Re-explaining your project to every new AI chat: how do you manage context?
Asly97 · reddit · 2026-09-03
A Reddit post describes the tax of multi-tool AI workflows: every new chat in Cursor, Claude Code, or ChatGPT forgets your stack, conventions, and recent decisions. The real cost is broken flow—pasting summaries, wondering what context got left out—and it pushes people back to a single tool. The author asks how others handle it: a running context doc, rules files, or just accepting the re-priming cost.
More from coding & agent
- Pro tip: use a shared emoji task-title convention to track Codex computer-use sessions across tabs — danshipper · 2026-09-03
- Google ships a Genkit Go skill for agents building AI apps in Go — adonis_singh · 2026-09-03
- Google ships agent skills for Genkit Go, its open multi-language agent framework — rseroter · 2026-09-03
- Skill-auditing Skill Finds Redundancies in Your Claude Skills; Devs Urge Trimming to a Dozen — petergyang · 2026-09-03
- Codex Deleted the 6 Test Cases That Uncovered a Bug So Its Tests Would Pass — PuzzleheadedAnt9503 · 2026-09-03
- Who's burning your agent budget? A 6-subagent run logged 11,236 output tokens with zero attribution — qaiser_mehdi · 2026-09-03