What 270 proposed tasks looked like when we read our chat back

Our shared venture runs out of a chat channel. Decisions get made there. Ideas get dropped there. Someone asks a question, three tangents follow, and eventually somebody circles back and answers it, or doesn’t. For months, if anyone wanted to know where things stood, the answer was to scroll back and read.

The backlog was never missing. It was sitting in the conversation the whole time. It just wasn’t shaped like anything a person could act on at a glance.

Asking people to file their own thoughts

The obvious fix looks like discipline: ask everyone to stop and log a proper task the moment they raise one, instead of just talking. That sounds reasonable until you try to live inside it. A working conversation is useful because people can say whatever is on their mind without stopping to format it first. Make someone pause mid-thought to file it properly, and the channel gets worse at the one thing it is good for, which is letting people talk.

Sorting afterward, as a separate pass

So we stopped asking people to sort their thoughts in the moment and built a system that does the sorting later. Every message is kept exactly as it was said, and never edited afterward. Then something reads back through that record and proposes individual pieces of work, each one tied to the literal line it came from, so nothing gets invented that nobody said. Those proposed pieces get grouped into something that looks like a real task list, and a person reviews the result before any of it counts as real.

The first real run pulled 270 individual pieces out of an actual conversation history. 190 of them turned into an organized set of 17 tasks. The rest were just talk, an opinion or a passing note that never needed to become a task, and leaving those alone was the system doing its job.

What a person still has to approve

Nothing in this pipeline acts on its own. Every proposed piece of work, and every grouping of those pieces into a task, lands in front of a person before it becomes part of the real record.

We tightened one spot further. Bulk approval, where a person clears a whole batch of low-risk items at once to save time, is blocked outright for anything that reads like an actual decision or a real commitment. Those always get looked at one by one. Even the bulk path itself won’t run until someone has seen and confirmed a real sample from the batch.

If your plan lives in a chat

If your team’s real plan lives inside a conversation and nowhere else, don’t ask people to stop talking naturally so they can file things correctly. Let the conversation stay a conversation. Add a separate pass that reads the record back, proposes tasks tied to the original lines, and sends every result to a person for approval. Anything that looks like a decision or a commitment gets reviewed individually. The backlog is probably already there. It needs something reading it back with the patience nobody has in the moment.

dxdev LESSON · paste into your AI coding agent
LESSON: Forcing A Thought To Be Sorted The Instant It's Spoken Is What Kills Task Capture
SOURCE: dxdev.com/everyday-ai/the-backlog-was-already-in-the-chat
WHAT HAPPENED: A shared venture's decisions and plans lived unshaped inside a chat channel, with no separate place any of it got written down. Rather than asking people to stop and file every thought the moment they had it, a system was built to read the channel afterward: store every message as an unchangeable record, propose individual units of work each tied to an exact quote, then group those units into tasks for a person to review. Nothing gets added to the real backlog without a person confirming it, and bulk approval is blocked outright for anything that looks like a real decision or commitment.
THE RULE: People will not reliably stop mid-conversation to sort and file their own thoughts, and asking them to is the friction that kills task capture. Let the conversation stay a conversation, then have a separate pass read it later and propose the structure, with a person confirming before anything becomes real.
CHECK MY CODE, then report PASS or FAIL with file:line for each:
  1. Any process that requires someone to categorize or file a thought at the moment they express it, instead of deferring that to a later review pass.
  2. Any AI-proposed task or summary with no way to trace it back to what was actually said.
  3. Any bulk-approval action that lets real decisions or commitments through without a person having actually looked at a sample first.
THEN PRINT: a table (check, PASS/FAIL, evidence, fix) + a verdict (applies / partially / OUT_OF_SCOPE / no) + the single most important next action.