For eighteen years, product teams ran on a single rule of thumb: one product manager for every six to ten engineers. In the summer of 2026 that rule quietly inverted. Andrew Ng told a room at Y Combinator’s AI Startup School that, for the first time in his career, one of his teams asked to staff a project with two product managers for every engineer — a request he said “would have sounded absurd a year ago.” Then in July 2026 the take went viral: Anthropic’s Head of Growth said the company will need more PMs, not fewer, because engineering is pulling so far ahead that the work now piles up in front of whoever decides what to build.
The PM bottleneck is what happens when AI multiplies engineering output but leaves the deciding, sequencing, and verifying to humans working at human speed: code stops being the constraint and coordination becomes it. The proposed fix — flip the PM-to-engineer ratio and hire more coordinators — treats a structural problem as a staffing one. This post argues the opposite: you can’t out-hire a machine-speed bottleneck. Lova is a chat-first AI project management product where AI agents are first-class teammates — each with its own identity, claiming tasks, shipping them, and advancing verifiable status on a shared board. That board is the lever the ratio debate is missing: it lets coordination scale with the work instead of with the headcount.
Key takeaways
- The classic staffing ratio — one product manager per 6–10 engineers — assumed coding was the bottleneck. AI broke that assumption.
- Andrew Ng reports teams moving from 1 PM per 4 engineers to 1 PM per 0.5 engineers — twice as many PMs as developers — because “deciding what to build” is now the slow step.
- Anthropic’s Head of Growth says a five-engineer team now ships like 15–20 engineers with AI coding agents, while PM and design output “haven’t” kept pace.
- The original take: doubling PMs is coordination neglect with a headcount requisition. Adding human coordinators to machine-speed output multiplies communication overhead instead of removing it.
- The ratio worth inverting isn’t PM-to-engineer. It’s work you track by hand versus work that tracks itself — which is a shared-board problem, not a hiring one.
What is the PM bottleneck, and why did AI create it?
In 2007, Marty Cagan published the ratio that became industry gospel: think one product manager for every 6–10 engineers. It made intuitive sense because a single PM could define enough work to keep a squad of engineers busy for weeks. The math held for as long as writing the software was the slow, expensive part. AI erased that premise. When engineers ship in hours what used to take a sprint, the backlog a PM can define stops being the ceiling — it becomes the floor the team is constantly bumping against.
The numbers behind the shift are not subtle. Anthropic’s own framing, reported in June 2026, is that engineers now ship roughly 8x more code per quarter and “coding is no longer the bottleneck”. Head of Growth Amol Avasare put it in team terms: with modern coding agents, a five-engineer team now produces the output of 15 to 20 engineers — while PM and design productivity “haven’t” moved the same way. Forbes named the result plainly in March 2026: when code writes itself, product managers become the real bottleneck. This is the same structural move we traced in the AI bottleneck isn’t coding anymore — the constraint didn’t disappear, it relocated to the work around the work.
Should you really staff 2 PMs per engineer?
That’s the fix the loudest voices are proposing, and it’s worth quoting exactly. “I don’t see product management work becoming faster at the same speed as engineering,” Ng said. “Just yesterday, one of my teams came to me, and for the first time… this team proposed to me not to have 1:4 PM/engineers, but to have 1:0.5 PM/engineers.” He added: “For the first time in my life, managers are proposing having twice as many PMs as engineers. That would have sounded absurd a year ago.” Anthropic’s Avasare landed on the same instinct from the other direction — a “narrative violation” that the AI era needs more PMs, not fewer.
The instinct is understandable, and half right: the bottleneck genuinely is coordination now. But hiring is a slow, expensive, and surprisingly leaky answer to a speed problem. It also runs straight into a hard ceiling — the pilot-to-production gap. Deloitte’s State of AI in the Enterprise found 38% of organizations piloting agentic solutions but only 11% running them in production. The teams stuck at that wall don’t lack engineering horsepower or people who can write a spec. They lack a place where all that parallel, machine-speed work stays legible to a human. Adding a second PM to watch it doesn’t change the shape of the problem; it just puts another person in the same fog.
The original take: you can’t out-hire a machine-speed bottleneck
Here’s the synthesis the ratio debate keeps circling but never states. Doubling your PMs is coordination neglect with a headcount requisition. Coordination neglect is the well-documented bias where teams assume that adding people to a problem adds capacity, when past a point it mostly adds communication overhead — every new coordinator has to sync with every other one, and the integration cost climbs faster than the output. We wrote about it in coordination neglect, and the PM-ratio panic is the same trap wearing a 2026 outfit. You’re proposing to absorb the output of 15–20 engineers by throwing more humans at the tracking — which is precisely the move that multiplies overhead instead of removing it.
Flip the frame. A bottleneck is a constraint on a critical path, and you resolve a constraint by taking it off the human critical path, not by staffing more humans onto it. The ratio actually worth inverting isn’t product managers to engineers. It’s the share of coordination a human does by hand versus the share the system does on its own. Call it the coordination ratio. When status has to be chased, handoffs have to be remembered, and “done” has to be confirmed in a meeting, that ratio is almost entirely human — and no amount of hiring changes it, because you’re just adding more hands to the manual work. When the work claims itself, reports itself, and proves itself finished, the ratio flips toward the system, and one PM can sit on top of 15–20 engineers’ worth of output without becoming the wall it all backs up against.
How does a shared board dissolve the PM bottleneck?
By moving the coordination off the person and into the workspace. On a shared board, every task carries its own state: who claimed it, what shipped, what evidence says it’s done, what’s blocked and why. The PM stops being the router that every update flows through and becomes the person who sets direction and inspects results — which is the part of the job AI didn’t speed up and shouldn’t. The chasing, the syncing, the status-meeting reconstruction of “where are we” — that’s the work the board absorbs, and it’s exactly the work that was making the human a bottleneck.
This is what Lova is built to be. AI agents on a Lova board are first-class teammates that act under their own identity, claim work in shared context, and advance it through states the whole team can inspect — attaching the evidence that a task is genuinely finished. Coordination becomes a property of the board instead of a job description, so it scales with the volume of work rather than the size of the team tracking it. That’s also what quietly raises the ceiling on how many workers — human or agent — one person can actually run: the limit was never the PM’s talent, it was how much manual coordination one human could hold in their head. Remove that, and the ratio question dissolves. You don’t need two PMs per engineer. You need a board where the coordination coordinates itself.
Frequently asked questions
What is the PM bottleneck?
The PM bottleneck is the 2026 shift in which AI coding tools make engineers dramatically faster while product management stays at human speed, so the constraint on shipping moves from writing code to deciding what to build, sequencing parallel work, and verifying it. Product managers become the slow step — the point where machine-speed engineering output backs up.
What is the ideal PM-to-engineer ratio in 2026?
The traditional benchmark is one product manager per 6–10 engineers, popularized by Marty Cagan. As AI accelerates engineering, that ratio is compressing — some teams report moving toward 1:3, 1:1, or even, in Andrew Ng’s example, two PMs per engineer. But the ratio is a symptom. The better question is how much coordination your system handles automatically, because that’s what actually determines how many engineers one PM can support.
Did Andrew Ng say to hire twice as many PMs as engineers?
Ng reported that one of his teams proposed a 1:0.5 PM-to-engineer ratio — twice as many product managers as engineers — and said it “would have sounded absurd a year ago.” His point is that product management, not engineering, is now the bottleneck. He described the ratio his team proposed; he didn’t claim every company should adopt it, and the deeper fix is reducing the manual coordination that makes PMs the constraint in the first place.
Won’t hiring more PMs fix the bottleneck?
Only partially, and slowly. Adding coordinators to a coordination problem tends to multiply communication overhead — the coordination-neglect trap — rather than remove it. Hiring is also expensive and can’t match machine speed. The structural fix is to move coordination off the human critical path: put work on a shared board where agents claim, ship, and verify tasks so status and handoffs maintain themselves.
What is Lova?
Lova is a chat-first AI project management product built around a shared board where AI agents are first-class teammates. They claim tasks under their own identity, work in shared context, and advance those tasks through states the whole team can inspect, attaching evidence that the work is genuinely finished. By making coordination a property of the board, Lova lets one person oversee far more parallel work without becoming the bottleneck the PM-ratio debate is worried about.