Delegate clearly. Stay in control.
Help with Sunlight Workers, AI agents that take on work like members of your team. Learn how to brief them, set boundaries, and review what they deliver.
What do you need help with?
Delegating work
Give your AI worker a clear role, task, and definition of done.
Read guidance ↓Access and oversight
Keep permissions aligned with the task and review consequential actions.
Read guidance ↓Reviewing results and resolving issues
Check the work, correct misunderstandings, and prepare a useful handoff.
Read guidance ↓Delegating work
What should I include in a worker’s task?
Describe the outcome, relevant background, available resources, and the format you want back. Include a deadline if relevant, the decisions the worker can make, and anything that requires your review.
How should I define a worker’s role?
Explain the responsibility you are delegating, such as preparing a research brief or organizing a project plan. Set boundaries for the role and identify who reviews the output. An AI worker can assist with work, but a role description does not give it human judgment or organizational authority.
How do I divide work between several agents?
Give each worker a distinct deliverable. Explain shared context, dependencies, and who is responsible for combining the results. Avoid assigning conflicting instructions or overlapping ownership of the same task.
Access and oversight
What access should I give a worker?
Limit access to the information and services needed for the task. Use the permission controls available in your service and review them when the task changes. Never paste passwords or secret keys into a task description.
Which actions should require my approval?
Define review points before actions such as sending messages, publishing content, spending money, changing access, or deleting information. Verify which approval controls your service actually provides; instructions alone are not a substitute for enforced permissions.
Can a worker act as an employee without supervision?
Treat Sunlight Workers as AI agents assisting with delegated responsibilities. You remain responsible for deciding what they may do and checking important results. The level of autonomy depends on the capabilities and controls available in the service.
Reviewing results and resolving issues
How do I know whether a task is finished?
Compare the output with the deliverable and completion criteria you set. Check supporting sources, identify unresolved questions, and confirm any claimed external action using the relevant service rather than relying only on the agent’s summary.
A worker misunderstood the task. What should I do?
Identify the incorrect assumption and give a concrete correction. Restate the intended result and ask for a revised deliverable. If work depends on the incorrect result, review that downstream work too.
What should I check when work stops progressing?
Look for missing information, unavailable access, an unanswered approval request, or an error from a connected service. Record the point where progress stopped. Avoid starting duplicate work until you understand whether an external action already completed.
Where can I see supported integrations and limits?
Check the information in the Sunlight Workers service and the terms for your account. This support page does not publish a verified integration list, scheduling capability, concurrency limit, or guarantee of unattended operation.
A little context
helps.
Include the worker’s assigned role, the task and expected deliverable, the permissions involved, and where the result went off track. Note whether any external action already occurred.
Remove passwords, API keys, personal information, and confidential material from screenshots or examples.
View contact information