The suggested next prompt

After a turn ends, the prompt box sometimes shows a line of grey text with a small Tab key beside it. That is the next prompt as the tool itself guesses it, for example "run the tests now" after it wrote the code. Press Tab and the words become real text in the box, ready for you to change or send.

Partly built. Claude Code and Grok give suggestions. Codex has nothing like it, so a Codex session never shows one. The studio also does not write suggestions of its own.

Take it or leave it

• Tab or → puts the suggestion into the box. The cursor stands at its end, so you can add words or cut some before you send.
• Start typing anything else and the suggestion goes away. It does not come back for that moment, even if you empty the box again.
• Enter on an empty box sends nothing. A suggestion is never sent by one key press, so a prompt cannot go out by accident.

If you leave the session and come back with the box still empty, a suggestion you never touched can be there again.

When you will see one

The tool says a suggestion only when it has one obvious next line. After an error, in plan mode, or when the next step is not obvious, the tool often says nothing, and so does the studio. In a trial with Claude Code, two turns out of three gave one; with Grok, one out of three.

The suggestion is drawn only while the box is empty, nothing is attached, the session is not working or waiting for you, and no prompt of the line is being edited. A suggestion of more than one line, or of more than 300 characters, is not shown. Only sessions the studio runs get one; a session of VS Code or a terminal does not.

What it costs

It is not free, and the studio does not say otherwise. Each suggestion is a short request to the session's own model, and it counts in your usage. The request reads the conversation from the prompt cache, so its cost grows with the size of the conversation, and a turn that gets no suggestion still pays for the request. In one trial, with a conversation of about 43,000 tokens, each turn's request read around 43,000 to 45,000 tokens from the cache. At list prices that was between one and two cents per turn. On a subscription it is usage of your limits instead of dollars.

If that is not worth it to you, turn the feature off.

Turn it off

1. Open Settings.
2. Under How sessions work, set Suggestion of the next prompt to Off.

With it off, nothing is asked. For Claude Code that counts from the next session you start, because the tool is told at its start. For Grok it counts at once. What already stands in a box stays until its session's next prompt.

More on the box in the prompt box, and on what each request costs in how AigoPilot saves tokens.

Small tips

Questions