KelsoCodes.com

A sales guy who probably shouldn't be building software.

25 years in enterprise sales. No CS degree. No dev team. Just real problems worth solving and enough curiosity to be dangerous. Building AI-powered applications on my own time, with my own money, sharing everything I learn along the way.

Nobody taught me how to build software. In fact I still don't know how to build software.

Maybe one day. Maybe not.

I'm a 25-year sales professional who would honestly rather be standing in a river chasing steelhead. But the fish aren't biting and the AI is.

Over the last six months I've built four AI-powered applications, on my own time, with my own money, learning everything from scratch. My accountant is not impressed. But some of my peers are. That's good enough for me. For now.

If you're curious about how to do the same, here's what someone told me:

01. Start with the spec, not the code Draw your data model before you write a single line of code. Map your entity relationships. Define how your data moves between services before those services exist. The functional design document isn't paperwork. It's the moment you find out whether you actually understand the problem. Skip it and you'll rebuild the same thing twice. I know this because I rebuilt the same thing twice. Or more.
02. Treat security like it's day one Because it is. Identity management, MFA, role-based access control. These aren't features you add later. They're decisions that shape everything underneath them. By the time you're thinking about security at the end, your architecture is already working against you. And so is your sleep schedule.
03. Your framework is a long-term relationship Whatever you choose, understand what you're committing to. Flexibility early has a cost as complexity grows. Changing it later isn't refactoring. It's starting over. With coffee. Lots of coffee.
04. Stop thinking about single models Start thinking about systems. I use Perplexity for real-time research, Claude for reasoning and strategy, and Gemini when I need a different perspective on a hard problem. A visual workflow automation tool ties it all together. Google NotebookLM turns documents into audio intelligence. The real power isn't any one tool. It's knowing how to connect them.
05. Data design is the job nobody talks about How your services pass information to each other, what goes in, what comes back, what gets stored, determines whether your system is reliable or fragile. Get the contract right. Your future self will either thank you or have some very strong feelings about your past self. I'm prepared for harsh words from my future self.

And the one thing no tool gives you: knowing the problem deeply enough to build the right solution. That still comes from you.

I'll keep sharing what I learn as I build. If this is useful, follow along.

Thanks to Ryan Clark, Stephen Thorne and Kevin West. And to Rob Wyse and Shamis Thomson.

#AI #BuildingWithAI #LearnWithMe #ClaudeCode #Python #Auth0 #AILeadership #BuildingInPublic

Every two weeks a new entry drops. Honest. Humorous. Useful. The wins, the rebuilds, the 11 PM debugging sessions, and whatever the AI taught me that I probably should have already known. No polish. No filter. Just the actual journey.

Get new entries straight to your inbox. No spam. Just a sales guy learning to build in public.

You're in. Next entry lands in your inbox.

Something went wrong. Please try again.