June 2026 · Anthony Sogg

The Org Chart Is the Demo

Two weeks after a record quarter, Salesforce cut jobs around its fastest-growing product. The most honest demo a vendor ever gives is what it does to its own org chart.

Two documents came out of Salesforce in the span of two weeks, and they are worth reading as a pair.

The first, on May 27, was a victory lap: record revenue, a 34.8% non-GAAP operating margin, and $1.2 billion in Agentforce annual recurring revenue, up 205% year over year. Marc Benioff's quote had tens of thousands of businesses transforming into "Agentic Enterprises."

The second, on June 10, was a California WARN notice. Eighty-six jobs across the orbit of Agentforce, MuleSoft, and Marketing Cloud: sales, general administration, technology and product, with more roles affected in Washington state and overseas.

Most of the coverage filed this under irony. The company selling you agentic labor is trimming the humans around its agents, cue the eye-roll. I want to put it in a different drawer.

This wasn't a contradiction of the keynote. It was the keynote, performed.

Columbo never solved anything in the interrogation room

I've written before that every vendor demo is a magic trick, staged and rehearsed, with the interesting part happening off camera. That hasn't changed. What I should have added is that there is exactly one demo a vendor cannot stage: the one it runs on its own payroll.

Columbo understood this. The confession never came from the suspect's prepared story. It came from the shoe, the car, the small domestic detail that contradicted the script, the thing they did when they assumed nobody important was watching. A WARN notice is that detail. Nobody writes a regulatory filing for applause.

So when a company tells Wall Street its agents delivered 3.8 billion "Agentic Work Units" this quarter, and then quietly files paperwork on the people who sold and supported those same products, you are watching the act and the tell in the same scene. The raincoat is rumpled, but the man is paying attention.

What the filing actually says, fairly

Let me be fair to Salesforce here, because the strongest version of this argument requires it.

Eighty-six people out of a company of roughly 80,000 is a rounding error. The severance is genuinely decent: up to 30 weeks of pay, payroll through August 7, healthcare continuation that beats what Oracle offered its own laid-off staff this cycle. Reports say the core Agentforce team was not directly affected. If you stop reading there, there's no story, and plenty of people would tell you to stop reading there.

Size is the wrong lens, though. Watch the direction of travel.

These cuts didn't land in a dying division. They landed around the growth engine, the product growing 205% a year, in the same quarter the company returned $27.1 billion to shareholders through buybacks. This follows a January round of just under 1,000 roles. For most of the software industry's history, growth like that bought headcount. That was the deal: the rocket ship hires. Now the rocket ship files WARN notices mid-burn.

And there's a quieter signal sitting in the earnings release that I haven't seen anyone connect. Salesforce now reports "Agentic Work Units" as a headline metric. Work, counted in units, by the billions. Here's the thing about counting work in units: sooner or later, someone in finance divides those units by the humans standing next to them. The denominator always comes for somebody.

Place Your Bets: four ways to read one WARN notice

Now for the speculative part of the program, because a) honesty requires me to admit the Columbo frame has a limit and, b) we can wager on just about anything and speculation seems to be closer to fact than fiction these days.

A small filing is a clue, and clues support more than one suspect. Here are the four explanations I think are live, with the evidence for and against each. Bring your own probabilities.

1. Boring acquisition math. Salesforce closed its Informatica deal, which is already contributing $1.1 billion of cloud ARR per the Q1 release, and Informatica's integration business overlaps MuleSoft's almost exactly.

The Contentful agreement sits on top of Marketing Cloud's content tooling the same way. Post-acquisition rationalization is the most ordinary explanation for cuts in exactly those two orgs, and ordinary explanations win most coin flips.

Evidence for: the specific units hit.

Evidence against: it says nothing about the Agentforce-adjacent roles, and the Contentful deal hasn't even closed yet.

Could also be a result of Marketing Cloud Next, the agentic rebuild of the legacy platform announced last June, and redundancies there.

2. The bet is being repriced, not abandoned. Business Insider's November reporting found fewer than half of Agentforce's 12,500 customers were paying, and fewer than 2% of total customers were running more than 50 agent conversations a week. (Salesforce pushed back, calling those numbers "one cut of data.")

The flagship reference from the 2024 launch demo, the Saks voice agent, was rebuilt with Amazon and NLX. The broader market rhymes: an MIT report last July found 95% of organizations seeing no return on $30 to $40 billion of enterprise GenAI spend. And yet the ARR grew 205%, driven by premium SKUs anchored in Sales and Service, with more than half of bookings from existing customers.

Put all of that together and a quieter story forms: the revolution did not sell as a standalone product, so it is being folded into the core SKUs and metered in work units. If the product is no longer sold separately, you need fewer humans dedicated to selling it.

Evidence for: it explains who was cut (the sell side), who was protected (the build side), and the pricing migration.

Evidence against: bookings mix is circumstantial, and none of us can see inside the contracts.

3. Margin math in an AI costume. An activist investor raised its stake by nearly 50% last summer, revenue growth slipped to single digits in 2025 for the first time in the company's public history (it has since recovered to 13%), and the same quarter as the record results saw $27.1 billion in buybacks.

On this reading the cuts are ordinary opex discipline wearing the AI narrative as a costume, a reflex I've written about before.

Evidence for: the buyback, the margin pressure, and a January round of cuts that predates any plausible internal agent payoff.

Evidence against: 86 people does not move a margin needle at this scale. As cost theater goes, it's a strangely small performance.

4. Drinking their own champagne. The agents really are absorbing work, and Salesforce is starting with its own org, exactly as an integrity-first version of this industry would.

Evidence for: the company does run its own products internally, and severance this generous reads more like a managed transition than a panic cut.

Evidence against: agents that fewer than 2% of customers use heavily are not yet absorbing anyone's job, including Salesforce's. The usage data has to inflect before this story can be true, and nobody has published that data.

Forced to bet, I'd put about a third on the acquisition math, a third on the repricing, a quarter on the costume, and the small remainder on the bubbly, but then again, I get no kick from...never mind, I digress for fear of Cole Porter's copyright lawyers, though I bet I'd get a kick out of them.

But that's one operator reading public documents.

Some of you are closer to this than I am. Tell me which scenario you'd bet on, or name a fifth: hireme@crmjedi.com.

You're not buying software. You're subscribing to a staffing philosophy.

Whichever scenario wins, one thing holds if you run a CRM program. I've run enterprise lifecycle programs on Marketing Cloud, and I can tell you what never shows up on a pricing sheet: the humans.

When a triggered send hangs at two in the morning the night before a launch, the thing that saves you is not a dashboard. It's a person, usually one specific person, who knows your org, your quirks, and which escalation path actually moves.

The roles in this filing, sales and support and product adjacents around Marketing Cloud and MuleSoft, are exactly that layer. The connective tissue between you and the product.

So, treat the vendor's org chart as part of the product spec, because it is. Three questions worth asking at your next renewal, in writing:

Who, specifically, owns my escalation path, and what happened to that team's headcount this year? If agents are now handling part of my support load, what is the human service-level commitment behind them when the agent taps out? And if your "work units" keep growing while the team around my products keeps shrinking, where exactly does my support quality land in two years?

There's a contract angle hiding here too.

Investors have spent the year worrying that agents will shrink seat counts, which is the polite explanation for the stock being down more than 30% this year. A vendor whose seat model is under pressure will protect revenue per account some other way, and a metric like Agentic Work Units is consumption pricing with its coat still on.

If your renewal quietly shifts from seats to units, you'll want to have seen that coming. You have now seen it coming.

Why none of the scenarios saves the keynote

Step back from the scenario board and notice the one thing all four readings share: not one of them is the story being told on stage. Not even the friendliest one.

If the layoffs were the honest demo, the integrity move every AI vendor claims to be making, then the demo belongs in the keynote.

Stand up in Denver and say: our product reduces the need for these specific roles, and we know because we started with ours. That sentence would be the most persuasive pitch in the history of enterprise software, and nobody will ever say it into a microphone. Likely because no one wants to bring back vaudeville tomato throwing but hey, anything goes. (If you get that, please drop me a line.)

Instead, the ARR slide gets the spotlight and the WARN notice gets a quiet regulatory filing. The confidence lives in one document, the consequences live in the other, and the gap between them is the tell.

Watch the payroll, not the keynote. The keynote is written for you. The payroll is written for the auditors, and the auditors don't clap.

Just one more thing, though: when your account executive calls to talk about the agentic future, ask them how many people are left on the team behind it. The answer is the product roadmap.

Anthony

← More writing · If this was useful, or wrong in an interesting way, tell me: hireme@crmjedi.com.