Anthropic Runs 17 Engineers per PM. The Ratio Is the Wrong Question.

Anthropic published data on its own workforce on October 16. Nathan Wilmers, Regina Catipon, Maxim Massenkoff, and Peter McCrory on its economics team wrote it up, and one sentence in it is going to end up in a lot of board decks: "the ratio of engineers to product managers, for example, sits at around 17 to 1 and has increased since 2025."

Please don't copy the number. Read the two next to it.

"The number of commits from managers has also doubled." And the junior share of new technical hires fell from 27 percent to 9 percent.

That's the org-shape story in three lines. The ratio went up because the people who coordinate now also build. And the bottom of the ladder got thinner while it happened.

This is part 4 of The CPO Transition. Part 1 argued the CPO stops owning the roadmap and starts owning the operating model, with the Product Builder as the unit of the org. Part 3 split the substrate with the CTO. This one is the org chart that sits on top.

The short version

The PM-to-engineer ratio is a reading, not a target. Anthropic runs about 17 engineers per PM, but that number is the output of an operating model where managers ship code, agents absorb coordination, and decisions get written in the open. A CPO who copies the ratio without the model just has fewer people relaying the same work. The design decisions underneath are spans, the new work, and the bottom rung. Spans are widening, Amplitude went from 3.6 to 5 reports per manager, so product managers who manage have to ship or decide. The new work is evals and agent management, which is where Anthropic moved its product writing team, so redeploy before you cut. And the bottom rung is thinning everywhere: Anthropic's junior technical hires fell from 27 to 9 percent and Aakash Gupta's data shows fewer people entering PM at L1. Whoever stops hiring at the bottom has no senior bench in six years. That one is the CPO's call, and it should be made on purpose.

The ratio is a reading

I've written about why the PM-to-engineer ratio is the first thing AI breaks, and part 1 covered why there's no new magic ratio to chase. Anthropic's report is the best public evidence yet for both, because it shows what produced the number.

Meta-work went from 10 percent of the average employee's interactive Claude sessions in fall 2025 to 22 percent in spring 2026, and the authors say much of that rise was agent management. Standing instructions for agents went from a few hundred to 8,500 in a year, per Jason Lemkin's summary of the report. And the report says something I'd put on the wall of every product org: "Private work, in unlogged meetings and direct messages, is invisible, confusing and discouraged in an AI-heavy workplace."

That's an operating model. Agents do the relay. Decisions live where agents and people can both read them. Managers build. Run that, and one PM can cover a lot more engineering. Skip it, cut to 17 to 1 anyway, and you get a PM drowning in status meetings that used to be split three ways.

The authors are careful that causality is unclear, and Anthropic is not a typical company. Its headcount tripled in a year. So use it as a picture of the mechanism, not a benchmark.

Spans widen, so managers have to ship

Rich Holmes passed along Amplitude's numbers from Spenser Skates on October 11. Over 18 months, managers went from 158 to 135 while individual contributors went from 567 to 656. Average span went from 3.6 to 5.

Put that next to Anthropic's managers doubling their commits and the pattern is plain. The manager who mostly moves information between people is the role that compresses. I made that argument in AI Deletes the Director of Product, Not the PM, and now there's data from two companies pointing the same way.

For a CPO the design question is what a product manager who manages is now for. My answer: they ship or they decide, and ideally both. They own the bar for their area, the eval set that says what good looks like, and they can still build a prototype that settles an argument. A manager of five Product Builders has no room for a status layer, and shouldn't want one. If your directors spend most of the week in syncs, that isn't a people problem. It's the operating model telling you which layer it no longer needs.

Redeploy into the new work before you cut

The most useful line in the Anthropic report for a CPO isn't a number. It's what happened to a product writing team: "The team then rotated into agent prompting and began writing model evals and prompts for Claude." The authors' summary: "Writers who used language to drive human behavior are now using it to drive Claude behavior."

That's the redeployment map. AI created real work in a product org. Writing evals. Writing and maintaining standing instructions. Running agent pipelines and reviewing what they produce. Somebody has to own each, and The Eval Is The Spec argues the first one is product's job outright. The people who already know your customers and your voice are usually the best fit, and Anthropic found employees with a year of tenure were 50 percent more likely to switch departments each month in 2026 than in 2025. Mobility is part of the design.

So before a reduction, list the new work and ask who in the current org could own it in a quarter. Product Leadership Through Layoffs covers the cut itself. This is the step before.

The bottom rung is the CPO's problem

Aakash Gupta's PM compensation and hiring report came out on October 22, built on Levels.fyi, Live Data Technologies, and TrueUp data. Outside PM hires are down about 45 percent since 2022. L1 fell from 17 percent of Levels.fyi PM reports in 2022 to 11 percent in 2026, while L3 to L5 rose from 50 to 61 percent. AI PM is about 46 percent of open PM postings and about 7 in every 100 PM hires. He's careful that fewer hires doesn't mean PM teams shrink, since promotions fill seats. His conclusion is the one I'd underline: "the outside door is narrower," and fewer people are entering PM at the bottom.

Anthropic's 27 to 9 percent is the same shape on the technical side. Lemkin made the uncomfortable point in his write-up that redeployment helps people already on payroll, not the ones who never got an offer.

Here's why that's the CPO's problem and not HR's. Every senior PM, every director, and every future CPO was once an L1 who got to make small calls with a safety net. If the whole industry stops hiring at the bottom, the bench is gone in about six years, and the people who'd have learned to spot slop by refining junior work never get the reps.

Atlassian made the opposite call. Sherif Mansour told Jason Lemkin the pipeline is now "weighted to both ends with less in the middle." Juniors are "up to 38% more likely to use AI." Seniors adopted more slowly "but are much better at spotting slop." That's a bench, built on purpose, with the pairing designed in. The Product Builder job ladder is where I'd start for the levels. The rest of hiring and leveling is a later part.

Next in the series: margin is product's job now.

One thing to try this week

Pull two counts. First, every product hire from the last 12 months, by level. Second, for each product manager who manages, what they shipped or decided last month against what they relayed. If the bottom row of the first count is zero, that's a decision somebody made by default. Make it on purpose, in writing, with your CEO, and say what the bench looks like in 2032.

Sources: Nathan Wilmers, Regina Catipon, Maxim Massenkoff, and Peter McCrory, "How employment is changing at Anthropic," Anthropic, October 16, 2026, anthropic.com. Jason Lemkin, "5 Interesting Learnings From Anthropic's Data on Its Own Workforce," SaaStr, October 19, 2026, saastr.com. Aakash Gupta, "The Definitive Report on PM Compensation and Hiring in 2026," Product Growth, October 22, 2026, news.aakashg.com. Rich Holmes, Department of Product, October 11, 2026, departmentofproduct.substack.com, for Spenser Skates's Amplitude figures. Jason Lemkin, "Atlassian's Head of AI: We Bolted AI Onto 20 Apps, Almost Didn't Ship Chat, and Flipped Our Hiring Toward Juniors," SaaStr, October 10, 2026, saastr.com, with Sherif Mansour.

Part of the running argument on Product Leadership: the product org chart was calibrated to a build cost that collapsed, and the CPO's job now includes deciding who enters the org at the bottom, not only how many sit at the top.

Related answer: Should product orgs still hire junior PMs when AI does the entry-level work?