Say what you want.
Your assistant asks the rest.
Press New mission: the project's assistant interviews you one question at a time, until it knows what done means. The brief writes itself next to the conversation. Then the work is split into tasks that start one after the other when you click Launch.
For people who do not code, and for developers. It runs on the AI plan you already have, through Claude Code, Codex, Antigravity CLI or another agent CLI.
How will we know it is finished?
- The page is online
- The client approved it
- Other
- Goal
- Done when
- Constraints
- Decisions taken
- Open questions
- Write the text
- Find three references
- Prepare the offer
One question at a time, the page written live, then a card to validate.
Most work handed to an AI agent fails on what was never said: what finished looks like, what must not change, what already exists. An agent told to get on with it fills those gaps with guesses, and you find out at the end.
A mission interview closes the gaps first. The assistant asks, you answer with a click, a few words or your voice, and every answer lands on a page you can read and change. When nothing is open any more, it proposes the work, and the app starts it only after your click.
How it works
From a sentence to tasks that run, in six steps.
- 1
Press New mission
On the screen that asks what you are working on today, or in the Mission tab. On the phone, behind the + button. The project's assistant takes it from there.
- 2
One question at a time
Cards with clickable options, your assistant's advice marked. Click, type or dictate. It rephrases what it understood before moving on.
- 3
It asks what a brief usually misses
What done means, the constraints, the deadline, and what already exists. It stops as soon as it could brief a colleague.
- 4
The page writes itself
Goal, done when, constraints, decisions taken and open questions, in the Mission tab of the right panel. Edit it, annotate it, read its earlier versions.
- 5
One card to validate
Each task with its assistant, model, effort, skills and finish line. Launch, Change or Cancel. The card is written by the app from the list, so what you read is what gets created.
- 6
One task after the other
The tasks land on your board under one mission. Each starts when the one before closes, and the assistant sums up at the end.
What you get
A definition of done
The question most briefs skip. The assistant keeps at it until the answer can be checked, for the mission and for each task.
A page you can touch
Written live in the Mission tab. Versions are kept, a dot marks what you have not read, and a change of yours is never overwritten.
Your world, asked once
The first interview fills a My world page: who you are, your activity, your clients, your tools, your words. Every assistant of the project reads it before working.
The right assistant and model
Each task goes to the role that does that work, with a model and an effort sized to it, and with the saved skills that apply. You see all of it before launching.
Nothing without your click
The assistant can only propose. The tasks are created and started by the app, from the card, after you press Launch.
Tasks in order
Each task waits for the one before it. Close the mission on the board and the chain stops.
On your phone too
Start a mission, answer the cards, read the page and follow the tasks from the iOS and Android app, while your computer does the work.
On any agent CLI
Questions, page and validation go through the same tools on Claude Code, Codex, Antigravity CLI and the other supported CLIs.
With or without code
If you do not code
Plain words, no jargon, no project to set up. Say what you want done for a client, a launch or a report, and the Writer, the Researcher or Finance take their part.
If you do
The same interview at the start of a feature. Tasks go to your frontend, backend or QA agents, sized by the same rules as your tickets, and land on your backlog with their links.
What to expect
- The work runs on your computer. The next task starts by itself on the computer where you pressed Launch, while the app is open. From elsewhere, Start next on the board does it.
- The interview and the tasks use your own AI subscription, like any agent you open.
- A task moves on when it is closed. Its assistant is asked to close it once its finish line is met; if it stops before, the task waits for you.
- Tasks run one by one, not in parallel. A mission holds 12 tasks at most: past that, split it.
Frequently asked questions
What is a mission interview?
A guided conversation with your project's assistant. It asks one question at a time, writes the answers on a page next to the conversation, then proposes tasks with who does them and how you will know they are finished. You launch them in one click.
Can anything start before I validate?
No. The assistant can only propose the list. The tasks are created and started by the app, after you press Launch on the card. Change and Cancel create nothing.
Is it the same as an interview or grill-me skill?
The idea is the same: be questioned before the work starts. A skill stays in the conversation. Here the answers become a page you can edit and find again, and the end of the interview is a list of tasks that actually start, one after the other.
Which AI tools does it work with?
Claude Code, Codex, Antigravity CLI and the other agent CLIs AgentsRoom supports. The interview runs on the assistant of your project, with the subscription you already have.
Do I need to know how to code?
No. In the Non-dev mode the assistant uses plain words and the tasks go to business roles such as Writer, Researcher or Finance. Developers use the same button in their projects.
Can I change the page while the assistant writes?
Yes. Edit it or annotate a passage for the assistant. Earlier versions are kept and can be restored, and if you both write at the same moment your draft is kept, never overwritten.
How do I stop a mission that is running?
Close its mission ticket on the board: no further task starts. A task already running stays open until you close it, like any agent.
What is the My world page?
A page of the project's notebook, filled during the first interview: who you are, your activity, your clients, your tools, your vocabulary. Every assistant of the project is told to read it. It is private to your account and you can edit it at any time.
Does it work from my phone?
Yes, as long as your computer is on with AgentsRoom open: start a mission from the + button, answer the cards, open the page from the mission's ticket, and follow the tasks.
You may also like
Non-dev mode
Not a developer? The Non-dev mode of AgentsRoom gives you a room of AI assistants instead of projects and code: 17 business roles, running on your own AI account.
Structured questions
Your AI coding agent asks in a card with clickable options, right above its input box, and waits for your answer. Codex, Claude Code, Cursor, Copilot CLI.
Plan review
Your agent proposes a plan: read it as a document, comment on, replace or delete any passage, then approve it or send it back in one message. Every version kept.
Task management
Visual kanban board to orchestrate AI agents with drag-and-drop. Create tasks in TODO, drag them to Development to automatically spawn a Claude agent, and track completion through Pending and Done columns. Backlog-driven development for parallel AI coding.
Project Memory
Give your AI coding agents a long-term memory. Decisions, architecture and pitfalls persist across sessions and are shared with everyone on the project. Free, unlimited.
Voice dictation : dictate prompts, your agent codes
Voice dictation built into the composer : dictate your prompts to your AI agents instead of typing them. A microphone button turns speech into text right in the draft. Server-side transcription with nothing to configure, 21 languages plus auto-detect, on-device speech recognition on mobile. Dictate long, precise prompts in seconds, send more context, burn fewer tokens.
Start with what you want done
Download AgentsRoom, press New mission, and let your assistant ask the rest.
Companion app: monitor your agents on the go
Bring your own: Claude, Codex, Antigravity CLI, or other AI provider.
Push bugs and requests straight to your public backlog.