When a team says “we want an AI agent,” what they usually need is something smaller, cheaper and easier to govern. The right shape depends on three questions: Does a person trigger the work? Are the steps predetermined? How much freedom does the capability need?
A Skill is user-invoked
A Skill is a capability your team starts whenever they need help with a recurring task. It combines instructions, approved source material, examples and tests — and, when it helps, a connection to another system. Preparing a recurring document, reviewing information against approved criteria, or producing a structured first draft are typical Skills.
A Workflow runs on a schedule or an event
A Workflow is an operating process with predetermined steps, business rules, approvals and exception paths. It can coordinate actions across systems and take care of routine steps on its own, so work gets done faster and more consistently. Because it runs on its own, Managed AI keeps production Workflows monitored and supported.
An Agent chooses its next step
An Agent receives a defined goal and dynamically selects among approved tools and actions to reach it, so it can take on work that calls for judgment. A governed Agent operates inside explicit permissions, and we always begin with a controlled pilot that has limited production exposure.
Why the distinction matters
Each shape carries different implementation effort, different ongoing support, and different risk. Choosing the smallest shape that produces the outcome is usually the right decision — and it is one of the things an AI Assessment settles before any money is spent on a build.