An AI workflow is a repeatable process for getting a specific kind of work done with AI: the input arrives in a known shape, saved instructions do the explaining, AI produces a draft, and you review the draft before the work is finished. Rough call notes turned into a follow-up email work this way. So do receipt photos turned into a monthly expense summary, or a job posting plus your resume turned into a tailored cover letter draft.
The four parts
A workflow has four parts. The input is the raw material: your call notes, your receipts, the job posting. The instructions are the saved explanation: who the output is for, the format, the tone, what to leave out. The draft is what AI produces from the two. The review is you, checking the draft against the source and finishing the job.
Remove a part and the shape changes. No saved instructions means you re-explain the task on every run, which is a chat, not a workflow. No review step means a first draft goes out unchecked, and first drafts can be wrong in ways that are easy to miss: a plausible number, an out-of-date fact, a promise you never made.
Workflow, chat, or agent?
A chat is one request and one answer: you explain what you want, AI responds, the conversation ends. A workflow is the structure around repeated requests: the saved instructions, the standard input, the same review each time. An agent is what appears when software runs the steps instead of you, making decisions along the way, like a scheduled task that collects the sources and drafts the digest on its own.
The order matters more than the labels. A workflow you run yourself shows you where the instructions miss, because you're there to see the draft fall short. Once the steps have stayed stable for a while, handing them to software on a schedule or to an agent might make sense. Starting there, before the steps are written down, automates a process that hasn't been defined yet.
The hand-off test
The hand-off test is simple: write down how the task runs and give the page to someone else. If they can produce something close to your output, the workflow holds enough of your method for AI to run its part. If the instructions never say whether the client or the internal team gets the short version, that missing detail is an instruction you haven't saved, and the test found it before a bad draft did.
The test also tells you when a workflow is ready for routine use. Once AI can follow the setup and produce a draft worth checking, keep it running, and add instructions whenever a run misses something.
Build your first one
Pick a task that comes up about weekly, has a stable shape, and produces output you can check. Follow-up emails from call notes are a good first candidate, and so are expense summaries and tailored cover letters.
Do the task once with AI and save everything you had to explain. For the follow-up email, that might be one line:
Follow-up email: from the call notes I paste, draft a short recap email. Agreed action items with owners and dates first, then a two-line summary, warm sign-off. No new ideas, nothing I didn't say on the call.
Decide the review step before you need it: facts against the notes, action items against what was agreed, anything headed to the client read in full before it goes. Our verification guide covers how much checking different stakes call for.
Run the workflow twice, fix what the drafts miss, and you have a workflow worth repeating.
FAQ
What is an AI workflow?
A repeatable process for one kind of work: a defined input, saved instructions, an AI-produced draft, and a human review step before the work counts as finished.
What are some examples of AI workflows?
A list of numbers into a weekly status update, meeting notes into a recap for the team, a voice memo into a to-do list. The pattern is repetition: same input shape, same output shape, run again the next time the task comes up.
Do I need to code to build an AI workflow?
No. The workflows in this post run in a chat window with a notes app holding the instructions. Scheduling and app-to-app automation can come later, and the setup is configuration rather than code: our automation guide covers what the free ChatGPT plan can run on a schedule.
What's the difference between an AI workflow and an AI agent?
A workflow is the process; an agent is software that runs steps of a process on its own and makes decisions along the way. A workflow you run yourself is the starting point, and an agent is one way to hand it off once the steps are stable. Our guide to AI agents covers what changes when software runs the steps.
Want to go further?
- How to Build a Personal AI Workflow (Step by Step) walks through the full build for one task: the setup document, the draft, and the review.
- How to Automate Repetitive Tasks with ChatGPT covers what to do once a workflow is stable: save it, schedule it, keep the review step.