AI Layoffs Are Starting to Feel Different

This week felt different.

Not because tech layoffs are new.

They are not.

What feels different is who is saying it, how openly they are saying it, and what they are pointing to.

Coinbase disclosed a plan to cut about 700 employees, roughly 14% of its workforce, while saying the company was adjusting for current market conditions and optimizing its operations for the AI era. Brian Armstrong’s note to employees went further. He wrote that engineers were using AI to ship in days what used to take weeks, that non-technical teams were shipping production code, and that Coinbase would rebuild around AI-native pods and people managing fleets of agents.

Two days later, Cloudflare said it would reduce its workforce by more than 1,100 employees globally. Matthew Prince and Michelle Zatlyn wrote that Cloudflare’s internal AI usage had increased by more than 600% in three months, that employees across engineering, HR, finance, and marketing were running thousands of AI agent sessions each day, and that the company was reimagining internal processes, teams, and roles for the agentic AI era.

Those are not random companies.

They are not legacy businesses trying to look modern.

They are technical, infrastructure-heavy, product-led companies that were already close to the AI shift.

That is why this feels like a signal.

My prediction is that we will keep seeing this. I would not be surprised if there is at least one major AI-linked layoff or restructuring story per week over the next several weeks.

Some of it will be real automation.

Some of it will be market pressure.

Some of it will be companies using AI as cleaner language for cost cutting.

But even if the story is messy, the direction is pretty clear. Companies are now asking a brutal question:

How much should we invest in compute, AI tools, and agents instead of people?

That is a hard question.

It is also a very human one, because the answer lands on people who now have to look for work in an already crowded market.

The numbers are already uncomfortable

I do not want to overstate this as a clean “AI caused every layoff” story.

That would be too simple.

But the numbers are no longer theoretical.

Challenger, Gray & Christmas reported that technology companies announced 52,050 job cuts in the first quarter of 2026, up 40% from the same period in 2025. In March, AI led all stated reasons for job cuts, accounting for 15,341 cuts, or about 25% of the month’s total. CBS reported that AI was the leading cited reason again in April, with 21,490 AI-related cuts, or 26% of all April cuts.

That is already tens of thousands of people.

And that is before this wave really settles into company planning cycles.

I think that is the part people underestimate. A company does not need to believe AI can replace everyone to change hiring behavior. It only needs to believe that AI lets smaller teams do more work, or that future AI will let smaller teams do more work soon.

That belief alone changes budgets.

It changes backfills.

It changes whether a team gets three junior roles or one senior engineer with agent tooling.

It changes whether a manager is allowed to hire now or told to wait another quarter.

The labor market feels the delay before the productivity gains are proven.

I keep thinking about new grads

This hits experienced engineers.

It also hits people trying to get their first real shot.

I graduated from SUNY Potsdam in 2017 with the basic promise that if you studied computer science, worked hard, and could write code, there would be a job path waiting for you.

That was not guaranteed, obviously.

But it felt structurally true.

I do not think it feels that way anymore.

The New York Fed’s recent college graduate data shows how much harder the entry point has become. In early 2017, the unemployment rate for recent college graduates was around 3.8%, and by December 2017 it was about 3.5%. In the first quarter of 2026, the New York Fed put recent-grad unemployment at about 5.7%, with underemployment at 41.5%.

For the latest outcomes by major, the New York Fed reports a 7.0% unemployment rate for recent computer science graduates and 7.8% for computer engineering graduates. Those majors still show strong wages for people who land, but the first job is not the same promise it used to be.

NACE’s Class of 2024 outcomes tell the same story from another angle. The overall career outcomes rate for bachelor’s graduates rose to 85.7%, but slightly less than 55% reported full-time employment within six months of graduation, down from more than 57% for the Class of 2023. NACE’s Job Outlook 2026 called the market cautious and initially projected only 1.6% growth in new college hiring compared with the Class of 2025.

Then there is the AI-specific entry-level data. Stanford researchers using ADP payroll data found that early-career workers ages 22 to 25 in the occupations most exposed to generative AI saw a 16% relative decline in employment, while more experienced workers in the same occupations were stable or still growing.

That is the apprenticeship problem.

Companies still want senior judgment.

They still want people who can own systems.

They still want people who can review AI output, connect it to business context, understand risk, and make production decisions.

But those people do not appear out of nowhere.

They usually start as juniors.

They learn by doing the “easy” work. They fix bugs. They write tests. They ship small features. They break things in low-risk ways. They sit near better engineers and slowly build judgment.

If AI eats too much of that first layer, the industry does not just lose entry-level jobs. It weakens the path that creates experienced engineers later.

That is bad for graduates.

It is also bad for companies.

The production-risk part matters

I use AI to write code every day.

I am not anti-AI.

I am building with these tools because they are genuinely useful. They make me faster. They let small teams do more. They lower the cost of trying ideas. They are changing software engineering in a real way.

But there is a difference between AI helping engineers move faster and companies treating production judgment as something that can be casually spread across the organization because the tool can generate code.

Coinbase’s note said non-technical teams are now shipping production code. I do not know what that means internally. It might mean small, reviewed, well-guarded changes. It might mean internal tooling. It might mean something riskier.

I want to be careful here. I have seen people try to connect specific Coinbase service incidents to that new operating model. The May 7 disruption might be a coincidence. It might be related in some broader operational way that is not public. I do not think anyone outside the company really knows. What we can say from the public record is narrower: Coinbase’s status page attributed the degraded performance to an AWS outage and then to increased temperatures in the affected AWS service. All markets were re-enabled early on May 8. That makes it a cloud dependency and resilience story first, not public evidence that AI-generated or non-engineer code caused the issue.

The broader point is still important.

If companies want more people and agents producing changes faster, the review and reliability system has to get stronger too.

The last year has given us enough reminders of that. OpenAI launched Codex as a cloud software engineering agent in May 2025. Claude Code became generally available later that same month. By late 2025, coding agents were normal enough that a lot of engineering teams were changing their workflows around them.

At the same time, major infrastructure outages kept reminding us how much damage a single change can do. Cloudflare’s November 2025 outage was triggered by a database permissions change that doubled the size of a Bot Management feature file, propagated it across the network, and caused core traffic delivery failures. Microsoft’s October 2025 Azure Front Door outage came from a configuration change that affected nodes across the global fleet.

I am not saying Codex or Claude Code caused those outages.

There is not evidence for that.

What I am saying is that modern infrastructure is fragile enough when expert teams move carefully. If we accelerate code generation, broaden who can make production changes, and reduce the number of experienced humans in the loop, then guardrails matter more, not less.

The future cannot just be:

Ship faster because AI can write it.

It has to be:

Ship faster because AI can help, and because humans built stronger review, testing, observability, rollback, permissions, and ownership around it.

Without that, companies are not becoming AI-native.

They are just moving risk around.

The human cost is easy to abstract away

Markets like efficiency.

Executives like cleaner org charts.

Investors like stories about leverage.

AI makes all of those stories easier to tell.

But behind every “AI-native restructuring” there are people who now have to re-enter a market where job postings are crowded, applicant pools are huge, resume screens are automated, and entry-level workers are already under pressure.

That part bothers me.

I feel for the Coinbase employees.

I feel for the Cloudflare employees.

I feel for the new CS grad who did what they were told to do, picked a hard major, built projects, sent out hundreds of applications, and now has to compete with laid-off engineers who have years of production experience.

I also feel for the engineer who is still employed but now has fewer teammates, more agent output to review, more systems to own, and maybe fifteen direct reports because the org got flatter overnight.

None of this is as simple as “AI bad.”

AI is going to create new work.

It is going to make some people much more capable.

It is going to help small teams build things that used to require large teams.

But the transition is going to be uneven. Some people will land in the new work quickly. Some will get squeezed by timing. Some will be told to reskill while the ladder they need to climb is being pulled up.

That is the part we should not casually wave away.

What I would want companies to do

If companies are going to make this shift, I think they owe employees more than vague AI language.

Be specific about what changed.

Say which work is automated.

Say which roles are being rebundled.

Say where human review still matters.

Say how production risk is controlled.

Say how junior people will learn if the junior work is being automated.

Say whether the company is cutting because AI is actually replacing work, because the market is weak, because margins are under pressure, or because leadership is making a bet on future productivity.

People can handle hard truths better than they can handle polished ambiguity.

The best version of this transition is not pretending that nothing changes.

It is building AI into work while preserving the judgment, apprenticeship, accountability, and trust that make technical systems reliable.

That means AI agents should be treated like powerful teammates that need permissions, evals, logs, review, and blast-radius limits.

It also means humans still matter.

Especially the ones who know the system.

Especially the ones who can say, “This generated diff looks right, but this rollout path is dangerous.”

Especially the ones who understand the difference between a working demo and a production-safe change.

A small note on what I am building

I have never really used a blog post to promote a product.

I am cautious about doing it here because the last thing I want is to turn someone else’s layoff into a marketing hook.

But I am building TrySignalHire because this market feels broken from both sides.

Candidates are using AI to write and optimize applications. Employers are using AI to screen and filter them. Application volume keeps going up. Trust keeps going down. People with real work can still disappear in the noise before a human ever sees them.

TrySignalHire is my attempt to build a better signal layer for that market. The goal is not to replace human hiring judgment. It is to make real candidate evidence easier to collect, review, and trust.

If you were impacted by layoffs, or if you are a new grad trying to make your work legible in an AI-heavy hiring market, you may find it useful once it launches.

You can join early access at trysignalhire.com.

I would also just like to hear what people are experiencing right now. If this post hits close to home, you can email me at jesse@trysignalhire.com.

I do not think one tool fixes this labor market.

But I do think the old resume funnel is getting worse at exactly the moment workers need better ways to show what they can actually do.

That is the problem I am trying to work on.

And after this week, it feels more urgent.