We taught computers how to talk to us. Now it might be time to let them teach us how to speak.
Brian Connelly
"On it."
Two words, and they may be the most successful phrase in the history of American business. The boss leans into the office with her coat still on and says, "Can you get the Feeney job moving?" and the office manager, without looking up from his clipboard, says, "On it." She leaves happy. He goes back to the six other things he was on. And nothing has happened. On what, exactly? Moving how far, by when, and how would anyone know? Nobody said, which means nobody can ever be wrong, which is the whole beauty of the thing.
Corporate English has a full vocabulary built for this, a set of phrases that sound exactly like agreement and commit you to absolutely nothing. "I'll try to get to it." "Let me circle back." "I'll take a look." "We're aligned." "Let's put a pin in it." Notice that nobody who says "let me circle back" has ever been seen circling back. There is no record of anyone, anywhere, removing the pin. "I'll try to get to it tomorrow" is my personal favorite, because it contains three separate escape hatches in seven words. Try. Get to. Tomorrow. Tomorrow, as every child knows, never actually arrives.
And the person saying it isn't lying. That's the part that makes it interesting. He means it kindly, and everyone who hears it, including him, knows it isn't quite a promise. Then Thursday comes, the customer sits down at the table, and there's no estimate, and everyone acts mildly surprised, the way people act surprised when it rains after a week of forecasts calling for rain.
The professional circlers-back
Most of what it costs to run a company is coordination, and most coordination is people asking each other for things and saying yes. A surprising share of those yeses are attached to nothing. So companies hired an entire layer of people whose job is to find out which yeses were real. We call them middle management. Their main tool is the status meeting, a gathering where people report on the promises they made at the last status meeting, and promise to have more to report at the next one. Then the minutes go up two levels, and somewhere on the way the yellows turn green. Nobody lied. The truth just got a little help on its way upstairs.
Talk isn't cheap
Everyone has been told that talk is cheap, usually by someone who wanted you to stop talking and start doing. The philosophers who actually studied it came to the opposite conclusion. J. L. Austin and John Searle noticed that a great deal of language doesn't describe the world, it changes it, and they called these speech acts. "I promise" creates an obligation that didn't exist a second earlier. "You're fired" doesn't describe anything either, and it changes quite a bit. Talk is how nearly all work in a company gets started, assigned and declared finished, and when it's sloppy it may be the most expensive thing the company does. It just never shows up as a line item.
In the 1980s Fernando Flores and Terry Winograd put that idea into software called The Coordinator, which treated work as conversations for action. Someone asks, someone agrees, the work gets done, and the person who asked says whether it was what they needed. The theory was right. The product died, because it asked every user to label every message before sending it. Is this a request? A promise? A counter-offer? It turns out people do not want to fill out a form in order to say yes to their boss. They would rather say "on it" and get on with their day.
I have a small stake in that history. Years ago I inherited a stranded multi million dollar workflow project that Flores's firm had analyzed, beautifully, and when the client asked them to build what they had mapped, the answer was that they did not do automation. I did. What I learned building it was that the theory was never the problem. The problem was the paperwork it put on the people using it.
Enter the coach
Language models take the paperwork away. A machine can now read "I'll try to get to it tomorrow" and recognize three escape hatches without anybody filling in a form. So here is the idea I've been working on: an AI coach that sits in the places people already talk, Slack, Teams, email, and pays attention to the asking and promising going on there.
It isn't a hall monitor. When someone writes "when you get a chance," the coach doesn't send a report to anyone. It asks that person, privately, a better question: Priya's meeting is Thursday at two, so is tomorrow real, or would it help to tell her which day is? You can ignore it. Plenty of people will. But a question asked at the right moment, by something with no ego in the game and no career to protect, gets a surprising number of people to say what they actually mean. And the coach keeps a record of the commitments that come out the other side: who asked, who agreed, by when, and what happened.
Which brings up the only question that really matters.
Who keeps the book?
A record of who promised what is worth exactly as much as people's belief in it. The day the executive who paid for it, or the administrator who runs it, can quietly change an entry, the ledger becomes one more place for office politics, and everyone goes back to trusting the meeting minutes, which is to say not much.
Get that right, though, and something odd happens to the company. The status meeting loses its reason to exist, because the status is already known, and nobody has to sit through an hour of people reading their own spreadsheets aloud. The executive finally gets the truth without it passing through three translators and a PowerPoint. The quiet person in accounting who has kept every promise she ever made for eleven years, and has never once been mentioned at an all-hands, turns out to have a record, and it is a better one than the vice president who gives the best presentations. "You never told me" arguments get settled in thirty seconds instead of three meetings. New hires can see who actually gets things done, which used to take about two years and a few bad lunches to figure out. And the customer starts getting dates that mean something, which may be the only part of this the customer will ever notice, and the only part they care about.
Keeping that book honest is not a new problem. It is the oldest problem in money. Whoever keeps the books has power over everybody whose name is in them, and for most of history the answer was to trust the bookkeeper, or hire a second bookkeeper to watch the first one, and then a third to watch the second, and eventually you have a bank. In 2008 the bookkeepers broke their commitments to just about everyone and were rescued for it. Satoshi Nakamoto put a newspaper headline about the second bank bailout into the very first block of Bitcoin, which is about as clear a note as anyone has ever left about why they built something. Bitcoin's answer was simple to say and hard to build: don't trust the bookkeeper. Let anyone check the books.
A commitment ledger doesn't need a cryptocurrency, mining, or a token with a cartoon dog on it. It needs one thing Bitcoin made practical, which is a way for anyone to confirm the record hasn't been altered without having to trust whoever holds it. Each entry gets a cryptographic fingerprint linked to everything before it, so changing an old entry breaks the chain and shows. Every so often a fingerprint of the whole book gets stamped onto the Bitcoin blockchain, which costs next to nothing and leaves permanent public proof that the book looked exactly like that on that day. The fingerprint can't practically be turned back into the contents, and each entry is salted with random data first so nobody can guess their way in, so the proof says the book is intact without saying a word about what's in it. Anybody holding a copy, an employee, an auditor, a lawyer on the other side of a dispute, can check it without asking permission from the company, the vendor, or me.
What we don't borrow from Bitcoin is the part where everyone can see everything. A company-wide public record of everybody's promises would be a surveillance system with good intentions, and those are the worst kind. So being able to verify the book is kept separate from being able to read it. Everyone can confirm it's intact. You can see your own commitments. Wider views are a matter of policy, and an executive looking for where things stall does not need to read every exchange between two people to find it.
People who follow Bitcoin will ask about quantum computers, so briefly: the quantum threat is to Bitcoin's signatures, the part that proves who owns a coin, not to the hashing that makes these fingerprints, which holds up even against the quantum attacks we know about.
Grandma, with a Harvard appointment
An append-only book sounds harsh until you remember that people renegotiate constantly, and telling someone on Tuesday that the Thursday estimate is going to be Friday is exactly the right thing to do. Nothing in the book gets erased; the renegotiation just goes in as a new entry. Over time the book shows who cleans up their commitments early and who lets them quietly rot, which tells you far more about a team than counting missed deadlines ever will.
Here's the part I enjoy. Michael Jensen, the Harvard economist whose agency theory helped launch the era of stock-option pay to keep executives honest, spent the later part of his career arguing, with Werner Erhard and Steve Zaffron, that what actually makes organizations work is integrity, which they defined as honoring your word: keep it, or when you can't, say so as early as possible and clean up the mess. It took one of the most cited economists alive, several decades, and a couple of corporate accounting scandals to arrive at what your grandmother told you when you were six. I don't say that to knock Jensen. I say it because he had the nerve to change his mind in public, which is more than most of us manage.
One honest limit. A stamp on the blockchain proves the book hasn't been changed. It doesn't prove the book is true. If the coach writes down what it thinks you meant, and nobody checks, you have a very secure record of a machine's guess. So nothing goes in the book until the person confirms it: "Was that a promise to have the draft by Friday?" Yes, or no, or "make it Monday." The book belongs to the people in it, or it isn't worth keeping.
Back to the office
The office manager in the opening isn't a villain, and neither is the estimator who said he'd try. They're busy people talking the way busy people talk, and the bill for all that looseness gets paid somewhere else, usually by a customer, usually later. The point of a coach isn't to make anyone talk like a contract. It's to ask the occasional better question, keep a book nobody can quietly cook, including the person who paid for it, and let people look at their own record and decide for themselves what it says about them.
So, the last time you said "on it." What were you on, and is it done?