Twenty Four

The studio runs on a small set of rules. They are boring on purpose.

Reversible work with no outside effect. Writing code, running tests, opening branches and deploying to staging.

Work with a real but recoverable effect. Merging to a shared branch, changing a schema or sending a draft for review.

Money, production deploys, secrets, authentication, public posts and anything that cannot be taken back.

A venture earns more build time by passing a gate, not by being interesting.

  • Gate A: pull

    Do strangers show up on their own?

  • Gate B: retention

    Do they come back after the first week?

  • Gate C: willingness to pay

    Does anyone pay, or clearly would?

Before an agent runs unattended on a codebase, the codebase needs tests and CI. Without a check that can fail, an agent loop repeats its own mistakes faster. So the harness comes before the autonomy, and each lesson from a failure becomes a new check.

No agent loop runs on code without a verification floor. Work is done when the tests pass and the output has been looked at, not when it was written.

Studio feed Simulated activity

An illustration of how the tiers and gates show up in daily work. Not live data.

  • researcher · Cozy Compute

    looked for signs of willingness to pay

    GreenGate C

  • reviewer · OpenHOA

    proposed a schema change for approval

    Yellow

  • researcher · TankTrue

    reviewed unprompted visits for the pull gate

    GreenGate A

  • builder · RunSite

    opened a branch for onboarding copy

    Green

  • builder · TankTrue

    queued a production deploy for the operator

    Red

  • researcher · OpenHOA

    reviewed return visits for the retention gate

    GreenGate B

  • reviewer · TankTrue

    requested a merge to the shared branch

    Yellow

  • researcher · RunSite

    reviewed unprompted visits for the pull gate

    GreenGate A

--:--