Learn
Run parallel agent work without losing control.
Multiple tasks can start in separate windows, move through RAM
worktrees, and still converge into one clean line.
- Multi-task lanes
- RAM worktree flow
- Clean rebase before land
Future capability, not a day-one requirement.
RAM worktree: /AIT_RAM/ait/lt-0001
feature/lt-0001
agent brief
- 09:14Make the homepage feel more agent-forward.
- 09:28Keep the motion strong, but do not hurt readability.
- 09:37Open task and start implementing the hero interaction.
RAM worktree: /AIT_RAM/ait/lt-0002
feature/lt-0002
agent brief
- 09:18Push the Learn page toward a clearer multiple-task story.
- 09:31Add more agent presence without crowding the copy.
- 09:41Open task and implement the extra agent cues for Learn.
Convergence
verify each slice
rebase to current main
land one honest result
That is the promise: parallel work that still converges into one
reviewable line.
What you are learning
This page starts with control, not concurrency. The goal is to learn the
local loop that later lets you split work safely.
The developer role
State the need, narrow the scope, choose the next honest slice, and decide when the work is ready to move forward.
The agent role
Help shape the work, execute the slice, verify the result honestly, and report what still blocks completion.
Why the site starts local
- It keeps the first loop simple.
- It teaches workflow rhythm before shared coordination.
- It preserves human review authority.
- It avoids forcing server, database, or team setup on day one.
First-day rules
- Ask for shaping before implementation.
- Refine until the slice is honest and executable.
- Say “local-only” when you want the first loop to stay simple.
- Ask for verification before asking for land.
What comes later
Parallel task lanes, self-hosted control-plane routing, broader service
surfaces, and shared browser workflow views can come later. They are not
required to understand the first public learning path.
