Agiled Docs

Update worker behavior, context access, tool categories, status, and iteration limits.

Edit a worker when its behavior, permissions, or readiness changes.

Pause a worker before making risky changes if it can affect records, messages, tasks, finance data, or customer-facing output.

Before Editing

Review the last few runs before changing instructions. Confirm whether the problem is the worker prompt, missing context, tool access, source data, or user expectations.

If the worker writes comments, creates records, sends messages, or updates tasks, pause it before broad edits. This prevents new runs from using half-reviewed instructions.

Copy the current instructions before making a risky edit. If output quality drops, you can restore the previous wording instead of trying to recreate it from memory.

Edit a Worker

Open AI Workers.
Open the worker.
Select Edit.
Update name, description, status, instructions, context sources, tool categories, output mode, or max iterations.
Select Save changes.

When to Edit

Edit a worker when:

Its output is too vague.
It needs less or more context.
Tool access is too broad.
It should be paused.
It needs a clearer role.
A failed run shows a permission or scope problem.

Make instructions specific enough that another teammate could predict the worker's behavior. Include the target outcome, data it should use, tone or format requirements, and anything it must not change.

Avoid broad instructions such as "handle everything for this customer." Split large responsibilities into smaller workers when the task needs different context, tools, or review rules.

Context and Tools

Give the worker only the context sources and tool categories needed for the job. More access can improve output, but it also increases the review surface.

When adding a new context source, run the worker on a safe record and check whether the output cites or uses the new information correctly.

Remove context or tools when the worker starts using irrelevant information. Adding more instructions is not always the right fix if the worker can see too much.

Status Changes

Use Paused when a worker should stop being used while you review it. Use Draft for setup. Use Active only when instructions and access are ready.

After Editing

Run the worker on one safe record or narrow context. Compare the new output with the previous behavior and review run detail before returning the worker to broad usage.

If you reduced context or tool access, confirm the worker still has enough information to complete the job. If you expanded access, review the safety page before activating it.

Edit Review Checklist

Before reactivating a changed worker:

compare the latest run with a previous successful run
confirm the source record was safe for testing
confirm output format still matches the instructions
confirm no extra tools or context were added unnecessarily
confirm queued or active runs are not using old assumptions
document the reason for the change if the worker affects shared work

Troubleshooting

If saving fails, check required fields and shorten overly long instructions.

If the worker still behaves the old way after saving, refresh the worker detail page and start a new run. Existing or already queued runs may have used the previous configuration.

If output quality drops after an edit, compare the latest run with the previous successful run and roll the instruction back to the smaller change that caused the regression.

Edit an AI Worker