Give Plexii a goal with the Autonomous agent

In a normal conversation, Plexii proposes changes one reply at a time. The Autonomous agent takes a whole goal instead, such as "set up a lead tracker with columns and a research agent wired to it", and works through it in rounds: it plans a step, carries out what it is allowed to, looks at what actually happened, and plans the next step, until the goal is done or it needs you.

Use it for set-up work with several parts, where you would otherwise go back and forth with Plexii many times.

Before you start

• Pro and Team plans include the Autonomous agent.
• It needs AI to run. See AI credits and where your AI runs.
• Open the desk you want it to work on. The agent works on the desk that is open when you click Run.
• Check your autonomy setting. It decides which changes the agent may make without asking you. See Choose how much the agent may do on its own.

Open the Autonomous agent

1. Click the Plexii button at the bottom right of the screen.
2. Click the Agents tab under the panel's header.
3. Make sure Autonomous agent is selected at the top of the tab. The other view, Desk agents, lists the agents on your desks.

Run a goal

1. Type your goal in the box. It suggests e.g. Set up a lead tracker with columns and a research agent wired to it.
2. Click Run, or press ⌘ Enter on a Mac or Ctrl Enter on Windows.
3. The button changes to Working… while the agent works. Your goal shows above the rounds as it goes.

Only one run can go at a time. While it works, the goal box is locked.

When the run has finished, click New goal to clear it and start again.

What you will see

Each round

Every round appears as a numbered card, #1, #2 and so on, with a sentence saying what the agent is doing. Under it:

• a tick for each change that was made, or a cross for one that failed, with a short message
• a line, Deferred for your approval:, naming anything it held back for you

The agent sees exactly which changes worked and which failed, so it can correct course in the next round.

Checking its own work

When the agent believes the goal is done, it checks what it actually changed against your goal. A round says, for example, Verified the goal is met (100%). If something is missing, it lists the gaps and takes up to two more rounds to fill them.

How the run ends

The final line shows one of these:

A run has at most eight rounds. If the goal is still not confirmed by then, it stops with Stopped after 8 rounds without confirming the goal was complete. Here is where it got to.

Waiting for your approval

Anything the agent held back appears at the end under Waiting for your approval, as the same cards Plexii uses in a conversation. Review each one and apply the ones you want. See Review and apply Plexii's changes.

Choose how much the agent may do on its own

The agent follows Assistant autonomy in Settings > AI:

Whatever the setting, these always wait for your approval:

• deleting a widget
• putting time in your calendar
• acting on email, such as moving or flagging a message
• drafting an email or a chat message
• starting a focus timer

Your organisation can set a limit that your own choice cannot go above. The setting is read when a run starts, so changing it mid-run applies from the next run. See Assistant autonomy.

Undo a run

Everything the agent applied in one run is a single step you can undo, labelled with the start of your goal. Undo it as you would undo any change Plexii made. See Review and apply Plexii's changes.

Tips

• Describe the end result. "A desk for the Q3 launch with a task list, a budget table and a running summary of the meeting notes" gives the agent a clear finish line.
• Open the right desk first. The agent builds on the desk you have open.
• Use it on your inbox. If you have connected mail, the agent can see your recent unread messages, so "draft replies to the emails that need one" works. Each reply arrives as a draft for you to review and send. See Mail in Plexii.
• Try Act on low-risk for set-up work. The agent can then build tables and widgets without stopping to ask, while still asking before anything consequential.

Common questions

The agent finished but nothing changed on my desk. On Ask before acting or Suggest only, the agent changes nothing itself. Look under Waiting for your approval and apply the cards you want, or change Assistant autonomy in Settings > AI.

Can it send email for me? No. It can draft a reply, which opens for you to review and send yourself.

It says Needs you. Read the line beside the status, then click New goal and run again with the missing detail in your goal.

Is this the same as a desk agent? No. The Autonomous agent works once on a goal you give it. A desk agent is a widget that stays on a desk and repeats a job. See Desk agent widget.

Related tasks

• Open, move and resize the Plexii panel
• Review and apply Plexii's changes
• Assistant autonomy
• See and manage all your agents
• Everything Plexii can create or change