A prediction · Part four of The Feed Turns Around ·
Tab, Enter
A few messages into a project, the box is already full. Gray letters, written by the machine, saying what you were about to say. Two keys and the work goes on. It is the smallest version of the future I have been writing about, and it is already on my desk.
- Tab
- then
- Enter
You are three prompts into a project. The machine finishes its answer. You reach for the keyboard to type the next instruction, and it is already there, in gray, in the box: Now write the tests for the new endpoint and run them.
That is exactly what you were going to say. Maybe a little better. You press Tab and the gray turns white. You press Enter. The work goes on, and you did not write a word of it.
What you could have predicted, you did not need to say
In 1948 Claude Shannon gave the world a way to measure information, and the thing he measured was surprise. A message that was certain to arrive tells you nothing when it arrives. A message you could not have guessed tells you a great deal. The formula fits on a napkin: the information in an event is the negative log of its probability. When the probability is one, the information is zero.
Three years later he turned the idea on ordinary English. He had a person guess a passage of text one letter at a time and counted how often the guesses were right. Most letters were easy. After a q comes a u. After “the quick brown” comes “fox.” His conclusion was that most of written English is predictable from what came before it, and that the real information in a sentence is a small fraction of its letters.
Almost everything we have built since then is a way of squeezing out the predictable part. Compressed files. Phone keyboards that guess your next word. And now the language models themselves, which are, at bottom, machines for guessing the next piece of text.
The turnThe machine started guessing you
For a few years these machines guessed their own next word. The gray text is something new. It is the machine guessing your next word. Not its reply to you. Your instruction to it.
OpenAI put this into its Codex desktop app as a beta on October 9. After the model finishes a response, it can draft your next message from what is already in the thread and show it in the box. Tab drops the draft in, editable. Nothing is sent until you send it.
Put that next to Shannon. If the machine can write your next prompt and you agree with it, the prompt carried almost no information. What actually came from you was two keystrokes. Tab means: yes, that is what I meant. Enter means: go.
You did not become less important. You became concentrated. Everything that was yours in a paragraph of typing collapses into one keypress: keep it, or change it.
This is the part most people will miss. When the next move is predictable, your contribution shrinks to the one thing that is not predictable, which is the decision. The writing moved to the machine. The authorship stayed with you, packed into a key.
And the moments you do not press Tab, when you delete the gray and type something else, are the most informative things you will say all day. That is the surprise. It is the only part the machine could not have written for you.
Tab, Enter is Try it at the size of a sentence
In The Feed Turns Around I predicted that every personal AI app would become a feed: cards tagged IRL for what your AI did, SIM for what it would do, and under every SIM card one button, Try it. The gray text is the same idea, shrunk to one line. The machine reads the pattern, proposes the next move, and waits for a yes. A SIM card predicts your next job. The gray text predicts your next prompt. Tab, Enter and Try it are one gesture in two places.
Right now the gray text only knows the thread. OpenAI says the predictions use what is already in the conversation, not your memories and not your connected apps. That limit is what keeps it small. Take the limit away, let it see your email, your calendar and your calls, and it stops predicting your next sentence and starts predicting your next week. At that point it no longer belongs in a text box. It belongs in the feed.
You open your AI. There is no box. There are three cards. The contractor’s invoice came in above the quote, and a reply is drafted. Your daughter’s flight lands at 4:10, and a car is ready. The third one you did not see coming: it wants to move Thursday’s class to the bigger room, because the sign-ups doubled overnight.
You tap the first two without reading past the headline. You stop on the third. You had not thought of that. Not having thought of it is the whole point.
Where two keys go from here
Every AI input box opens pre-filled.
Within two years an empty chat box will feel the way a blank form feels today, like something failed to load. The default state of the input will be a guess, and blank will be the thing you have to ask for.
Tab and Enter become one key.
Two keys is one too many for a yes. For low-stakes work that stays inside the machine, accepting and sending merge into a single gesture. A separate Enter survives only where the action leaves the machine and touches other people, money, or anything you cannot take back.
The guess leaves the thread.
Once predictions can read memory and the apps you already use, they stop guessing your next sentence and start guessing your next job. That is the day the gray text turns into a card.
“Not this” becomes the most valuable thing you type.
The best assistants will learn more from what you delete than from what you accept, because a rejection is the surprise and an acceptance is not. They will make saying no exactly as cheap as saying yes.
People will be judged by what they refuse.
When anyone can produce the predictable next move with two keys, the visible skill is knowing when the prediction is wrong. The good operator is the one who stops on the third card.
It showed up on my own screen
I have spent the last few weeks writing about SIM cards and Try it as a forecast: the feed, the business model, how to ask your AI to build one. Then, a few messages into a project in OpenAI’s desktop app, the box filled in gray with what I was about to ask for. I read it. It was right. I pressed Tab, then Enter. That is when it stopped being only a forecast for me. Nobody announced the next phase of personal AI. It arrived as a shade of gray in a text box, and the only thing it asked of me was to agree.
The claim ledgerHow to check me
- Already true
- An OpenAI desktop app drafts your next message from the current thread, shows it in the input box, and lets you accept it with Tab as an editable draft. It launched as a beta in October 2026, on by default for eligible users, and nothing is sent without you.
- What has to happen
- The guesses have to be right often enough that Tab becomes reflex. Then they have to see past the thread, into memory and the apps you already live in, so the guess is about your life and not your last paragraph.
- Where I am probably wrong
- Being predicted can feel like being presumed. People were asking for prompt suggestions to be turned off long before this, and the switch is right there in settings. If the guesses are correct but feel pushy, people will turn them off and this stays a convenience for programmers. The difference between a welcome guess and an unwelcome one is manners, and the industry has not yet shown it has them. The deeper risk is mine to name too: if two keys start replacing thinking instead of typing, we stop producing the surprise that made us worth predicting. Manners are not the bottom of it, though. I go deeper in Add One Egg: manners are the egg, and underneath them is identity.
Back to the box. Gray letters, a cursor blinking at the end of a sentence you did not write. It is what you were about to say. Your part is no longer the sentence. Your part is the yes.
Tab. Enter.
Background
- Claude E. Shannon, A Mathematical Theory of Communication (1948).
- Claude E. Shannon, Prediction and Entropy of Printed English (1951).
- OpenAI, Composer Predictions, product documentation.
- OpenAI Developer Community, Codex can now suggest your next message (October 2026).
- John Rector, The Feed Turns Around.
- John Rector, Try It Is the Business Model.
- John Rector, Ask for the Feed. Let It Build the Cards.
Part four of The Feed Turns Around. Part one: The Feed Turns Around. Part two: Try It Is the Business Model. Part three: Ask for the Feed. Let It Build the Cards. Part five: Add One Egg.
1 thought on “Tab, Enter”