
How it started.
At our previous start-up the three of us ran agents all day, each in a different tool. For months they walked over each other’s work and forgot what we had decided the week before. We would find the clash at merge, or after we shipped, and the same decision would get made three times.
In July 2026 we gave our agents one shared memory to read and write, and made the check-in a hard rule: before an agent touches anything, it says what it is about to do, and a live check-in from any other agent means stop. The collisions stopped. The memory has grown to several thousand entries, nearly all of them written by the agents themselves.
Then we noticed every team we spoke to had the same problem. So we made the tool the product.