The scope of ai-augmented-agents
Shape: scope table
The table below is copied from the front page of ai-augmented-agents as it stood on 2026-11-03. The member owns it; if it changes there, this copy is a record of the example, not of the member.
Scope
| What | Why | |
|---|---|---|
| in | delegation: when handing a step to another agent beats doing it, and what the brief must hold for a cheaper agent to be right | the maintainer's rule is to spend judgment, not volume; the rule needs a measure |
| in | coordination: how several agents on one family of repositories keep one place per fact and one decision per question | six sessions built six repositories in one day; what they did is the first data |
| in | incentives: what a person and a model each optimise, and the game they are in, from the principal-agent problem to team theory and mechanism design | economics has the models; whether a model agent has an incentive at all is open |
| in | learning: when several agents improve at coordinating, and what multi-agent reinforcement learning says about it | the repeated game, with and without a model in it |
| out | the one-agent case | ai-augmented-project |
| out | building or training the agents' models | others do it; this arranges them |
| out | the software | ai-augmented-ide, later |
How to read it
- Each in row is a noun phrase and a reason, so a neighbour can compare its own rows with it.
- The first out row links the member that holds the one-agent case.
- The second out row says that others do the work, which is the other way an out row may end.
- The last row links a member that is not built yet; its Why is the link and one word.