AI agent
Build an agent that does one of your recurring jobs without being asked. You describe the job; it proposes what the agent should do, what it needs access to, and when it should run — five steps, each showing you a guess to correct rather than a form to fill.
- Asks for
- The job, described as you would describe it to a colleague
- What the result should look like, and who it is for
- Which of your connections it may read, if it needs one
- Gives back
- A proposed agent you correct rather than configure
- What it does, what it reads, and what it produces
- When it runs, asked last and only once the job is right
- An agent in your account you can task, follow and change
- Plan
- Pro and Max
- Use next
- Agent team
What it gives you
You said — "Every Monday I pull the sprint numbers together for my lead." Here is what I think you mean. Correct anything wrong: Does Reads the current sprint, writes a short status for one reader. Reads Jira, your connected project. Nothing else. Produces Six lines: committed, done, moved, blocked, and why. Runs Monday 08:00, your time. ← we will come back to this Goes to You, to send on. Not your lead directly, for now.
How it works
The reason most people never build an automation is the setup: a blank form asking for a trigger, a schedule, a prompt and a set of permissions, before you have decided what the thing should do. So this guesses first. You say "every Monday I pull the sprint numbers together for my lead" and it shows you the agent it thinks you mean.
Five steps, and each one is a correction rather than an entry: what the job is, what the agent produces, what it needs to read, when it runs, and who sees the result. A schedule is the last thing it asks for and only once the rest is right — an agent running every Monday on the wrong definition of the job is worse than no agent.
What it produces is a real agent you can then task directly, follow, and change. It is your agent, built in your account, doing one job you described in a sentence.
Who it is for
A PM with a recurring Monday job — a status roll-up, a ticket triage, a weekly read of feedback — who has never got round to automating it.
Where it stops
An agent does the job you described, so a vaguely described job produces a confidently wrong result every Monday. Start it producing for you rather than for your lead, read what it writes for a fortnight, then change who it goes to. It also cannot do what your connections do not let it read.
Questions people ask
What kind of jobs are worth giving to an agent?
Recurring ones with a clear input and a clear output: a weekly status roll-up, triaging new tickets into themes, reading the week’s feedback and flagging what is new. The test is whether you could describe the job to a new colleague in three sentences and expect them to do it the same way twice. Jobs that need a judgement call every time are the ones to keep doing yourself.
Why does it guess instead of asking me to configure it?
Because the configuration is the part that stops people. A blank form asking for a trigger, a schedule and a set of permissions demands decisions you have not made yet, in an order that makes no sense until you have seen the thing. Showing a guess means your first interaction is a correction — much faster, and it immediately reveals whether what you said describes the job you actually do.
Can I change the agent after it is built?
Yes. It is a real agent in your account: you can task it directly with a one-off, follow what it has done, change what it reads, change when it runs, or stop it. The sensible pattern is to let it produce for you for a couple of weeks before pointing the output at anybody else, because the first version of any agent is working from a description that turns out to be slightly wrong.
Tools people use next
Run it on your own work
AI agent needs an account, because it writes against what you give it and remembers what you have told it before. It is part of Pro and Max. Or see what the plans cost first.