Some AI startups are hiring almost as many PMs as engineers. Others are hiring none.

The traditional PM ratio looks surprisingly resilient

In 2007, Marty Cagan offered a durable rule of thumb: one product manager for every six to ten engineers. His original “roles and ratios” article became a reference point for software teams. A counter-tradition also persisted: founders and engineers carrying product decisions themselves. Index Ventures describes Stripe reaching more than 250 people before hiring its first PM.

Today’s hiring picture suggests that the traditional model still has life in it. Among companies hiring PMs, the median was one PM opening for every 4.5 PM-facing engineering openings.

Job openings are not the same thing as staffing, and our count focuses on PM-facing engineers. Still, the direction is notable: PM capacity does not appear to be shrinking. If anything, companies that continue to invest in the role look slightly more PM-dense than the familiar rule of thumb would suggest.

What has changed is not the disappearance of the old model. It is the arrival of two stronger alternatives alongside it: companies hiring PMs almost as densely as product engineers, and companies building products without hiring any new PMs.

Three models are competing inside AI

Of the 39 AI startups and scale-ups, 13 had no PM opening. The other 26 were hiring at least one PM and had a median hiring mix of 1:3.8. That median describes the full PM-hiring group. Underneath it sit two very different models.

Model 1: PM-dense product teams

Application and workflow companies had a median hiring mix of 1:2.7. Glean listed 12 PM openings against 14 PM-facing engineering roles. Sierra listed 19 against 43, Lovable four against ten, and Decagon ten against 29. At lower volume, Runway listed two PM roles and two engineering roles.

These companies are not treating PM as a thin coordination layer. They are adding PMs as a meaningful part of product-building capacity. For enterprise-facing companies such as Glean, Sierra, and Decagon, that may reflect the work of turning complex customer needs, integrations, and workflows into a coherent and differentiated product.

Model 2: A small PM layer inside a large engineering push

Model, infrastructure, and developer-platform companies had a median hiring mix of 1:7.7. Mistral AI listed three PM openings against 32 PM-facing engineering roles. Cohere listed two against 38. LangChain listed one against 29.

Perplexity is the clearest exception: an application company hiring like an infrastructure company, with one PM opening against 28 PM-facing engineering openings.

PM has not vanished from these companies. It is simply a much smaller part of the current buildout. The closer the product gets to an end-user workflow, the more room there may be for PMs to shape differentiation. The closer it gets to models and infrastructure, the more the next bottleneck may be technical capacity. There are exceptions in both directions, but the gap is large enough to deserve attention.

Model 3: Product building without new PM hires

Cursor, ElevenLabs, Character.AI, Cognition, Writer, Dust, Hebbia, Poolside, Hippocratic AI, fal, Clay, Wonderful, and Figure had no PM openings. Together, they still listed 181 PM-facing engineering roles.

Product work did not disappear. These companies were adding builders without adding PMs at the same time. Decisions may sit with founders, engineering leaders, designers, product engineers, forward-deployed teams, or PMs already inside the company. Cursor is the clearest high-volume example: 34 PM-facing engineering openings, no PM opening, and 121 public jobs overall.

A careers page cannot tell us who wrote the roadmap. It can tell us where the next hires are going.

What might be driving the split?

Job openings do not reveal the decision made in a leadership meeting. But the pattern points to three plausible forces.

1. 1

   When builders resemble users, engineers can hold product context for longer.

   Developer tools are often built by people who feel the problem themselves. Founders and engineers can make product calls directly for longer. In the wider sample, developer and product tools had a 1:8.0 median hiring mix, and 42% of the companies advertised no PM opening.

2. 2

   Enterprise AI creates a translation problem.

   Glean, Sierra, and Decagon have to turn messy customer workflows, integrations, and repeated requests into products that can scale. That creates obvious work for dedicated PMs. Enterprise software and data companies had a 1:3.7 median mix, and nine of ten were hiring at least one PM.

3. 3

   At the application layer, product judgment may be the moat.

   For application companies such as Lovable, model quality alone may not be enough, especially as AI labs move closer to end users. Choosing the right workflow, experience, and customer becomes a bigger part of differentiation. That helps explain why application companies in our sample hired PMs more densely than technical-layer companies.

These explanations are not mutually exclusive. They also point to a better question than “What is the correct PM-to-engineer ratio?” The useful question is where product judgment lives, and when a company decides it needs a dedicated person to own it.

PM presence and PM density are different decisions

Look only at the median hiring mix and AI companies appear unusually PM-heavy. Look only at the companies with no PM opening and the picture reverses. That is not a contradiction. It reveals two separate choices: whether to add a PM at all, and how densely to hire PMs once the answer is yes.

The broad AI startups, scale-ups, and applied-AI group had the most PM-dense median at 1:3.4, yet one-third of the companies advertised no PM opening. Enterprise software and large tech were less PM-dense, but 90% of companies in both groups were hiring at least one PM.

Enterprise software treats PM as a recurring function. AI startups and scale-ups treat it more like a strategic choice. Many either hire PMs densely or do not add them at all. That polarization is more revealing than any single ratio.

Company size changes the odds of seeing a PM opening, not PM hiring density

Companies with at least one PM opening had a median of 144 open jobs. Companies with no PM opening had a median of 23. Among the 30 companies with the fewest open jobs, 19 had no PM opening. Among the 30 with the most open jobs, only one had none.

Part of this is simple probability: a company with hundreds of openings has more chances to advertise at least one PM role. But once a company was hiring PMs, a bigger careers page did not make it meaningfully more or less PM-heavy. The relationship was almost zero, with a Spearman rank correlation of 0.03.

The disagreement continues in AI labs and big tech

AI labs are making different bets

OpenAI listed 14 PM and 163 PM-facing engineering roles. Anthropic listed 20 PM and 96 PM-facing engineering roles. xAI had 30 PM-facing engineering openings and no PM opening.

Anthropic’s current mix was more than twice as PM-heavy as OpenAI’s. xAI resembles Model 3 at lab scale: 30 PM-facing engineering openings and no PM opening. Frontier research does not lead automatically to one product-management model.

Scale does not produce a standard ratio

The same disagreement appears in large technology companies. Among the eight familiar names below, the mix ranged from 2.1 PM-facing engineering openings per PM at Meta to 8.7 at Apple.

Meta listed 28 PM roles against 60 PM-facing engineering roles. Apple listed 62 against 541. Both are enormous product organizations, but scale does not push them toward the same hiring mix.

What this means if you are interviewing at an AI company

The ratio matters less than which operating model the company has chosen. Ask who currently owns customer research, prioritization, and the roadmap; whether the company is deliberately PM-light or simply between hires; and what decisions the new PM would actually own.

The visible market also skews toward experienced, technical PMs. Across the sample, 69% of PM titles included a seniority or leadership marker; without Amazon, the share was 59%. Technical domains such as platform, developer, API, infrastructure, data, security, hardware, or cloud appeared in 21.1% of PM titles, or 18.6% without Amazon. These are title signals, not applicant requirements, but they show the work employers are naming.

The market is testing three models at once

PM is not disappearing. It is becoming a more deliberate organizational bet.

In one model, PMs turn customer complexity into differentiation. In another, a small PM team focuses a much larger engineering push. In the third, product judgment stays embedded in founders, technical leaders, designers, and hybrid builders while the company adds no PMs at all.

The market is now testing which of these models scales. We plan to repeat the analysis in six to twelve months and pair it with interviews about where customer research, prioritization, and roadmap decisions actually sit.

The question is not whether companies need product judgment. Every model has it somewhere. The question is where they put it, and when they decide it deserves a dedicated product manager.

How we counted

We analyzed 39,916 open roles listed on official company career pages and hiring systems on August 30, 2026.

The comparison includes 1,551 PM roles and 7,092 PM-facing engineering roles. We counted a role as PM when it owned product strategy, priorities, a roadmap, or meaningful feature trade-offs. We counted engineers when they built product capabilities for users or customers, including reusable platform work that a PM could help shape.

PM openings were concentrated in a few large employers. The ten companies with the most PM openings accounted for 1,000 roles, or 64.5% of all PM openings in the sample. Amazon alone accounted for 518, or 33.4%. That is why company comparisons use medians, giving every company one vote.

Research, data science, DevOps, security, support, quality assurance, internal analytics, and sales engineering stayed outside the ratio unless the role clearly owned product-building work. This matters because titles such as “platform engineer” or “forward-deployed engineer” can describe very different jobs.

Full descriptions were available for 99.8% of PM listings and 74.3% of PM-facing engineering listings. For roughly 1,800 engineering roles without a usable full description, we counted the role only when the title and function clearly indicated product-building work. That likely understates PM-facing engineering and makes some company ratios look slightly more PM-heavy.

We also reviewed a random sample of 200 roles classified as PM-facing engineering. We confirmed 195; the other five exposed repeatable issues that we corrected across the data.

We deliberately chose a mix of leading AI companies, large technology businesses, enterprise software companies, established product companies, and developer tools. This is a comparison of those 120 companies, not an estimate for the entire technology industry.

We use “AI startups and scale-ups” as a practical label for 39 independent AI businesses, regardless of age or size. Their current products center on AI applications, models, developer tooling, or infrastructure. Frontier AI labs and large technology incumbents are analyzed separately. For the within-AI comparison, we classified the 26 PM-hiring companies by their primary offering. The full membership of both groups is shown with the result.

Limits

• Openings measure current hiring, not current staffing.
• A company with zero PM openings may already have enough PMs or may have recently filled a role.
• Smaller companies advertise fewer roles, so the absence of a PM opening is weaker evidence when overall posting volume is low.
• A few new roles can change a smaller company’s ratio quickly.
• Career pages change continuously. These numbers reflect what was listed on August 30, 2026.

Company data and sources

Explore every company, check the official careers source, or download the role-level data behind the totals.