Why summaries do not change behaviour

Every chat tool now ships a summary button. Open a channel you have been ignoring, press it, and a tidy paragraph appears: the team discussed the migration, Priya raised a concern about the timeline, a decision was deferred to Thursday. It reads well. It is usually accurate. And a week later the migration is still stuck, because a summary is a description of the past and nothing in it points at anybody.

That gap between reading and doing is not a discipline problem. It is a design problem, and it is worth being precise about where the design goes wrong.

A summary answers the wrong question

When you open a backlog of unread messages, you are not really asking what happened. You are asking a narrower and more uncomfortable question: what does this cost me. Which of these threads is waiting on me. What did I say I would do. Who is going to ask me about it, and when.

A summary answers what happened. It compresses a hundred messages into eight sentences with the volume turned down evenly across all of them. The sentence where you promised to send the revised scope by Friday gets the same weight as the sentence where somebody shared a link. Compression treats every message as information. But messages are not all information. Some of them are debts.

The moment you compress a debt into prose, you strip the two properties that make it actionable: it has an owner, and it has a deadline. What is left is a description of a debt, which is a different object entirely, and one you cannot close.

Reading is not a commitment device

There is a second failure underneath the first. Even a summary that does mention your promise does not put you on the hook for it, because reading has no state.

Consider what happens after you finish reading. Nothing is different in any system anywhere. No record now says that you owe Priya a document. No process will check on Friday whether it arrived. The only thing that changed is that some information passed through your working memory, where it will decay at the usual rate, competing with everything else that passed through today.

Behavioural research on this is not subtle. Intentions predict behaviour poorly on their own. What closes the gap is a specific plan attached to a specific cue: not "I will send the document", but "I will send the document after the standup on Friday morning". Peter Gollwitzer's work on implementation intentions found the effect repeatedly across decades: people who form a concrete when and where follow through at much higher rates than people who simply intend to act.

A summary gives you neither the when nor the where. It gives you a paragraph.

The unit that works is an obligation, not a sentence

The alternative is not a better summary. It is a different object.

An obligation has structure that a sentence does not:

• An owner. Not "the team", but a person. Either you or somebody specific.
• A direction. You owe it, or you are owed it. These are not the same list and

they do not behave the same way.

• A due date, or an explicit note that no date was given, which is itself

useful information about how the promise was made.

• A link to the message it came from, so the context is one click away rather

than reconstructed from memory.

• A state you can change: open, nudged, done, dropped.

That last property is the one that makes the difference. State is what allows something outside your head to check up on you. A record that is still open on Friday can produce a nudge. A record that has been open for nine days without movement can be surfaced as a problem. A paragraph you read on Tuesday cannot do any of that, no matter how well written it is.

The half that goes missing entirely

There is a further asymmetry, and it is the expensive one.

Summaries, when they are action oriented at all, orient towards your actions. The output looks like a personal to-do list. But most of the delay in a working week is not caused by tasks you failed to do. It is caused by tasks other people failed to do for you, which you failed to chase.

Nobody keeps that list. It is socially awkward to write down, so it lives as a low-grade background feeling that something is outstanding. Then, some days later, a deadline arrives and the answer is that Marcus never sent the file, and Marcus is not being negligent. Marcus made the promise in a thread that scrolled away eleven days ago and no system anywhere retained it.

Tracking what you are owed is not about mistrust. It is about the fact that the other person is in exactly the same position you are: buried, well intentioned, and working without a record. A polite nudge in the original thread on the day it becomes relevant is the whole intervention. It just requires that somebody wrote it down.

What to take from this

If you evaluate a tool that promises to help with chat overload, ask it one question: after I use this, what exists that did not exist before.

If the answer is "a paragraph", the tool has saved you reading time, which is worth something but is not worth much. Reading was never the constraint. If the answer is "a list of specific obligations, each with a name, a date, a link and a state I can close", the tool has changed what happens next week, because next week something other than your memory is holding the record.

Summaries lose to the passage of time. Obligations survive it, because they were built to be checked.