A clear request. A reviewable result.
Turn a specific piece of work into inputs, steps and a result you can inspect.
Runs in your browser. No account needed.
One specific request
Prepare Maya’s next follow-up- Inputs
- Use the conversation notes. Find one helpful next step.
- Deliverable
- A draft and the questions still to confirm.
Make one request easy to review
An agent task is a specific piece of work assigned to an agent. It has a request, the context needed to answer it and a result a person can inspect.
Prepare Sam Taylor's customer-success interview is more useful than help with hiring. It identifies the person, the role and the deliverable. A good task also states which facts to use, what is missing and where the work should stop for review.
Put it to work
- Prepare an interview
- Use a candidate's role and interview stage to draft consistent questions for the hiring team.
- Triage an import issue
- Review Cedar Works' column-mapping problem and identify the example data needed to investigate.
- Review a vendor renewal
- Prepare a decision note for North Supply, including the owner and terms that still need checking.
Follow a request through its proposed steps
Explore agent tasks, then follow the example there:
- Choose a task and review its request, assigned role and fictional input records before approving creation.
- Edit the task context and checklist, then save your changes and mark the steps you have reviewed.
- Review and run the local simulation to inspect the scenario's proposed response beside your saved instructions.
- Reopen the task's creation chat to see the request and scripted preparation steps.
How it fits in your workspace
Common questions
- What should I include in a TulipFarm agent task?
- State the deliverable, give the relevant facts and say where the work should stop for review. For example: prepare interview questions for this role, use the supplied candidate notes and return a draft for review. Asking the agent to name missing information works better than asking for a conclusion the facts can't support.
- Are agent task results in the Explore tour real agent work?
- No. The Explore tour uses scripted, fictional requests and results so you can inspect the flow. Editing a task changes its saved definition but doesn't generate a new answer. Checking off steps doesn't send a response or do any work outside your browser.
Demo tasks are independent local examples. They are not assigned to your saved demo agents or executed in a real workspace.
Follow a request through its proposed steps
Start from an empty workspace and build something you can inspect and edit. Your changes stay in this browser.