The Chat Thread That Disappeared
Reviewed by Human · Updated October 19, 2026
Real world
A scientist runs an experiment, scribbles the key result on a sticky note, and slaps it on the monitor. Two weeks later the sticky note is gone, the monitor's been wiped, and the result is just... vapor. Compare that to a lab notebook: dated, bound, sitting on the shelf where anyone can open it next month and read exactly what was decided.
AI parallel
A decision you make in a chat thread is a sticky note. The agent that starts a fresh session tomorrow can't reach into yesterday's conversation. The repo is the lab notebook — the only surface the next session can actually open and read.
You told the agent 'don't touch the auth module, it's frozen' — in chat, on Tuesday. On Thursday a new session refactors the auth module. What happened?
Do AI Coding Agents Remember Previous Chats?
Short answer: no, not in the way you'd hope. A fresh agent session starts from whatever is loaded into its context — usually the repo's instruction files and the code — and not your earlier conversations. **System of record**: the single authoritative place where a fact is stored, such that anything else is just a copy or a rumor. **Ephemeral context**: information that exists only for the length of one conversation and vanishes when the session ends. So when a fact you care about lives only in an ephemeral chat thread, the next session is flying blind. This builds directly on what a harness actually is — the instructions subsystem only works if the facts it points to are written where the agent reads. It's also exactly why strong models still fail on real repos: the intelligence is fine, but the memory has nowhere to land.
Your turn
Look at your own current project (or the last one you paired with an agent on). List three facts that currently live ONLY in your head or a chat thread — decisions, constraints, or 'don't do X' rules the agent would have no way to know.
Reflect
Every fact you listed is a sticky note. The rest of this course is about turning each one into a lab-notebook entry.
You tell your agent in chat: 'Skip the /legacy folder, we're deleting it.' A week later a new session edits files in /legacy anyway. What's the root cause?
Go deeper
You've finished this module.
Mark it complete to earn your XP and keep your streak alive.
Progress saved locally · Sign up to earn XP