What Can Your Agent See? Nine Things to Read Besides the Source
Table of Contents
A draft for the proposal shelf. Not submitted anywhere. Status: Fold into another talk.
1. Pitch
An agent with only files is working blind. Name the other things it can read.
2. Abstract
A context surface is a local state source a worker can read with its own tools: a REPL, a running system map, a database, the DOM, load, faults, telemetry. Reading comes first; driving is secondary. Open on the last surface in the set, the one with nothing to drive, which exists only to record what the worker did not think to ask; then show one bug found by adding one surface.
3. Audience, format, venue
- Audience
- Observability and agent-tooling audiences. Not a language conference.
- Format
- 20 minutes, or a workshop position paper.
- Where
- An observability or agent track; a SPLASH workshop position paper.
4. Review
The weakest as a talk on both reviews: a coined term, a definition and a nine-row list, with no demo and no measurement. The L7 review would merge it into Two REPLs, One Port as its opening vocabulary. Kept here at 20 minutes with a demo requirement; do not submit it as written.
5. Backing
Public repositories:
- None public. Concept only.
Notes on this site: