The previous post was about honesty. Why most cultures can't say what's broken. Here's what happens when they finally can: they discover the structure itself is the problem.
AI tools now let one developer do what used to require three. One designer can prototype, test, and iterate without waiting for engineering handoffs. One product person can ship a functional MVP in a weekend.
Individual productivity is exploding. Organizational productivity is not.
Faros AI tracked 10,000 developers across 1,255 teams.
The pattern: developers using AI write more code, complete more tasks, work on multiple things in parallel. Company-wide delivery metrics? Flat. A separate study found experienced developers were actually 19% slower when using AI on complex work in their own codebases. They expected to be faster. They weren't.
AI doesn't fix your organization.
It exposes it.
Where the leverage actually went
When individuals get faster but the whole stays the same, something is absorbing the gains.
Usually it's coordination overhead. Meetings to align. Reviews to approve. Processes to follow. Handoffs to manage.
Traditional product team structures assumed human-speed work. Everyone roughly equally productive. Coordination costs justified by specialization benefits. Cross-functional teams made sense when crossing functions meant waiting for people.
Now a single person with good tools can cross those functions in an afternoon. The coordination overhead stays. The justification disappears.
Meanwhile, at the other end of the spectrum:
Midjourney: $200M annual revenue, ~50 people. Cursor: $100M ARR in 21 months, ~20 people. Bolt: $20M ARR in two months, ~15 people.
These aren't outliers.
They're what happens when small teams with AI tools don't have coordination overhead to absorb their gains.
Your company isn't Midjourney. But the principle still applies. Every coordination layer that exists only to manage other coordination is overhead. AI doesn't eliminate it. AI makes it more expensive.
Anthropic, building some of the most sophisticated AI in existence, gives everyone the same title: Member of Technical Staff. Their experimental Labs team started with two people. CEO Dario Amodei calls talent density "the most important thing."
The feature factory trap
Most established companies can't operate like a 10-person startup.
But they can stop operating like a 1990s software department.
The standard product team model creates local optimization. Each team ships features. Each team hits OKRs. Nobody owns what spans teams: design language, architecture patterns, coding standards, shared infrastructure.
At casavi, we lived this. Feature teams delivered constantly. Building a coherent design system across products?
impossible. Resources went to features because features had deadlines and stakeholders. The horizontal work had neither.
We only made progress when we created a mission team. Five people, clear objective, fixed timeline, organizational backing. Not a committee. Not a working group. A team with one job and permission to finish it.
The feature factory couldn't solve a feature factory problem.
Team structures have evolved in waves. Each shift promised more agility, but coordination overhead grew with it:
The last shift isn't about methodology. It's about unit size and scope clarity.
What this looks like in practice
Mission teams for bounded problems. Small groups with clear outcomes, time limits, and the authority to ship. When the mission ends, the team disbands or reforms around the next one.
Platform teams for shared concerns. Design systems, infrastructure, tooling. Not governance committees. Teams that ship products other teams consume. If your "platform" is a Confluence page of guidelines nobody follows, you don't have a platform team.
Enabling over managing. People whose job is making others more effective. Not process enforcement. Capability building.
And critically: permission to use the tools. Many organizations restrict or ban AI tools while wondering why productivity isn't improving. Your people can't work at AI speed if policy keeps them at 2019 speed.
The test
Look at your current team structure. Pick any team. Ask:
Could three people with AI tools and a clear mission accomplish what this team does? If yes, what is the rest of the structure actually doing?
If the answer is coordination, ask the next question:
coordination of what? Work that needs to exist, or work the structure itself creates?
The honest answer usually hurts.
Companies that confront this will build faster than those still debating sprint lengths. The constraint isn't talent or tools anymore.
It's the org chart.
The org chart is the constraint. But changing it requires something most leadership teams avoid: honest conversations about which work actually matters. That's where this series goes next.
Sources
Faros AI: Measuring AI Impact on Engineering Teams
Study: Developer Productivity with AI Tools
Dario Amodei on Lex Fridman Podcast #452: Talent Density
Midjourney Revenue and Team Size
Cursor ARR Growth
Bolt: Zero to $20M ARR in Two Months