Lesson 03 · Part 5 10 min

Proving the Workflow

Prove that your workflow is reusable by running a completely different topic through the same four building blocks you just assembled.

Watch on YouTube

In this series

Where we are

You've built a complete workflow and iterated a first draft from roughly 75% to production-ready. The context.md now includes everything: brand system, persona, tool map, process sequence, and the verification and feedback rules you encoded along the way. The real question is whether all of that investment was worth it. Time to prove it.

You spent the majority of your time planning. Then the first document went from a rough draft to something you'd actually send in a few passes. But the real test isn't whether the workflow works once. It's whether the same context.md can produce a brand-consistent first draft on a completely different topic, with zero additional setup. That's what separates a one-off project from a reusable workflow.

The reusability proof

A good test to run: ask Claude Code to create an input document on a random topic, then run it through the same context.md. I asked Claude to write a dummy two-page input on the importance of weightlifting and its contribution to long-term health and longevity. Then ran one prompt: read the input, apply context.md instructions, create a PDF document.

Input Random topic

A two-page document about weightlifting and longevity, written by Claude. Zero overlap with the original task.

Setup One prompt

Read the input, apply context.md instructions, create a PDF. No additional configuration. No new rules.

Results

Same logo, same brand colours
Proper structure and layout
Verification passes ran automatically

The whole thing took about seven minutes. The verification passes ran automatically because they were baked into context.md. This is feedback encoding in action: the rules you encoded in Part 4 are now permanent.

Key Insight

This took seven minutes to make. That's the crazy part. Same quality, different topic, zero additional setup. The investment in planning pays for itself by the second run.

The compounding time curve

The investment is front-loaded; the returns compound over every future run.

First document 60-90 minutes

Task selection, context library, planning, execution, iteration, feedback encoding.

then
Every document after ~7 minutes

One prompt, zero additional setup, same quality output.

This is the same pattern whether you're building branded PDFs, weekly reports, client proposals, or any other repetitive, research-heavy task. The setup time pays for itself by the second run.

The four building blocks recap

Throughout this demo, you assembled and used all four building blocks from Lesson 1. Here's how each one showed up in practice.

LLM & Environment Claude Code inside VS Code. Folder integration, read and write. Context window management with /clear and /compact. Model selection: Opus for planning, Sonnet and Haiku for execution
Context & Knowledge Base The folder system with input, research, brand guidelines, templates, inspiration, and output folders. Different workflows will have different context libraries, but the structure is the same
Tools Internal Python tools: Playwright and Chromium for screenshots, Plotly and Kaleido for charts, Pillow for PDF assembly. External tools like Slack and GitHub mapped via MCP
Instructions context.md as a living, working document. Persona, process sequence, brand rules, verification steps. Once set up, it produced similar quality output in a fraction of the time
Key Insight

When thinking about how to apply this to your own work, think of tasks that are repetitive, research-heavy, instruction-driven, and tool-dependent. The more of these traits a task has, the stronger the case for building a workflow around it.

Your first AI agent workflow

You started with the task. You defined what you wanted to build and what "done" looks like. Then you spent the majority of your time planning. You asked the questions around the persona it should adopt, the process it should go through, the tools it could use. That gave you some inspiration that you actually ended up using.

Then you created the first iteration and iterated either surgically or in bulk. Every correction was encoded back into context.md as a permanent rule. And you proved the whole system works on a completely different topic with zero additional setup.

This is your first AI agent workflow. The four building blocks, the planning-heavy approach, the feedback encoding loop, all of it applies to any task that meets the criteria from Part 1. The format changes, but the method stays the same.

Download PDF

Enter your details to download

Ask me a question

About this lesson or anything AI-related

Ask me on LinkedIn Send me a message directly