Build Your First AI Workflow in 30 Minutes

A workflow sounds like something that needs software, arrows, and a subscription. It doesn’t. A workflow is a task you do repeatedly, handled the same way every time, with the thinking done once in advance. You can build your first one in half an hour with nothing but a chat window and a notes app.

The difference between using AI and having an AI workflow is reuse. Most people open ChatGPT or Claude, type something from scratch, get a decent result, and close the tab. Next week they do the same task and start from zero again. The prompt that worked is gone, the corrections they made are gone, and the quality resets to average.

Step 1: pick the right task (5 minutes)

Good first workflows share three traits. The task repeats at least weekly, the input changes but the format of the output doesn’t, and you currently find it boring. Weekly status updates, meeting notes into action items, turning bullet points into client emails, summarizing reports for a manager who reads nothing past the first paragraph.

Skip anything high-stakes for now. Not the board presentation, not the salary negotiation email. You want a task where a B+ result saves real time and a mistake costs you nothing but a re-read.

Step 2: write the instructions once, properly (15 minutes)

This is the part people rush. Open a document and write down what a great version of this output looks like, as if briefing a smart temp who starts today. Both Anthropic’s prompting guide and OpenAI’s best practices converge on the same advice: be explicit about role, format, audience, and constraints, and show an example of what good looks like.

A reusable template for a weekly status update might say: you write my weekly status update; audience is my manager, who skims; format is three sections titled Done, In Progress, Blocked; each item one line, starting with the outcome, not the activity; tone plain and direct, no adjectives; here is last week’s update as a style reference; here are my raw notes.

The raw notes are the only part that changes each week. Everything else is written once. (If prompting itself is the part that intimidates you, we covered the fundamentals in Prompt Engineering for Non-Developers.)

Step 3: run it, fix it, save it (10 minutes)

Paste the template plus this week’s notes into Claude or ChatGPT. The first output will be 80% right and 20% off in ways that annoy you. That 20% is the valuable part: every correction you make is an instruction you forgot to write down.

Fix the output by editing the template, not just the result. If it padded every line with “successfully,” add “no filler adverbs” to the instructions. If it buried the blocked items, say blocked items always come last and always name who you’re waiting on. Then save the corrected template wherever you’ll actually find it next week. A pinned note works. So does a doc titled “AI templates.” The tool matters less than the habit.

The workflow isn’t the prompt. It’s the prompt plus the discipline of never starting from scratch again.

What you’ve actually built

Run this for three weeks and the template stops needing edits. At that point a task that took 25 minutes takes 5: paste notes, paste template, skim, send. You’ve also built something more useful than the time savings, which is a worked example of you applying AI to a real job task. That’s the kind of specific story that holds up when someone asks what you actually do with these tools.

When the first one is boring and reliable, build a second. Most people who do this stop at four or five templates that quietly absorb a few hours a month of repetitive work. Nobody needs twenty.