Agentic workflow

A fixed script is predictable and cheap to run, but it breaks when reality differs from what it expected. An agentic workflow copes with messy or varied input because it can change course. The price is cost and predictability. It uses more computing per run, can take a different path each time and needs limits so it does not wander. The skill is knowing which tasks justify that price.

• Use agentic workflows for judgement-heavy work: research, inbox triage, compiling a report from several sources.
• Keep ordinary workflow automation for steps that never change.
• Give the agent a clear goal and the tools to reach it, not a rigid list of steps.
• Cap how many times it can loop or retry, so a confused agent stops instead of spinning.
• Require it to report what it did, and send unclear cases to a person.
• Review a sample of runs regularly. A workflow that drifts usually drifts quietly at first.

A scripted automation follows the same steps every time: when a form is submitted, copy the fields to a spreadsheet. An agentic workflow gives the agent a goal and a set of tools, then lets it choose the steps. It plans, acts, looks at the result and adjusts, looping until the job is done or a limit is reached. Triaging an inbox is a good example, because each message may need a reply, a forward or nothing at all.

• Using an agent for a job with one correct path, where a script is cheaper and safer.
• Giving broad tool access when the task needs only reading.