Three is the magic number
AI agents make useful skills easier to reach. I reckon the smallest team that can own a real outcome is shrinking.
Yes, the title is a deliberate De La Soul reference.
For years, I thought about six people was the right size for a product team. Enough people to cover the skills. Small enough that everyone still knew what was going on.
A bigger team bought you access to more skills. It also bought you more meetings, more handoffs and more chances for nobody to quite own the result.
AI agents change that trade-off.
I can now reach a decent first pass at analysis, research, design, code and writing without adding another permanent role. The work still needs judgement. But access to the skill is cheaper.
I reckon the useful unit is shrinking to three people.
- Leader — owns the outcome. Decides what matters, what gets cut and whether the work actually worked.
- Subject matter expert — knows the job well enough to spot plausible nonsense.
- Engineer — builds the thing and wrangles the unit's agents.
That engineer role is changing fastest. I don't mean one developer quietly doing six jobs until they fall over.
I mean someone who can give agents context, tools, tests and boundaries. They can send work out, inspect what comes back and wire the useful bits together.
The subject matter expert matters more too, not less. Agents make it cheap to produce answers. Somebody still needs to know when the answer is wrong.
The leader stops the whole thing becoming a clever side quest. One person owns the result and has enough authority to keep everyone pointed at it.
Three people won't suit every problem. Some work needs deep specialist skills, independent checks or simply more hands. If there are six full-time jobs, hire six people.
But I think we should stop treating a six-person squad as the default unit of serious work.
Three humans, with the right agents around them, can now have a crack at things that used to need a much bigger team.
There is another reason I like the shape. Smaller, specialised cells can make a smarter organism. Each cell can stay close to its problem and call on agents for skills it doesn't keep permanently.
Push the biology metaphor too far and it gets silly. People aren't cells. But the organisational shape is interesting.
I also don't think the job titles survive every context. The leader is probably still a leader. The other two might be roles rather than titles.
If the engineer is really the person turning intent into working systems and wrangling the agents, what is that role in finance, legal, operations, marketing or service?
And what is the equivalent subject matter expert in each of those places?
The pattern might travel even when the titles don't.
What does a company look like when it is built from lots of three-person units?