In six months, you will not remember what this workflow does.

This is not a failure of your memory. This is a universal fact. You'll have built new workflows. You'll have modified old ones. Your work context will have shifted. And the decision you made about why a particular step filters for only Tuesday-through-Friday emails? It will be gone.

Documentation is not bureaucracy. It's not extra work. It's a gift you give your future self — the difference between "I can fix this in five minutes" and "I have to reverse-engineer my own workflow from scratch."

The good news: you don't need a lot of documentation. You need the right documentation. Specific, practical, and written in a way that will actually make sense to you later.

What to Document for Each Workflow

For every workflow you build, write down these seven things. You don't need essay-length answers — bullets and short sentences are fine.

1. Name and Purpose

One sentence. What does this workflow do, and why does it exist?

Example: "Morning Brief — Fetches the top stories from news sources, summarizes them with AI, and emails me a daily briefing at 7am."

Not: "This scenario fetches data from an API."

2. Trigger

What starts this workflow? How often? Any conditions?

Example: "Scheduled daily at 7:00am UTC. Runs every day, including weekends. No conditions."

Or: "Webhook from Shopify when a new order is placed. Runs immediately."

Or: "Checks every 30 minutes for new rows in the 'To Triage' sheet. Only processes sheets where Status != 'Done'."

3. Key Steps (in Plain English)

List the 3-5 most important actions, not the technical detail, but what the workflow does.

Example: - Fetches the top 10 stories from NewsAPI - Sends each headline to Claude for a one-sentence summary - Formats the summaries into HTML - Sends the HTML as an email to me@example.com

Not: "HTTP module to api.newsapi.org with headers and query params."

4. Data Sources and Destinations

Where does data come in? Where does it go?

Example: - Input: NewsAPI (external API), my Gmail address (hardcoded) - Output: Email to my inbox, and a copy saved to a 'Daily Briefings' folder in Drive

5. AI Steps

If your workflow uses AI, document it: - What model? (Claude 3.5 Sonnet, GPT-4o, etc.) - What prompt? (Copy the exact prompt, or a link to where it's stored) - What are you asking the AI to produce? (One-sentence summary, category label, suggested action) - Any special handling? (Temperature settings, max tokens, etc.)

Example: - Model: Claude 3.5 Sonnet - Prompt: "Summarize this news headline in one sentence. Be factual, not sensational." - Expected output: 1-3 sentences, plain text

6. Known Edge Cases

What unusual inputs have you seen, and how does the workflow handle them?

Example: - If an API call times out, the workflow waits 5 seconds and retries. If it fails twice, it sends me an alert. - If an email has no subject, it uses the sender name instead. - Workflows that process very large files (>50MB) have been seen to timeout — consider splitting into batches.

7. Last Updated and By Whom

Date + who made the change. Even if "who" is always you, write it down.

Example: "Last updated 2026-03-01 by [your name]. Added error handling for missing API keys."