Writing
What we have had to work out while building Mirasim: parallelism, cost, evaluation, and where the boundaries of an honest wrapper are.
All pieces
Our own notes from building the product. The reading notes on other people's published research are a separate page: Blog.
- 2026Changing exactly one thing6 min read
Most tools that put a coding agent inside their own UI end up reimplementing it. There is a narrower move: launch the real CLI, and point one URL somewhere else.
- 2026Two agents, one repository6 min read
The moment you run a second coding agent on the same checkout, the two start overwriting each other. Linked git worktrees fix that, and any Mirasim session can start in its own.
- 2026Evidence, not a verdict from the model6 min read
An agent that reviews its own work grades the diff. What you actually need to know is whether a person could finish the task — and that question needs somebody using the thing.
- 2026Cheap per call, expensive per hour5 min read
Every individual model call looks too small to matter. An afternoon of four agents does not. A ledger kept on your machine shows where it went.
- 2026The name we changed5 min read
We shipped under "ADE" — agentic development environment — long enough that our own X handle still says it. Then we changed the words across the site to Agent IDE. Here is what the two names each get right.