Steer: agree the brief before the AI writes
Steer shows you what the AI guessed, so you can fix it before it writes.
Give an AI a vague request and it quietly picks the audience, goal and tone for you. Steer turns your request into a brief you can see. Its guesses show up as chips you can edit, and choices it shouldn't make alone are flagged for you to decide.
Live prototype: Steer. Data is synthetic, and the energy investor is a fictional company.
The problem: the AI guesses and doesn't tell you
The draft reads well, but it's built on guesses you never saw.
AI tools fill every gap in your request themselves. You only spot a wrong guess after reading the draft, when fixing it means asking all over again.
What the AI guessedclick one to change it
Your callthe AI won't pick this for you
Reader
-
You ask for something vague
"Write up the Q3 solar cost story for the board." It says nothing about the reader, the purpose, the format or the forecast.
-
The AI fills the gaps without telling you
It makes four or five decisions for you and doesn't mark any of them as guesses.
-
It writes a confident draft
The draft reads well, so nothing makes you look twice.
-
You find the wrong guess too late
It was written for strangers when you meant customers. Now you have to ask again and read it again.
The idea: a brief you can edit, between your request and the draft
Same AI, same request. What changes is whether someone filled the gaps on purpose.
Steer adds one step before the AI writes: a brief with its guesses marked as guesses, and the big choices left for you. It takes about as long as asking again, but it happens before the mistake.
Without steering and with it, side by side. Built live in the Minia design system, like the app above.
The AI's guesses, shown as chips you can edit
Each guess gets its own chip, marked as a guess. One click changes it.
Tone, length, format and sender each get a chip. Leave one alone and you've accepted it. Change one and the brief updates straight away.
The AI's guesses as a list you can edit. Built live in the Minia design system, like the app above.
Choices the AI shouldn't make alone are flagged for you
Some gaps aren't the AI's to fill, like who the reader is or what the piece is for.
Small guesses stay as chips. For the big ones, Steer gives you two options and says why it's asking. It won't write the steered draft until you pick.
An open choice, flagged for you. Built live in the Minia design system, like the app above.
Try it: compare the two drafts
With the brief in view, you can see why the draft says what it says.
In the app above, turn the request into a brief, change a guess and make the open choices. Then press Generate without steering to see what the same AI writes when it guesses. Or press Watch it happen for a guided run you can take over at any point.
How the design changed, from v1 to v6
Each version moved one more decision to before the AI writes.
It started as a bigger prompt box. By v6 the guesses were visible and the big choices were yours.
-
1
A bigger prompt box
More room to type, with hints. People still left things out, and the AI still guessed.
-
2
+ A summary of what it understood
It read back its understanding before writing. You couldn't edit it, but you could see the guesses.
-
3
+ Guesses you can edit
The summary became chips, so fixing a guess no longer meant asking again.
-
4
+ Flagged choices
Small guesses stayed as chips. Big decisions became a choice between two options.
-
5
+ No writing until you choose
While a flagged choice is open, the steered draft waits for you.
-
6
+ Side-by-side drafts · current
Run the same request with and without steering, and compare the two drafts.
The other three prototypes in the lab
Ground shows where an AI answer comes from. Oversee keeps AI agents in check. Recall shows what an AI remembers about you.