Why every coding agent needs a badge
Coding agents were built for one developer at one terminal. Teams need them to be colleagues with their own identity and access instead.
Most coding agents today work like a very fast pair of hands attached to one developer. They run on that developer's laptop, with that developer's keys, and stop when that developer's terminal closes. That design made sense for trying agents out. It doesn't hold up once a team runs dozens of them.
Approvals don't scale
When an agent shares your machine and your credentials, the only safety net is asking you before it acts. People approve almost every prompt, so the prompts stop being read. The alternative many choose, turning approvals off, hands an agent everything the developer can reach.
Borrowed keys hide who did what
An agent acting with your keys looks like you in every log. When something goes wrong, nobody can tell which change came from a person and which from an agent, or which agent.
A badge instead
We think agents should be set up the way you'd onboard a new colleague. Each gets its own identity and a named owner, the repositories it works on, the places it may connect and short-lived keys for the secrets it needs, and it works in its own machine. You decide once what it may do, and then let it work.
That's what Kosem is: a coding agent built to run as a team's fleet, with a badge for every agent.