Agentic workflow
An agentic workflow is a multi-step process in which an AI agent chains model calls, tool invocations, and decisions to complete a job — authenticate, create a customer, attach a payment method, then charge it — carrying state between steps and handling failures along the way. The workflow, rather than the single call, is the real unit of agent work against an API.
Real jobs are sequences. An agent booking a shipment or reconciling an invoice has to know which call comes first, which identifiers pass from one response into the next request, and what to do when step three fails after steps one and two succeeded. Reference documentation that describes each endpoint in isolation leaves that sequencing to inference, and inference is where agents guess wrong.
This is why workflow documentation — guides that chain endpoints into complete jobs, with prerequisites and the data flowing between calls made explicit — reads so differently to an agent than an endpoint list of identical shape. Error-recovery guidance belongs to the same layer: a workflow is only as reliable as the agent's ability to resume it after a failed step, an expired token, or a rate limit mid-sequence.
Discry's comprehension dimension includes a multi-step workflows check — whether the documentation chains individual calls into complete jobs an agent can follow — alongside an error-recovery guidance check for the failure paths those workflows depend on. Both results publish per profile with the rest of the comprehension breakdown.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free