Meetings have tripled since 2020. AI note-taking apps won't fix it. Every AI tool you adopt makes it worse — because AI broke the math on team size, and nobody is talking about it. The solution isn't firing people. It's reorganizing into 5-person strike teams with 5-10x larger missions.
Three disciplines — evolutionary psychology, military readiness, and software engineering — all converged on the same answer: the human brain sustains deep, high-context coordination with about five people. AI didn't change that number. It changed the consequences of getting it wrong.
12 hours/week average, 16 for people managers, 23 for executives. Nobody can explain where the new ones came from. The stand-up became a slide session. The alignment session produces another alignment session. AI note-taking apps treat the symptom, not the disease.
5 people = 10 pathways (everyone holds the full map). 10 people = 45 pathways. 20 people = 190 pathways. Robin Dunbar's research (1992), US Army fire teams, and Fred Brooks' 1975 software engineering law all point to the same number: five.
Before AI, adding a 6th person gave diminished returns but manageable overhead. After AI, each person on a 5-person team generates $2-3M in value. The coordination cost of person #6 is no longer a minor tax — it's measured in millions of lost productivity.
Teams of 20 optimize for volume. Teams of 5 optimize for correctness. AI made volume free. Harvard Business School (2025): teams with AI were 3x more likely to produce ideas in the top 10% of quality — not 3x more output, 3x more likely to be right.
Most organizations have neither. They have oversized teams too slow for exploration and too diluted for precision execution, burning their best people on coordination overhead.
Zero coordination overhead. Maximum speed. The constraint is one person's judgment. Peter Steinberger built OpenClaw in ~60 days with 4-10 coding agents, in a language he'd never used. Fantastic for exploration — but scouts map territory, they don't build roads. Production needs more eyes.
Every AI output passes through at least one other brain sharing enough context to catch meaningful errors. A team of 5 covers product, engineering, design, data, and domain expertise. Below 5: blind spots. Above 5: silos. In a team of 5, there is nowhere to hide — which is exactly what you want.
A 500-person company just acquired the productive capacity of 2,500-5,000 people without hiring anyone. The correct response is not 'I can run with 50.' It's 'What was I previously unable to do?' A SaaS company with 400 engineers restructured into 80 strike teams could build a platform with 10 products.
The layers track Dunbar's biological constraints: 5 for a strike team, 15 (3-4 teams) sharing a domain, 50 for full working relationships, 150 for stable connections. Management thins dramatically. The taste layer — people defining and enforcing the standard of correctness — becomes the most important role.
In a team of 5, each person occupies 1 of 10 communication pathways. A mediocre contributor doesn't just underperform — their mediocre judgment amplified by AI generates verification burdens on everyone else. They make the team actively worse by consuming the team's most precious resource: shared attention.
Traditional SaaS runs below $500K per employee. AI-native companies like Lovable, Midjourney, ElevenLabs, and Cursor run $2.5M-5M. The math is clear: small teams with AI produce disproportionate value.
776 professionals at Procter & Gamble on real innovation challenges. Teams using AI were 3x more likely to produce top-10% quality ideas. AI broke functional silos — R&D and Marketing produced more balanced, integrated ideas.
Toby Lütke mandated every team prototype with AI before building. AI fluency in performance reviews. Teams must demonstrate why AI can't do a task before requesting headcount. Every forced prototype becomes a training ramp for strike team skills.
Give someone a real problem your company has been ignoring, full AI tooling, one week, zero check-ins. What you're testing: can they define the problem without a spec? Do they know what 'right' looks like at the architectural level? Do they default to action or to permission?
You didn't get a cost reduction — you got an army. The companies defining the next decade aren't cutting headcount to protect margins. They're keeping their people, reorganizing into strike teams, and going after missions 10x larger than what they settled for when headcount was expensive.
AI made volume free. Getting things right — polished, free of subtle errors that look fine in demos and compound into real failures in production — that's what separates teams of 5 from teams of 20. Volume masquerades as progress. Correctness is progress.
Every five-person team now has the capacity of a 50-person department. The question isn't how small you can get — it's how big your mission can become.