跳到正文
DEV Community· Fathy Shalaby·· 3 小時前AI 評分59

產品經理與工程師比例的常見範圍、基準及 AI 帶來的轉變

PM-to-Engineer Ratio: Typical Ranges, Benchmarks and the AI Shift

AI 導讀

文章整理產品經理與工程師的常見配置比例,指出多數軟件團隊約為每位產品經理對應 4 至 10 名工程師,常見參考值約為 1:6。Pave 對 4,195 家公司的數據顯示,員工少於 100 人的公司約為 1:5,超過 3,000 人的公司約為 1:11。文章指出 AI 編碼工具可能令產品決策成為瓶頸,但 1:0.5 只是單一團隊的提議而非行業基準,並建議按工作等待、返工及原型積壓等情況判斷團隊配置。

正文

Most software companies run one product manager per 4 to 10 engineers, from about 1:5 at small companies to 1:11 at large ones. Benchmarks by stage, how to calculate yours, and what AI coding tools change.

The typical PM-to-engineer ratio in software companies is one product manager for every 4 to 10 engineers, with about 1:6 the most common rule of thumb. Company size moves it: data from 4,195 companies shows roughly 1 PM per 5 engineers at companies under 100 people and 1 per 11 at companies over 3,000. AI coding tools are now pushing some teams towards 1:1.

Key takeaways

  • Common range: 1 PM to 4-10 engineers. Andrew Ng calls 6:1 a common Engineer:PM ratio and 4:1 to 10:1 typical (The Batch, 2025).
  • Bigger companies run leaner on PMs. PMs fall from 13% of R&D headcount at 1-100 employees to 7% at 3,000+ (Pave, 2024).
  • AI speeds engineering more than product decisions. 50% of product leaders report high AI impact on engineering work, against 17% on strategic planning and roadmapping (Product Institute, 2026).
  • The 1:0.5 story is one proposal, not a benchmark. Ng said he still did not know whether it was a good idea.
  • Measure the bottleneck, not the ratio. Engineers waiting on decisions, rework and unowned prototypes tell you more than a headcount fraction.

What is the typical PM-to-engineer ratio?

There is no official standard, but published data and practitioner surveys cluster tightly. Here is what each source actually measured:

Source Year What it measured Engineers per PM
Pave, R&D org benchmarks 2024 Headcount data, ~233,000 R&D employees at 4,195 companies ~5 (1-100 employees) to ~11 (3,000+ employees)
Andrew Ng, The Batch 2025 Rule of thumb from his work across companies 6 common; 4-10 typical
Synq, 50 scaleups 2023 Public LinkedIn profiles, grouped by title ~8 median (0.12 product people per engineer); over 15 at GitHub, Webflow and GitLab
Ken Norton, informal poll 2017 Practitioner answers on Twitter 5-9
Saeed Khan, Pragmatic Institute n.d. Author's experience 10-20 at most; up to 50 for mature, low-innovation products

Sources: Pave, The Batch, Synq, Ken Norton, Pragmatic Institute.

Only the Pave and Synq figures come from headcount data; the rest are expert estimates. Still, they agree on the middle: most software teams sit between 1:5 and 1:10. Below 1:4 is unusual outside early startups and AI-heavy teams. Above 1:15 usually means either an engineering-led culture or PMs spread too thin.

What does the ratio look like by company stage?

Stage Typical engineers per PM Who does product work Source or basis
Pre-product-market fit, under ~10 people No dedicated PM A founder acts as day-to-day PM Ken Norton
Small company, 1-100 employees ~5 First PMs, often one per product area Pave
Scaleup ~8 median; 15+ in engineering-led companies PM per team or surface Synq
Large enterprise, 3,000+ employees ~11 PMs, group PMs, platform PMs Pave
Mature, low-change products Up to 50 PM covers maintenance roadmap Saeed Khan (opinion)
AI-native teams in 2026 Moving from ~8 towards 1 Generalists who build and decide Andrew Ng, 2026

Pave also found engineering grows from about 70% of R&D headcount at small companies to about 80% at large ones, while data science stays at 4-6% throughout (Pave, 2024). Large companies add engineers faster than PMs. One plausible reason, which Pave does not measure, is platform and infrastructure work that needs less direct product management.

Ken Norton's rule for the earliest stage is worth keeping: before product/market fit, a founder should be the day-to-day PM (Norton, 2017). If you are deciding when to hire one, see our guide to the first product manager at a startup.

How do you calculate your PM-to-engineer ratio?

Divide the number of engineers by the number of people doing product management, counting by time spent rather than by title.

  1. Count engineers who build the product. Include product engineers, front-end, back-end, mobile and ML engineers. Decide once whether to include engineering managers, QA and platform engineers, and stay consistent.
  2. Count product management capacity in full-time equivalents. A full-time PM is 1.0. A founder who spends half the week on product decisions is 0.5. A designer or engineer who runs discovery one day a week is 0.2.
  3. Divide engineers by PM capacity.
  4. Check it per team, not just company-wide. A company average of 1:7 can hide one team at 1:3 and another at 1:20.

A worked example, with illustrative numbers. Say a 30-person SaaS company has 18 engineers, 2 PMs, a CEO who spends 40% of the week on product and a lead engineer who spends 20% on discovery:

  • PM capacity: 2 + 0.4 + 0.2 = 2.6 FTE
  • Ratio: 18 ÷ 2.6 ≈ 6.9 engineers per PM
  • By title alone it would be 18 ÷ 2 = 9, which overstates the gap

The by-title figure would push this team to hire. The capacity figure says it sits mid-range, and the real question is whether the CEO's 40% is the right 40%.

Why is the PM-to-engineer ratio changing in the AI era?

AI coding tools speed up building faster than they speed up deciding what to build, so product decisions become the constraint. Ng put it this way in July 2025: deciding what to build is the new bottleneck, especially early in a project (The Batch, 2025).

The survey data matches the argument. In Product Institute's 2026 survey of 309 product leaders, 50% reported high AI impact in engineering and development and 45% in design and prototyping, but only 17% in strategic planning and roadmapping and 9% in cross-team collaboration (Product Institute, 2026). Google's 2025 DORA report found 90% of about 5,000 technology professionals now use AI at work, and that adoption is now linked to higher delivery throughput (Google, 2025).

If engineering output per person rises and product decision capacity does not, each PM can support fewer engineers before engineers start waiting. That is the mechanism behind a falling ratio.

What did Andrew Ng actually say about 1 PM to 0.5 engineers?

At Y Combinator's AI Startup School in June 2025, Ng described a team that proposed one PM for every half an engineer, instead of the usual 1:4 (YouTube, 2025). He called the idea extreme and added: "I still don't know if this is a good idea" (transcript excerpt, HackerNoon). It was one team's proposal for one project.

By April 2026 his framing was more measured: some teams are pushing engineer-to-PM ratios from about 8:1 to as low as 1:1, and small co-located teams of 2 to 10 people who learn adjacent skills can achieve much more (The Batch, 2026). The more durable point is that the line between engineer and PM is blurring, not that companies should hire two PMs per engineer.

What are the limits of the "AI makes engineers faster" argument?

The speed-up is real for many tasks but uneven and easy to overestimate:

  • In Atlassian's 2025 survey of 3,500 developers and managers, 68% saved more than 10 hours a week with AI, but 50% lost 10 or more hours a week to organisational friction, and developers spent only 16% of their time coding (Atlassian, 2025).
  • In METR's randomised trial, 16 experienced open-source developers took 19% longer on 246 tasks with AI tools while believing they had been 20% faster (METR, 2025). METR now describes it as a snapshot of early-2025 tools.

If coding is only 16% of an engineer's time, even doubling coding speed frees less capacity than the headlines suggest. Teams whose bottleneck is review, testing or unclear requirements will not fix it by adding PMs or removing engineers.

How should AI change your team's ratio in practice?

Treat the ratio as an output of how work flows, not an input to hiring. Ask where work waits:

Signal What it suggests What to try first
Engineers or agents idle waiting for decisions or specs Too little product decision capacity A shorter decision loop and bounded briefs; then more PM capacity
Prototypes pile up with no owner or production path Building outruns deciding Weekly keep, kill or ship review; name an owner for each
High rework: features reopened because the goal was unclear Weak briefs, not necessarily too few PMs A shared brief reviewed with engineering before work starts
PMs writing specs nobody reads; many status meetings Too many PMs, or PMs doing coordination Fewer handoffs; give engineers direct customer access
Release quality and support load worsen as output rises Review and testing are the bottleneck Invest in QA, evals and release checks before PM headcount

What ratio should a 3-10 person AI-native team aim for?

Aim for enough product decision capacity that nobody waits on a decision for more than a day or two, rather than a fixed ratio. An illustrative example: a 7-person team has 5 engineers, 1 designer and a founder-CEO who spends half the week on product. That is 5 ÷ 0.5 = 10 engineers per PM-equivalent, at the lean end of the range.

If those engineers use coding agents and now finish a typical change in two days instead of five, the founder's 0.5 FTE of product time becomes the queue. The options, in order of cost:

  1. Reclaim product time from chores. Much PM time goes on assembling context rather than deciding. Our piece on reclaiming the craft of product management covers what to hand off.
  2. Make one engineer a part-time product engineer, owning discovery for one area at about 0.2-0.3 FTE.
  3. Hire a first PM once the founder cannot run weekly customer conversations and keep the product brief current. That moves the team to roughly 5 ÷ 1.5 ≈ 3.3 engineers per PM-equivalent, close to the AI-era direction Ng describes.

Tooling changes how many builders one person can support; the options are compared in the AI-native stack for technical PMs.

For a small AI-native team, the practical answer is usually not a new ratio but generalists: engineers who do discovery and PMs who can prototype. Each person should still say which hat they are wearing at each stage. The operating model is covered in AI-native product management for small teams, and the PM role itself in what an AI product manager does.

Why do engineers build things that don't match what the PM asked for?

Engineers build something different from what the PM expected when the brief fails in one of two opposite ways. It says too little about the user, the problem or the expected behaviour. Or it prescribes implementation before engineering has evaluated the approach. A ratio change will not fix either.

Before work starts, make these points easy to find:

  • User and situation: who is affected, and what are they trying to do?
  • Evidence: which customer conversation, support pattern or usage signal led here, and what is still an assumption?
  • Desired outcome: what should improve, with a baseline if there is one?
  • Behaviour examples: the main case and two or three important edge cases.
  • Constraints and non-goals: what must stay true, and what is deliberately out of scope?
  • Open decisions: what needs engineering, design or legal input before committing?
  • Verification: how the team will check behaviour and effect after release.

Review the brief with the engineer before anyone, person or coding agent, starts building. The AI coding-agent PRD template gives a copyable version. For engineers who want to own more of the discovery themselves, see customer discovery for product engineers.

Common mistakes when using PM-to-engineer ratios

  • Using a benchmark as a hiring target. A ratio describes a team; it does not tell you whether the team understands customers.
  • Counting titles instead of time. Founders, designers and engineers often do real product work. Count it.
  • Copying a big company's ratio. A 1:11 enterprise ratio reflects platform teams that need little product management. A 15-person startup has different work.
  • Forgetting design. Ken Norton prefers one designer per PM on user-facing surfaces. Adding PMs without design capacity moves the bottleneck.
  • Assuming AI makes every engineer equally faster. Measured gains vary widely by task and codebase.
  • Reading the 1:0.5 story as a trend. It was a single proposal that its own source was unsure about.

Where Specky fits

When product decisions are the bottleneck, the fix is usually faster access to evidence, not more meetings. Specky is an AI-native product system that reads customer calls, feedback, Slack, Jira and analytics into a Product Graph and drafts specs with the customer evidence cited inline, so engineers and coding agents can see why something is being built. Teams of 3-10 hiring their first PM can see Specky for startups.

Frequently asked questions

What is a normal product manager to engineer ratio?

A normal ratio is one product manager to between 4 and 10 engineers, with 1:6 a common rule of thumb. Pave's 2024 data puts it near 1:5 at companies under 100 employees and 1:11 at companies over 3,000. The right number for your team depends on product complexity, customer access and how much product work engineers and founders already do.

How many engineers per product manager should a startup have?

A startup before product-market fit usually has no dedicated PM; a founder plays the role. Once there are several engineers across more than one product area, Pave and Synq data put small companies and scaleups at roughly 1:5 to 1:8. Count the founder's product time before hiring, since it often covers more than the title suggests.

What does one PM to 0.5 engineers mean?

It means one product manager for half an engineering role, or two PMs for each engineer. Andrew Ng described a team proposing it at a June 2025 talk and said he did not know whether it was a good idea. It is a single proposal that illustrates the shift towards deciding being the bottleneck, not an industry benchmark.

Does AI change the PM-to-engineer ratio?

AI is pushing the ratio down in some teams because coding speeds up more than product decisions do. Ng wrote in April 2026 that some teams have moved from about 8:1 towards 1:1. The evidence on engineering speed-ups is mixed, though, so most teams should fix decision and review bottlenecks before changing headcount.

How many product managers do you need for 100 engineers?

At the typical 1:6 to 1:10 range, 100 engineers would usually need about 10 to 17 PMs. Fewer is reasonable if many engineers work on platform or infrastructure, or if engineers do their own discovery. More may be needed for many customer-facing surfaces or regulated products. Plan by product area, not by headcount alone.

What is the ideal ratio of designers to product managers?

Ken Norton prefers one designer per PM on user-facing surfaces, with a PM, designer and tech lead forming the core team. Many companies run fewer designers than PMs, especially on platform and internal products. If design work is queueing behind decisions, adding PMs alone will not speed delivery.

Sources

  1. DeepLearning.AI, The Batch (Andrew Ng) — AI Product Managers Will Be in Demand
  2. Pave — R&D Team Structure Benchmarks: Engineer-to-PM Ratios by Company Size
  3. Ken Norton, Bring the Donuts — The Ideal Ratio of Engineers to Product Managers
  4. Synq — Data and product to engineer ratio at 50 tech scaleups
  5. Pragmatic Institute (Saeed Khan) — You Can Never Have Too Many Product Managers
  6. YouTube (Y Combinator) — Andrew Ng: Building Faster with AI
  7. HackerNoon — Andrew Ng: Product Team Ratios Evolving to Just One Software Developer for Every Two Product Manager
  8. DeepLearning.AI, The Batch (Andrew Ng) — How to Get Through the Product Management Bottleneck
  9. DeepLearning.AI, The Batch (Andrew Ng) — AI-Native Software Development Needs Generalists
  10. Product Institute — State of AI in Product 2026
  11. Atlassian — AI adoption is rising, but friction persists (State of DevEx 2025)
  12. METR — Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity
  13. Google — How are developers using AI? Inside our 2025 DORA report

This story was originally published on 16 July 2026 at https://www.specky.space/blog/pm-engineer-ratio-inverted-what-it-means.

來源:DEV Community · dev.to