Create and review tasks

A task keeps a request, its team, outputs, and review history together. Use chat to develop a request or create it directly from the board.

Create a task in chat

Select a product and message Jak or a specialist with an actionable request. The app creates a task and links it in the conversation. Ordinary questions stay in chat until you choose Make task on a response. Repeating Make task on that response returns the same task.

If work is Ready for review, a follow-up in the same thread can request a revision. That revision uses the task’s saved brief and plan. Use the task’s retry or review controls when its state does not allow a chat revision.

Create a task from the board

1. Open Tasks → New task.
2. Select the Product and enter a Title and Brief.
3. Choose a Lead agent. Use Jak when you want coordinated work.
4. Select contributors if you have a particular team in mind.
5. Leave Model for this task on Use each agent’s model setting, or choose a saved model for everyone on this task.
6. Optionally adjust Thinking under the advanced options, then choose Create task.

A task-wide model selection also applies to retries. Otherwise agents use their own model override or the installation default. Automatic thinking is a good starting point; more thorough settings can take longer and use more model tokens.

Write a useful brief

Include the deliverable, intended audience, source material, constraints, and definition of done. For research, name a time range and ask for source links. For private account data, connect the tool to the same product first.

Review the last 28 days of Search Console performance for example.com. Find five pages losing clicks, compare with the previous 28 days, and suggest one change per page. Include the property, dates, clicks, impressions, and source URLs. Deliver a table and a short prioritized action list.

The team follows an ordered plan. Contributors receive relevant prior outputs as the work progresses; selecting several contributors does not mean they all run simultaneously.

Follow the task board

Filter Tasks by product, agent, or status. A request has one card even when several agents contribute.

Status	What to do
Queued	The request is saved and waiting to start.
In progress	Agents are working through the plan.
Ready for review	Read the output, then accept or request changes.
Needs attention	Read the message and provide the requested answer.
Failed	Fix the reported issue, then choose Retry task.
Done	The work has been accepted.
Canceled	Work on this task has been stopped.

Open a card to read its outputs. Team & research shows contributions and research notes. Details includes the original request, model setting, rerun schedule, and task history.

Review and revise

Choose Request changes, explain what to improve, and submit the feedback. Refer to the exact section and desired result. The next revision remains on the same task, and earlier versions are available in Details → Task history → Saved output versions.

You can also edit an output when its edit control is available. Choose Accept work once the deliverable is ready. Acceptance marks the task Done; it does not publish or send anything.

Some models and task states offer Re-run with deeper thinking. Use it when the task needs another reasoning pass. Check the new output as you would any other revision.

Recover a failed task

Read the error first. Common causes include an unavailable model, expired tool access, a rate limit, or an app restart during generation. Fix the underlying problem and choose Retry task. Retries generate fresh output; previous attempts stay in history.

For an uncertain external action, check the destination service before doing anything else. I have verified the provider result records your manual check; it does not query the provider or resend the action.

Repeat the work

Open Details → Schedule reruns to turn a useful task into a daily or weekly routine. See Recurring tasks for timing, editing, and downtime behavior.