Before I wrote this post I had a conversation I did not expect to have.
I have two agents: a Science Officer and a Blogger. I was considering adding more. A Dev Lead, a BA, maybe one other I cannot recall now. So I did what felt natural. I asked them.
Both agreed the Dev Lead made sense. The reason was more useful than the answer. I had been giving both of them work that was not theirs. The Dev Lead was not a new capability. It was a missing boundary. Without it, coordination costs were bleeding into roles that should not be carrying them.
Then they wrote the job spec.
That is worth sitting with.
My existing agents identified the gap in the org structure, validated the hire, and defined the role. I approved it. That is the full hiring loop, except two of the three participants are not human.
This is Conway's Law running in a direction Conway did not anticipate. Systems mirror the communication structures of the organisations that build them. But what happens when the organisation is actively redesigning itself, and some of the designers are agents?
The question is not theoretical. Every person reading this who uses AI seriously is already inside it. You gave your agent a task. It hit a boundary. You worked around it. That workaround is your org chart. The boundary you did not design is shaping your system anyway.
The frontend/backend split existed because human brains could not hold the full context of a feature simultaneously. AI does not have that limitation. But we are still hiring, structuring, and coordinating as if it does. The structures we built for human communication bandwidth are now the ceiling on what AI can do for us.
The deeper question is not whether to add an agent. It is what adding that agent does to every relationship that already exists. Who does it talk to? What does it own? What were other agents doing that now belongs to it?
I got those answers before I made the decision. From the agents already in the room.
Conway's Law is not dead. It just got a lot more interesting.
