What should an AI agent take on first?
The first pilot should show whether an agent makes one piece of work easier. Choose a task you can measure before deciding what else to hand over.
- Pick one taskRepeatable work with clear inputs
- Set the review stepA person checks the draft
- Decide whether to expandCompare time and corrections
Choose a task with a useful first draft
Start with something your team does regularly, with inputs it can name and a clear standard for a good result. In an RFP, for example, an agent could draft answers to recurring questions using approved material. A reviewer could check a draft against the same evidence they already rely on.
Decide where a person reviews the work
Assign a reviewer before the pilot starts. For the RFP example, the agent links each claim to an approved source and flags questions it cannot answer. The reviewer checks the claims, fills the gaps, and decides what goes to the buyer.
Compare the whole workflow before expanding
Use a small set of comparable requests and agree on the threshold up front. Track time from intake to a review-ready draft, time spent reviewing, and how often a claim needs correction. If the team saves time after review and the quality holds, try a related question set. If review absorbs the savings, revise the inputs or choose a different task.