Language is how commitments get made

The ordinary picture of talking at work is that it moves information around: you know something, you say it, now I know it too. That picture covers a surprising amount of what is said and misses the part that matters, because most of what people say to each other at work is not describing the world. It is changing it.

Some sentences do things

"The invoice is overdue" describes something; it is true or false, and you can check. "I'll have it to you by Thursday" describes nothing. Before it was said, nobody was owed anything; after, somebody is. The sentence did not report a fact about the world, it added one — and no amount of checking the world beforehand would have told you whether it was true, because the saying is what made it so.

This is J. L. Austin's observation, in lectures published as How to Do Things with Words in 1962, and John Searle's after him: a small number of things people do with sentences, sorted by what the speaker is up to. Some state how things are. Some try to get somebody to do something. Some put the speaker on the hook. Some, said by the right person in the right setting, simply make it the case — the meeting is adjourned, you are hired, the release is cut.

Two of those categories carry organisations. The one that asks and the one that undertakes. Everything a company calls coordination is built out of them, and neither one is information.

An answer that cannot be checked later is not an answer

A promise arrives with terms attached, whether or not anybody says them out loud: what will have been done, by when, and how the person who asked will recognise it. The terms are the substance. Without them there is a warm feeling and no way to tell, later, whether anything was kept — and so no way to be let down honestly, which sounds like a small loss and is not.

The tradition calls these the conditions of satisfaction, and the word satisfaction is doing real work in the phrase. They are conditions on the customer being satisfied, not on the performer feeling finished. The two come apart constantly. Most of what gets called a quality problem was a conditions problem that nobody had the conversation about, usually because having it in advance felt like distrust.

The conversation for action

Fernando Flores, in a 1982 dissertation at Berkeley, and then Terry Winograd and Flores together in Understanding Computers and Cognition (1986), put the two acts into a shape: a conversation between exactly two parties, opened by a request or an offer, closed only by the person who opened it.

The reason it is a loop and not a line is the closing. Work does not finish when the person doing it stops; it finishes when the person who wanted it says so. Between those two events sits the most expensive gap in most organisations, and a straight line has nowhere to draw it.

Four moments, then, each one belonging to a named person: the asking, the answer, the report, and the acceptance. The interesting part is not that path. It is that at every moment there is more than one thing that can be said. The answer may be no, or may be a counter-offer. The report may be a withdrawal. The acceptance may be a refusal to accept. These are not errors in the process — they are the process, used properly by people who are still able to say what is true.

And it names what has no move at all: silence. There is no act for having been asked and saying nothing. That is deliberate. Silence is the commonest failure in every organisation anybody has ever worked in, and the honest thing a notation can do about it is refuse to draw it as an answer.

Breakdowns are where the work becomes visible

Winograd and Flores took a further idea from Heidegger, and it is the one that changes how you read a stalled project. Most of the time you do not notice the tools you work through — you notice the work. The tool appears as a thing in its own right at the moment it fails. The hammer becomes a hammer when the head flies off.

Coordination is like that. Nobody thinks about who owes what to whom while it is going well; it appears as a subject on the day something is late and three people discover they each believed somebody else had it. That day is not an interruption of the work. It is the only moment the structure of the work is available to be looked at — which is a good argument for drawing it, and a better argument for drawing it before that day rather than after.

The objection worth taking seriously

The first serious software built on this, in the mid-1980s, asked people to label their messages: this one is a request, this one is a promise, this one closes that one. It was not loved. Lucy Suchman's Do Categories Have Politics? (1994) put the case against sharply: a system that makes people classify their own speech hands whoever runs the system a new kind of leverage, and the categories, for all their air of neutrality, are somebody's categories.

That objection has not gone away and should not be waved off. The reply we would make is narrow. There is a difference between a system that makes every sentence you type at work into a tracked commitment and a notation you use, deliberately, to look at how a piece of work is actually coordinated. The first is surveillance with a theory attached. The second is a drawing, and it can be wrong, and you can throw it away.

This is a drawing tool. Nothing here watches anybody, and nothing is anybody's system of record. If what you draw makes an accountability visible that people would rather stayed vague, that is the tool working — but the choice to look is yours, and so is the choice to stop.

What the notation keeps

Two parties per commitment, never more. Four moments, each owned by one of them. Every alternative act drawn as a path rather than an error state. Conditions of satisfaction as the substance of the promise. Closure by the customer alone. That is all of it — and it is enough to show, on one page, most of the ways work goes wrong between people who are all doing their best.

What it does to software follows the same idea into schemas and workflow engines, where it has a forty-year history and a couple of famous failures.