Nope, haven’t seen the new movie yet..
I’m starting to think more and more about where we’re heading, especially with manager layers flattening and AI stepping into the junior engineer’s territory. AI can generate code, tests, even draft designs from specs—but those specs don’t magically appear out of thin air. A human has to write them, and write them well.
One
This is where 1 starts. The solo builder with AI is powerful, but let’s be honest: a team of one isn’t really a team. And even the best individual only has so much runway—productivity peaks, averages out, and is bounded by the reality of sustainable hours. No one codes 12 hours a day, every day, forever.
Two
So we quickly get to 2 - because as soon as there’s a second person, ideas can be bounced around, challenged, refined. And in the AI era, ideas matter a lot. Execution will increasingly be assisted, but deciding what’s worth executing is where humans are needed.
Two is stronger than one, but still fragile. If one person takes PTO, you’re back to solo mode (and maybe reviewing your own PRs while sipping a piña colada).
Three
Three feels like a sweet setup for an MVP - small, fast, tight alignment. But when two are out, the last one ends up holding the pager, the roadmap, and the existential dread all at once.
Four: The Sweet Spot
Four is different. Four feels like balance. With four, AI actually becomes a multiplier. You can run two workstreams in parallel, reviews don’t clog up, and all the big skills - frontend, backend, infra, data/AI - are covered without overloading anyone. PTO rotations work. On-call feels survivable. And when the inevitable “let’s sync with three other teams” meeting pops up, you can send one person without grinding progress to a halt - while still having enough capacity to spread the context back to the group. Knowledge-sharing has its cost, but at four it’s manageable without losing velocity.
The Spec-Driven Loop
The spec-driven loop looks like this:
human-written spec → AI generation → human review → deploy → feedback → updated spec
Notice the human bookends. Writing the spec and updating it based on real-world feedback are where humans are needed most. AI is brilliant at turning a spec into code, but it has no clue whether that code matches compliance rules, scales in production, or avoids waking up the SRE at 3 a.m.
Why Not Five+?
And why not five+? Because five is where the fun may start turning into overhead. Stand-ups get longer. More meetings sneak in. Suddenly there’s talk of sub-teams, “ownership areas,” and process. You’re not just solving customer problems anymore - you’re solving coordination problems.
Conclusion
That’s why the Fantastic 4 stands out. Small enough to stay nimble, big enough to survive vacations, outages, and endless cross-team syncs. At four, efficiency stays high, velocity stays steady, and the team delivers without getting dragged down by overhead.