Salesforce cut staff weeks after a record AI quarter. Should the board cut engineers next?

I had the same Salesforce headline forwarded to me three times this week, twice by founders and once by a board member.

The facts are clean. On June 9 and 10, Salesforce filed for another round of cuts: 86 roles in a California WARN notice, spread across the Agentforce, MuleSoft, and Marketing Cloud teams. Dataquest and American Bazaar both put it less than two weeks after Salesforce told the market that Agentforce had crossed $1.2 billion in annual recurring revenue, up 205% year on year. Third layoff round in nine months. Record AI quarter and fresh job cuts, same company, same month.

The headline a board takes from that is brutal in its simplicity: AI is so good now that the company selling it is firing the people who build it. And the unspoken follow-up, the one that lands on a CEO’s desk the next morning, is “so when do we do the same to engineering?”

The number that made the headline is not the number that matters

Here is the part that gets lost between the headline and the board meeting: 86.

That is how many roles Salesforce cut. Against an engineering organization that the company itself describes as roughly 15,000 people, 86 is a rounding error. And the most important thing Salesforce has said about that 15,000 is not in the layoff filing at all. It is that the number has stayed roughly flat for about two years, by choice. Fortune reported at the end of May that Benioff has held engineering headcount steady rather than cutting it, even while saying AI has changed how much each engineer ships.

So the move is not “agents replaced our engineers.” The move is “we froze engineering, redirected it, and trimmed 86 roles at the edges.” Dataquest was careful to say so directly: the developments “do not support a simple conclusion that Agentforce’s success is eliminating the jobs required to build and support it.” The core Agentforce teams were not touched.

And the company is spending into engineering, not away from it. American Bazaar noted Salesforce expects to spend close to $300 million on Anthropic tokens in 2026, most of it on coding work. That is a company buying every engineer a very expensive power tool, not one that thinks engineers are obsolete.

The board saw the 86 and the $1.2 billion next to each other and drew a line between them. The number that should drive a headcount decision is the 15,000 that did not move.

Double the output from the same engineers. When output per person doubles, a company has two honest options: ship twice as much, or keep the same engineers and cut. Salesforce, so far, chose the first one and held the line on people. That is the detail the headline strips out.

Whether cutting engineers pays, and who it hurts

When this comes up at the board meeting, it arrives as three questions in a trench coat. Answer them in order.

“Should we cut engineers like Salesforce did?” The data says be very careful here. A Gartner study from May surveyed 350 executives at companies above a billion dollars in revenue. Roughly 80% of the ones deploying autonomous AI had already cut headcount. And the cuts showed no correlation with better returns. The firms reporting strong ROI had reduced their workforce at almost exactly the same rate as the firms reporting weak or negative ROI. Cutting freed up budget. It did not produce results.

“Then what are the engineers we keep actually doing?” They are doing the one job agents cannot do for themselves: deciding whether the agent was right. The trust gap is well documented. Sonar’s developer survey found 96% of engineers do not fully trust AI-generated output, and only 48% always verify it before it ships. Double the code, the same skepticism, and a thinner bench of people qualified to apply it. That is the actual bottleneck, and it is staffed by exactly the senior engineers a panicked cut would target first.

When an agent writes twice the code, the scarce resource is no longer someone to write it. It is someone trusted to say the agent got it wrong.

“So where do the savings come from, if not engineers?” This is the uncomfortable one. The real erosion is already happening, and it is not at the senior level. Stanford’s AI Index found employment for developers aged 22 to 25 has fallen about 20% since 2024 while their older colleagues’ headcount grew, and a Harvard study of 62 million workers saw junior developer employment drop 9 to 10% within six quarters of generative-AI adoption, with senior employment barely moving. The cut is real. It is just landing on the pipeline that produces future seniors, which is borrowing from a future board meeting.

Salesforce held its engineers flat, cut 86 roles

If a CEO has one minute before the board picks this up, here is the whole thing.

Salesforce did not fire its engineers. It cut 86 roles, mostly outside the core teams, while holding roughly 15,000 engineers flat and spending hundreds of millions on coding tools for them. The headline is real; the layoff is small and selective, and the engineering story underneath it is a freeze, not a purge.

The broad evidence agrees. Cutting engineers to chase AI returns is the one move a large study found has no measured payoff. A late-May CEO survey caught the mood: only 27% of chief executives said their AI returns had met expectations, down from 38% a year earlier. Enthusiasm for cutting is running well ahead of the evidence that cutting works.

So the defensible position is not “match Salesforce.” It is to hold the line on engineering, protect the verification layer, and treat the junior pipeline like the strategic asset it is.

Watch whether layoffs name engineering, or sales

Watch whether the next “AI revenue plus layoffs” headline names engineering specifically or, like this one, names sales, support, and admin while engineering quietly stays flat. The distinction is the whole story, and most coverage will blur it.

Watch your own junior-to-senior ratio over the next two quarters. That is where the real decision is being made, usually without anyone deciding it.

And take a breath before the meeting. The data actually hands you a calm, board-ready answer here, and it is not the reflexive one. The companies getting AI right are not the ones cutting hardest. They are the ones who kept the people who know when the machine is wrong.