Lesson 002Field guide · AI Workflows & Systems
Input → Process → Output
The simple model behind every reliable AI workflow.
You'll learn
- The three-part model behind every reliable workflow
- Why inconsistent results usually trace back to inconsistent input
- Why the process is the prompt plus the review — never the prompt alone
- How to define good output before you run, not after
- How to map any recurring task through Input → Process → Output
01Section
Same prompt, different day, different result
You run your best prompt on Tuesday and get a great result. Friday, same prompt, and the output is mediocre. The prompt usually gets the blame — but the prompt didn't change. Something else did, and more often than not it's what you fed in: different notes, a different shape, a chunk of context missing.
Reliable workflows aren't built on magic prompts. They're built on controlling all three parts of a run — and there's a simple model for that.
The model in one line
Input → Process → Output. Control what goes in, fix what happens to it, define what should come out.
Checkpoint
02Section
The three parts
Input: Everything you hand the AI: the notes, the draft, the data — and, just as important, their shape and quality.
Process: What happens to the input: your saved prompt, and your review of what comes back. Both are part of the process.
Output: What comes out — in a shape you defined before running, so you can tell good from merely fluent.
Key takeaway
The process is the prompt plus the review. A workflow without a review step isn't finished — it's just unchecked output.
Checkpoint
03Section
Standardize the input
Here's the rule that fixes most flaky workflows: same shape in, same quality out. When results swing between runs, look at the input before you touch the prompt.
Friday's input
Meeting notes pasted as one wall of text — no date, no attendees, and halfway down it accidentally includes notes from a different meeting. The prompt never stood a chance.
The standardized version
The same notes pasted under a small fixed header — Meeting, Date, Attendees, then the raw notes. Thirty seconds of formatting, and the prompt now sees the same shape every single week.
- Decide what the input always includes — and what it never includes
- Give it a fixed order; a mini template beats your memory
- Trim what the prompt doesn't need — extra noise degrades output
- Note where the input comes from, so gathering it is step one, not a scramble
Privacy note: Standardizing input is also where privacy lives. Decide once what never goes in — customer details, credentials, other people's private information — and that decision is made for every future run.
Checkpoint
04Section
Define good output before you run
If you can't describe a good output before running the workflow, you can't recognize one after — you'll accept whatever sounds fluent. So write the definition first. It takes two minutes and turns "looks fine, I guess" into an actual check.
Output spec — fill in once per workflow
A good output for this workflow: - Format: (e.g. a table, a bulleted action list, a 150-word draft) - Must include: (e.g. an owner for every action item) - Must not include: (e.g. invented details, filler, anything not in the input) - Done when: (the check you run before using the result)
Checkpoint
05Section
Two workflows through the model
Here's the model applied to two everyday workflows. Notice how each part is one plain sentence — if you need a paragraph, the workflow isn't clear yet.
Workflow 1 — Meeting notes → action list
Input: raw notes pasted under a fixed header (meeting, date, attendees, notes). Process: the saved meeting-notes prompt, then a scan for wrong owners or invented deadlines. Output: an action list — task, owner, due date — pasted into the team channel.
Workflow 2 — Weekly priorities
Input: last week's list, this week's calendar, and open tasks — pasted in that order. Process: the weekly-priorities prompt, then a gut-check against what actually matters this week. Output: a top-five list with one line of reasoning each.
Pause and think: Pick one task you repeat. Can you say, in one line each, what its input, process, and output are? Whichever line comes out fuzzy is the part to fix first.
Checkpoint
Prompt exercise
Map one of your tasks through the model
Copy this prompt into ChatGPT, Claude, Gemini, Copilot, or whichever AI tool you have access to — the website doesn't run AI itself. Use the task you identified in the last lesson, or any task you repeat.
Help me map one recurring task into an Input → Process → Output workflow. Ask me one question at a time: what the task is, what information I start with, what I do with that information today, and what a genuinely good result looks like. Then lay the workflow out in three short sections — Input (what I gather, in what shape), Process (the prompt I should run, and what I should review afterward), Output (the format, plus the checks that tell me it's good). End by naming the single weakest link in my current version of this task.
Reflection: Look at the weakest link it named. In most first workflows it's the input — same task, different shape every time. Fixing that costs less than any prompt rewrite.
Quick check
3 quick questions — no pressure
There's no pass or fail here. Answer them all, and we'll show you the answers either way.