Through the summer of 2026, one line kept surfacing in every conversation about AI and engineering: coding is no longer the bottleneck. It started as a boast about speed and quietly became a warning about structure. When AI writes the code, the constraint moves to the work around the code — deciding what to build, keeping parallel work straight, and verifying that any of it is actually done. That work has a name: coordination. Lova is the chat-first AI project management product where AI agents act as first-class teammates on a shared board — claiming tasks, posting evidence, and moving cards through verifiable status alongside humans — and it exists because the AI bottleneck stopped being code and became coordination.
Key takeaways
- Anthropic engineers now ship roughly 8x more code per quarter than in 2025, and the team’s own framing is that “coding is no longer the bottleneck.”
- With Claude Code, Anthropic’s Head of Growth says a five-engineer team now produces the output of 15 to 20 engineers — but product and design productivity haven’t kept pace.
- The result is a leverage asymmetry: engineering scaled on an exponential (AI) curve while coordination stayed on a linear (human) one. The gap between the two is the new bottleneck.
- The proposed fix — hire more product managers — treats a systems problem as a staffing problem. You can’t hire your way across an exponential gap.
- Our take: coordination has to move onto a system that scales the way code did — a shared board agents operate directly — not onto more people watching a chat thread.
Why is coding no longer the bottleneck in 2026?
Because the numbers finally caught up with the vibe. In June 2026, Anthropic’s Fiona Fung, who leads the Claude Code and Cowork teams, said on Lenny Rachitsky’s podcast that Anthropic engineers ship on average eight times as much code per quarter as they did in 2025, and put it plainly: coding is no longer the bottleneck. That’s not an isolated brag. Google’s 2025 DORA report, the most rigorous annual study of software delivery, found that 90% of technology professionals now use AI at work, up sharply year over year, with a median of two hours a day spent working alongside it.
When a whole profession gets that much faster at its core task, the question stops being “can we build it?” and becomes “are we building the right thing, and can we tell?” The 2025 Stack Overflow Developer Survey caught the tension precisely: 84% of developers now use or plan to use AI tools, up from 76% a year earlier — yet trust fell, and 66% say they spend more time fixing AI-generated code that’s “almost right, but not quite.” Output went up. Certainty went down. That’s the shape of a bottleneck relocating, not disappearing.
What is the leverage asymmetry between engineering and coordination?
Here is the original claim worth carrying out of this piece. AI didn’t remove the bottleneck; it created an asymmetry. Engineering got exponential leverage — the same person, with the same day, now ships several people’s worth of work. Coordination didn’t. Deciding what to build, sequencing dependencies, reconciling three parallel efforts, and confirming that a thing is truly finished are still done at human speed, in human meetings, over human attention. Put an exponential curve next to a linear one and the distance between them widens every quarter. That widening distance is the bottleneck everyone is suddenly feeling.
Anthropic’s Head of Growth, Amol Avasare, described the mechanism cleanly in a widely shared conversation with Lenny Rachitsky: with Claude Code, a five-engineer team now produces the output of 15 to 20 engineers, while PM and design productivity haven’t kept up. One product manager is now absorbing the coordination load of a team three to four times the size on paper. This is the same relocation we traced in why vibe coding solved the wrong bottleneck — except now there’s hard headcount math behind it, straight from the company with the most AI leverage on earth.
Can you fix the coordination bottleneck by hiring more PMs?
The instinct across the industry is to add coordinators. Andrew Ng has recounted teams proposing one product manager for every half-engineer — twice as many PMs as engineers, a full inversion of the old one-to-four or one-to-six rule of thumb. Anthropic’s own read is a “narrative violation”: we’ll need more PMs, not fewer. Both are honest reactions to real pain. Both, we’d argue, are treating a systems problem as a staffing problem.
Think about what adding a coordinator actually does. It buys you more linear capacity to specify, track, and verify. But the thing outrunning you is exponential. Doubling the slow side of an asymmetry doesn’t close it; it just delays the moment you notice you’re still behind — now with a bigger payroll and more people to keep in sync, which is itself more coordination. This is why the DORA authors, after studying thousands of teams, concluded that the biggest returns come not from the AI tools themselves but from the quality of internal platforms, the clarity of workflows, and the alignment of teams. AI amplifies whatever system it lands in. Drop it into coordination that runs on meetings and threads, and you amplify the mess.
Why does coordination break in chat and win on a board?
Coordination is failing in exactly the place we do most of it: the chat window and the recurring meeting. Chat is a consultation surface — a linear stream where the last message is the only state. It’s wonderful for asking and hopeless for tracking, because the moment three efforts run in parallel there’s no owner, no shared status, and no gate between “the agent says it’s done” and “it actually is.” That gap is expensive at human speed and ruinous at AI speed, which is why your AI coding assistant needs a project board: faster individual output was never the same thing as a faster team.
The org data says the same thing from a different angle. Microsoft’s 2026 Work Trend Index, drawn from 20,000 workers across ten markets, found that active agents in its ecosystem grew 15x year over year — and that organizational factors like structure and manager support account for 67% of AI’s real impact, more than twice the 32% attributable to individual effort. Only 19% of AI users land in the “Frontier” zone where capability and readiness reinforce each other. The capacity to produce is everywhere; the structure to coordinate it is rare. Capability got cheap. Coordination got scarce.
A board flips the surface. A task becomes a first-class object with its own state: it gets claimed, so there’s an owner; it carries its own context — the goal, the constraints, the definition of done — so whoever or whatever picks it up needs no backstory; and it can only reach “done” when the evidence is attached: the merged change, the passing check, the finished document. Watch what that does to the three jobs a coordinator actually holds. Deciding becomes creating a card. Tracking becomes reading a column instead of reconstructing a thread. Verifying becomes the status transition itself. Run five agents or fifty and the board holds them all in parallel — the exact shape the Anthropic numbers say work is taking.
How does an AI-native board make coordination scale like code?
The reason engineering scaled is that code lives in a system built for scale — version control, automated checks, structured state. Coordination never got that. It stayed in prose: the standup, the DM, the “quick sync.” The way out of the leverage asymmetry is to give coordination the same treatment — move it onto a system agents can operate directly, so coordination capacity grows with the agents instead of with how many coordinators you can hire. That’s the bet behind Lova. It’s chat-first because that’s where people start — you describe what you want in plain language — but it resolves into a board, because that’s where delegated, parallel, verifiable work can actually be seen. When the agents themselves claim cards, post evidence, and move status, the coordination scales on the same curve as the output it’s trying to keep up with.
This is the piece most 2026 AI strategies are still missing. The teams pulling ahead won’t be the ones with the most AI leverage in engineering — nearly everyone will have that. They’ll be the ones who noticed the bottleneck move and rebuilt the coordination layer to match, rather than throwing linear headcount at an exponential problem. It’s the same argument we made in your company has AI agents and nobody manages them, now with the receipts: the constraint has moved, and coordination is where the second half of 2026 gets won or lost.
Frequently asked questions
Is coding really no longer the bottleneck?
For teams with heavy AI leverage, yes. Anthropic reports its engineers ship roughly eight times more code per quarter than in 2025, and 90% of technology professionals now use AI at work according to Google’s 2025 DORA report. When writing code gets that fast, the constraint shifts to deciding what to build, coordinating parallel work, and verifying it — the coordination layer.
What is the leverage asymmetry?
It’s the gap between engineering, which got exponential leverage from AI, and coordination, which is still done at human speed. Anthropic’s Head of Growth notes a five-engineer team now produces the output of 15 to 20 engineers while PM and design productivity haven’t kept pace. The widening distance between those two curves is the new bottleneck.
Why won’t hiring more product managers fix it?
Because it adds linear capacity to a problem growing exponentially. More coordinators buy time but don’t close the gap, and more people to keep in sync is itself more coordination. The DORA research points elsewhere: the biggest AI returns come from better platforms, clearer workflows, and team alignment — systems, not headcount.
What is Lova?
Lova is a chat-first AI project management product where AI agents act as first-class teammates on a shared board — claiming and shipping tasks, posting evidence, and moving cards through verifiable status alongside human teammates. It’s built so coordination scales the way engineering did: on a system the agents operate directly, not in a chat thread someone has to babysit.
How is this different from a normal project management tool?
Traditional tools assume a human is staring at a dashboard, typing updates. An AI-native board treats agents as participants: they claim work, attach the evidence that a task is actually done, and advance status through explicit, verifiable transitions. That turns coordination from a meeting into a system — which is the only way it keeps pace with AI-accelerated output.