AI and agents
Agentic workflow
An agentic workflow is a business process in which one or more steps are handled by an AI agent that decides what to do, while the rest of the process follows fixed rules.
How it works in practice
Most real processes are a mix. Receiving an invoice, checking it against a purchase order and posting it to the ledger are rule-based steps. Reading a supplier’s PDF in whatever layout they chose, or deciding which cost centre a vague line belongs to, needs judgement. An agentic workflow keeps the rule-based steps deterministic and gives the judgement steps to an agent.
This is usually cheaper and easier to trust than letting an agent run the whole process. The fixed steps are testable and predictable, the agent’s scope stays small, and the hand-offs are places to add checks and approvals. When people talk about agentic AI in business, this is what most working systems look like underneath.
A worked example
A property manager’s maintenance requests arrive by WhatsApp. A workflow logs each message and the unit number. An agent classifies the problem, checks the tenancy and suggests a contractor. The workflow then books the job, and urgent leaks go straight to the on-call manager.
Examples in this glossary describe typical builds for companies of 1 to 1,000 people. They are not named client results.















