging, edge cases, knowing why something broke.
- Use AI as leverage, not as a replacement for understanding. A small SL team that wields it well is exactly who wins this cycle.
The companies that became "too AI-pilled" are making a bet. Make sure you're on the side of the bet that understands the work.
