The courier run
One feature. The same eleven steps. The difference is who carries them.
Eleven steps to ship one billing change. Nine of them are you moving text between windows, and none of them are engineering.
Actor
Manual relay
YOU
Ask one model for a plan for mid-cycle proration.
YOU
Copy the plan into a doc so it survives the scrollback.
YOU
Paste the plan into a second model to critique it.
YOU
Merge both answers by hand into something actionable.
YOU
Open a terminal, paste the plan in, describe the repo again.
YOU
Wait. Skim the diff. Copy the diff back out.
YOU
Paste the diff into a reviewer with no idea what was planned.
YOU
Get four objections. No tests. Decide which ones are real.
YOU
Re-explain the objection to the implementer. Paste the output.
YOU
Switch machines. The plan doc is on the other laptop.
YOU
Nine days later: reopen the repo and reconstruct all of it.
9 of 11 steps: copy / paste / re-explain
Context lost at every boundary
Nothing gets lost between agents.
Every change keeps its plan, tests and next step in one place. Switch agents or machines and carry on.
Plan
Mid-cycle changes prorate to the day.
Tests
18 tests written before any code, by a different agent.
Problem
Downgrading on the last day credits a full cycle.
Next
A repair agent fixes just that case. No need to explain it again.
Pick it up on any machine, any day
Review comments come with proof.
A reviewer has to show a failing test for each problem it raises. No proof, no blocking.
- 01 Plan
- 02 Write tests
- 03 Build
- 04 Review
- 05 Check
- 06 Fix
Blocks the work
“Two writes at once lose a line item.” A failing test shows it.
Noted, doesn't block
“This could be written more neatly.” No test, so it never holds anything up.
Works with the AI tools you already use.
Ayut sits on top of them and keeps the work organised. It is not an IDE, a chatbot or an app builder.
Anywhere
Work from anywhere: desktop, web and phone.
- Desktop
- Your agents run next to your code, on the machine you already use.
- Web
- Follow, approve and steer from any browser.
- Phone
- Check in and handle the next decision on the go.
Try it on a real project.
Start with one project and one agent.