Roadmap
Agent-native programming
The series, one property at a time. This page is navigation, not an archive — entries get reordered, renamed, and dropped as the work teaches us better. Substantial articles ship a companion Agent Skill when the workflow is concrete enough to execute and verify.
| # | Article | Companion artifact | Status |
|---|---|---|---|
| 0 | Hello, agents. | — | Published |
| 1 | Errors are recovery protocols | agent-recoverable-errors | Published |
| 2 | A CLI is an API | agent-native-cli | Planned |
| 3 | Safe retries need evidence | retry & idempotency skill | Planned |
| 4 | Plan, apply, verify | side-effect workflow skill | Planned |
Further out
- Long-running operations need handles
- Make capabilities discoverable
- Documentation agents can execute
- Repository context is a product surface
- Authentication without dead ends
- Observability for automated operators
- Dogfood: this blog on Run402
Editorial rule: each article makes one falsifiable design claim, shows a bad and an improved interface, examines at least one failure boundary, and leaves behind a contract or checklist. A companion skill ships only when the workflow is concrete enough to execute and verify.