Productivity Yes. Profitability No. Why most AI layoffs are not what they claim to be.

Last Thursday, the CEO of ClickUp, Zeb Evans, cut 22% of the company and framed the cuts as a strategy, not a cost decision. He deployed 3,000 internal AI agents to do the work the laid-off humans were doing. He promised million-dollar salary bands to the ones who stay. He named the play out loud. “100x org.” Not a cost-cutting exercise, he said. A radical bet on AI.

This is the most explicit version of the AI layoff cycle the industry has produced yet. It is also the cleanest test case of whether the framing is prophecy or wish. We will know in a year.

The Pattern Around the Edge Case

ClickUp is the loud version of a quiet pattern. Same month, three other CEOs ran the same playbook. Cloudflare’s Matthew Prince. 1,100 people, 20% of the company. Coinbase’s Brian Armstrong. Around 700 people, 14%. Upwork’s Hayden Brown. 24%. Same month. 113,000 tech jobs gone YTD across 179 companies. Layoffs.fyi counts roughly half explicitly attributed to AI. The framing converged faster than the math did. That is the tell. We have seen this movie before. Five years ago it was called COVID.

The Press Release and the Spreadsheet Do Not Match

If the cycle is familiar, the scale is new. The AI layoff narrative has become so valuable to share prices that profitable companies are cutting anyway. Cloudflare posted record quarterly revenue on the same day it announced 1,100 cuts. $639.8M, up 34% year over year, best quarter in company history. That is not a company in distress. The market right now pays more for an AI-attributed cut than for the people being cut. Prince and his board are doing what the system rewards. The cost gets paid in trust with the team that stays.

Mark Zuckerberg said the quiet part at the Meta town hall on April 30. “Getting everyone internally to use AI tools is not the thing driving layoffs.” Then he said the louder part. “We basically have two major cost centers in the company: compute infrastructure and people-oriented things. If we’re investing more in one area, then we have less capital to allocate to the other. So that means we do need to take down the size of the company somewhat.” He told 80,000 employees that they are funding the GPU bill. He absorbed the market hit instead of taking the easy reward.

Some days ago, Jensen Huang said it harder. The CEO of NVIDIA, the company benefiting more than any other from AI capex, went on Channel NewsAsia and called out the framing directly. “I think the narrative that connects AI to job loss for many of the CEOs that are doing it, it is just too lazy.” Then the timing argument that breaks the whole story. “AI has just arrived. How is it possible they’re already losing jobs? How is it possible that AI became productive and useful only six months ago, and they were somehow laying people off two years ago because of AI? It doesn’t make any sense. It was just a way for them to sound smart, and I really hate that.” When the man selling the picks and shovels says the gold rush narrative is a cover story, the cover story is over. The system is the problem. Not the people inside it.

Productivity Yes. Unit Economics No. Worldwide.

Gartner ran the audit three weeks ago. 350 global executives at companies with $1B+ in revenue, all piloting or deploying autonomous AI. 80% reported workforce reductions tied to their AI initiatives. Zero correlation between the size of the cuts and the size of the returns. Helen Poitevin, Distinguished VP Analyst at Gartner, said it cleanly. “Workforce reductions may create budget room, but they do not create return.” That is the financial verdict on the AI layoff narrative, stamped by the most cited industry analyst on the planet, in writing, four weeks before ClickUp doubled down on the bet.

Behind the headlines sits the bigger math. $725B in hyperscaler capex this year. JPMorgan: $650B in annual AI revenue needed in perpetuity to deliver a 10% return on what is already committed. PwC surveyed 4,454 CEOs across 95 countries. 56% reported getting nothing measurable back from their AI investments. The gains are real. The profitability is not yet there for most. Even the profitable companies are cutting, because the market pays them more for the haircut than the haircut costs. The market is paying for the press release. Not for the product. Yet.

Three Groups, One Cycle

The math is not uniform across the industry. Three groups of companies are operating in this market right now, and the cycle is hitting each one differently.

Group 1: Foundation Model Labs

Big tech. OpenAI, Anthropic, Google, Meta, NVIDIA on the silicon side. They train and ship the foundational models, funding a circular AI economy where the same dollar gets counted three times. They set the prices everyone else pays. NVIDIA alone is roughly 7% of the S&P 500, larger than the entire energy sector. AI policy in Washington, Brussels, and Beijing is now bent around their cycle. Their layoffs are funding a race they are already winning.

Group 2: SLM Shippers

They train and ship their own models against domain KPIs. Small language models, fine-tuned stacks, bespoke architectures for the workloads that matter to their business. They do not depend on someone else’s API for the work that defines their margin. They protect unit economics religiously because the unit economics IS the moat. Most attempts at this fail. Training infrastructure, data quality, talent, model evaluation, deployment discipline. Every step is a place where Group 2 ambitions collapse back into Group 3 economics. The ones who do it right compound. The ones who do it half-right end up with the worst of both worlds. API costs and bespoke model costs at the same time. Gartner predicts enterprise SLM usage will run 3x LLM usage by 2027. You can spot the real ones by their gross margin.

Group 3: API Wrappers

They call themselves AI-native. They built the product on top of someone else’s API. When the model owner moves a price, their gross margin moves with it. Their moat is someone else’s P&L. Until their roadmap stops depending on someone else’s pricing decisions, they will struggle to make AI profitable while competing with five other companies running the same wrapper. Cursor saw this coming. They were among the largest customers of OpenAI and Anthropic for two years, then shipped Composer 2 in March, a model trained from scratch on their own infrastructure. Call it the migration. The smart Group 3 companies are racing to become Group 2 before the math catches up. Cursor is just the first one with the capital to do it visibly. ClickUp is running a different play. Not training a model, but rebuilding the org around the assumption that the agents do the work. Same direction. Different bet. Same exposure if the productivity does not show up.

The COVID Coin Has a Flipside

Five years ago COVID hit. Zero rates followed. Every spreadsheet looked smart. Hiring went vertical. Companies hired thousands for the cycle, not for the company. Same CFOs running the model. Same compensation committees signing it off. Same incentive loop. Five years later the AI bill is exposing every API wrapper that never had a real moat. Layoffs are going vertical. AI is the biggest technology disruption in modern history. The leadership cycle around it is not. Every major technology shift in the last 150 years produced a labor restructuring followed by a labor expansion at a new layer. The companies that built the bench during the cut survived to staff the next layer. The ones that did not, did not. The disruption is new. The circle is old.

The Two Seats That Owe the Truth

Which means somebody in the room has to name what is actually happening. Two seats can. No others.

The CTO knows the group. Token spend, model dependency, gross margin compression. The CTO is the only person who knows whether the company is shipping its own models or renting them.

The CEO knows the why. The strategy bet, the capex commitment, the market signal the cuts are answering. “We bet on shipping our own models and we ended up renting margin” is harder to say than “AI made you redundant.” It is also the truth the team is owed.

The ROI Returns When You Return to the Problem

And once the truth is named, the question shifts from who to cut to what to actually build. The 95% with no measurable return is not because AI does not work. It is because most companies are using AI to satisfy the market’s hunger. Not to solve a problem the user actually has. The ROI shows up when the technology gets pointed at a real workflow, a real friction, a real cost that real customers feel every day.

Stop doing GenAI for the sake of GenAI. Take the step back. Ask what the end user is actually paying for. Build the AI around the answer. The companies running this playbook quietly are the ones whose unit economics work. The ones doing AI to be seen doing AI are the ones cutting people to make the math look better. The technology was never the problem. The use case was.

Three Questions for the CEO

If you want to know which group your company is actually in, three questions answer it.

One. Does your core revenue depend on a workflow that runs through someone else’s foundational model API? If yes, you are Group 3 today, regardless of what the deck says.

Two. Has your team trained, fine-tuned, or shipped a model against a specific domain KPI in the last twelve months? If no, the Group 2 transition has not started. If yes, the next question matters more than this one.

Three. Does your gross margin move when OpenAI or Anthropic adjust their pricing tiers? If yes, the SLM work is real but not yet load-bearing. If no, you are Group 2.

Three questions. Five minutes. The honest answer tells you what the next quarter has to be about.

Build for the Jobs That Are Arriving

Same logic applies to the people. Stop mourning the jobs that are leaving. Start building the people for the jobs that are arriving. The same disruption removing roles is creating new ones nobody is qualified for yet. The most valuable hire in 2026 is the token economics analyst who can cut a company’s inference bill by 40% without touching the product. That role did not exist 18 months ago. There are maybe a few hundred people in the world who can do it well. The ClickUp version of this hire gets a million-dollar salary band. The non-ClickUp version gets stock options and a story. Either way, the role is the leverage point. The people getting cut this quarter are not failing. They are funding a transition. The leaders who survive are building the bench. The rest are doing layoff math.

If You Ask Me

AI exposes our deepest weaknesses as builders and as leaders. The companies pointing it at imaginary problems get exposed for not knowing what their users actually need. The companies pointing it at real-life problems, with the guts to take them to the next level, get the unit economics, the customer retention, the moat that holds. The API combiners are racing against the clock. Either they migrate or the model owner absorbs their use case. There is no third path. The big ones are shaping the economy of the world. The middle path is the one that wins, as long as they do it right. The competition is fierce.

Build something that solves a real problem. Ship the model that runs it. Defend the unit economics like your company depends on it. Because it does. ClickUp just bet the company on the version of this that fires humans and trusts the agents. We will know in a year whether that bet was prophecy or wish.

And do not be surprised when a CEO announces they are firing AI agents next, to feed the market’s hunger once again. History and markets run in circles. Nobody escapes. The market will reward that too, for a quarter.