Goodbye Power Platform

June 26, 2025

For 16 months, I was dedicated to learning how to build software with Power Platform. Internal tools, data flows, automation - it all worked beautifully. I was planning to become a certified Solution Architect (PL-600).

The problem was that I don't work for a big company. How was I going to break into consulting or an in-house position? I needed a portfolio project, and so I came up with an idea that aligned with the work I had been doing previously at Seattle Convention Center. That led me to the QR Cards project, which took on a life of its own.

It took me about a year to realize that I couldn't complete the QR Cards system I envisioned using Power Platform. It wasn't a technical limitation, but rather a cost limitation with Power Pages, which charges even for anonymous users.

I tried a workaround, a Power Automate flow that served up webpages in a way that's probably not what Microsoft intended. It worked enough for a demo, but nothing you'd want to build a business on.

I needed another way.

In April 2025, I fed my Power Automate flows into GitHub Copilot and asked: "Do you understand these flows?"

Yes, it did. Copilot wrote comprehensive documentation for each of my flows, for which I had spent days mapping out the logic on paper.

"Now can you rebuild these flows in Python?"

GitHub Copilot didn't just rebuild my flows - it improved them with cleaner logic, new capabilities, and comprehensive error handling. Things I knew I needed but couldn't figure out how to implement in Power Platform. Things I didn't know I needed but was eager to learn.

My next move was slightly unusual: I fed AI my published newsletter posts. And then my business plans.

From there, AI understood what I was trying to achieve, by reading my documented thinking and building systems that matched my intent.

I hadn't expected this outcome. It wasn't planned. It emerged through experimentation. But it changed everything.

I discovered that AI builds better systems when you give it documented thinking instead of vague prompts. The documentation-first methodology now drives everything I build.

My switch last month to Augment Code also made a big difference. I had been using Claude via GitHub Copilot, but as soon as I tried Augment Code, I was hooked. Augment offers deep codebase context, architecture awareness, project memory, and the ability to keep on going and going from a single documentation-based prompt without breaking things. It's a few more pennies per prompt, but you get immense value out of each one.

An analogy: If the AI model (e.g. Claude) is the horse, the coding agent (e.g. GitHub Copilot or Augment Code) is the trainer. And if everyone has a fast horse, having the best trainer can be a competitive advantage. Especially if that trainer keeps your horse from wandering off into the weeds.

The Numbers I've been rebuilding my QR Cards project for just over 10 weeks.

On average, I've generated over 10,000 lines per day through a combination of documentation, tests, and code (in that order).

Wanna guess how many lines of documentation and how many tests I wrote using Power Platform?

But lines of code is a misleading metric.

The real measure is the number of problems I solve each day.

On average: 25.

That's the average count of documents that I produce per day.

Each document corresponds to at least one (and sometimes several) decisions that have to be made surrounding a given problem, from a simple go/no-go decision to an analysis of competing approaches to solution architecture. At each step, I influence the direction of the project while tracking every decision that goes into it.

Documentation categories include:

Vision & Planning: Big picture documents, requirements breakdowns, and architectural decisions Implementation Guidance: Specifications, API documentation, and configuration guides Testing & Validation: Test plans, test results, and bug reports Learning & Discovery: Research notes, experiment logs, and decision records Project Management: Task completion reports, methodology refinements, and progress tracking.

I'm grateful to have learned so much from Power Platform. Power Platform taught me what enterprise systems need. Test-Driven Development taught me how to build them well. And AI coding taught me how to build more than I ever thought possible.

But the real secret is something important that I've been good at all along -- how to read.

See you at the races.