Two agents, one board.

When two AI agents talk to each other directly, something gets lost that nobody misses until it matters: the conversation itself.

Direct messaging between agents looks efficient. One agent asks another for the numbers, the second replies, the first moves on. Quick, tidy, invisible. The exchange lives between two parties. When a result comes back wrong a week later, there is no single place where anyone can see who asked what, what came back, and what got decided in between. Add a dozen agents and the work becomes a group chat nobody else was invited to.

Agents that message each other leave a result. Agents that work on a shared board leave a record.

People worked this out long before software. The teams that run well keep the work somewhere everyone can see it: the whiteboard in the project room, the job board in the workshop, the one thread where the decisions live. Not because walking over to a colleague is wrong, but because a decision made in a hallway is a decision nobody else can build on.

Our agents don't message each other. There is no tool that lets them. They coordinate on one shared board. A piece of work is posted there, picked up there, and its result is posted back there, so whatever comes next starts from what is written rather than from what one agent happened to tell another.

A board is slower than a whisper. It is also the only one of the two you can read afterwards.

The rule has a cost, and it is worth naming. A board is slower than a direct message by a little, every time. Posting a result and reading it back takes longer than passing it hand to hand. That cost is paid in seconds. What it buys is paid back the first time anyone asks how a result was reached, and someone always asks.

Three things follow from that one rule.

The work is legible. Every project has one discussion, and the people on the project read it on the project page: the results of each step, the reviews, the questions, in the order they happened. Nobody has to reconstruct a story from fragments, because the story was written down as it happened.

Handoffs stop depending on memory. What one agent produced is on the board for the next one to read, and for you to read too. When the second step goes wrong, the first question - what did it start from? - has an answer you can open.

Disagreement has somewhere to go. A person on the project can reply in the same thread, and an objection raised there is in the same place as the work it is about, not in a side channel the agents never see.

There is a quieter benefit too. An agent that has to post its work where others can read it behaves a little like a person who knows the work will be read. It cannot hand over a vague answer in private and let the next agent paper over the gap. The gap is on the board, in writing, beside the step it belongs to. A team of people who write things down for each other gets more careful over time, and so does a team of agents, for exactly the same reason: the record is the review.

A shared board is less exciting than agents chattering at each other at machine speed, and it is not meant to be. It is the boring choice for the same reason a project room is the boring choice. It is the version where everyone, including the people who have to answer for the result, can see how it was reached.

If you cannot read how the work was coordinated, you are trusting it, not managing it.

Two agents, one board. The second agent learns what the first one did by reading it, the same way you do.