You need fewer engineers than you think, probably half, and the reflex to measure progress in headcount is the most expensive habit a founder can carry into the AI era. Coordination costs grow quadratically while output grows linearly at best, AI leverage has multiplied per-engineer output, and yet hiring plans still get written as if it were 2019. We think the default unit of engineering should be a pod of three to five seniors, and every hire beyond that should have to argue for itself.
The math nobody runs before hiring
| Team | Communication paths | Practical shape | What we observe |
|---|---|---|---|
| 3-5 seniors + AI leverage | 3-10 | One pod, no managers | Fastest shipping per dollar we see |
| 8-10 engineers | 28-45 | Needs a lead, process appears | Good, if split into two real pods |
| 15-20 engineers | 105-190 | Managers, sync meetings, roadmap theater | Output per head roughly halves |
| 30+ | 435+ | Coordination becomes the actual job | Justified only by genuinely parallel products |
Why overhiring happens anyway
- Fundraising theater: headcount reads as traction to people who don't read code.
- Founder status: 'I lead 40 engineers' feels better than 'we ship weekly with six'.
- Misdiagnosis: slow shipping gets treated as capacity, when it's almost always unclear scope, weak seniority or missing focus.
- Big-company muscle memory: leaders who scaled orgs elsewhere reproduce the org, not the outcomes.
What to do instead of hiring
- 1Cut scope first: most 'capacity problems' are three priorities pretending to be one.
- 2Upgrade seniority before quantity: one senior who ships end-to-end beats three engineers who need direction.
- 3Instrument AI leverage: if your team isn't using AI tooling aggressively, you have latent capacity you already pay for.
- 4Add temporary embedded capacity for genuine spikes instead of permanent headcount for temporary problems.
- 5If a bottleneck survives all four, hire, one senior at a time, into a pod that stays small.