John Saxon
About
Outside of my day job, I've been creating solutions for small businesses for over twenty years. It started with Webpark in 2003. A consulting and hosting company that grew from two eBay servers in a basement into a real operation running on dedicated hardware in a downtown datacenter. Fourteen years in, it was growing faster than I expected. With a young family at home, I chose day-job stability over the entrepreneurial leap. It was exciting to build something real, and I learned what it takes to run a business from the ground up.
I spent sixteen years at AESO, working my way from desktop support through Windows Server teams, NetApp storage administration, infrastructure design, and team leadership. Alberta's power grid is the kind of environment where mistakes aren't theoretical. That's where I learned how to build things that scale, stay secure, and don't fail.
Then came eight years where every role I held was the first of its kind: managing an IT team, then vendor management across hundreds of enterprise applications, then AI Operations once AI got real. During that stretch I started coding with AI the day ChatGPT 3.5 dropped in late 2022. It was rough. The models hallucinated, the code barely worked, every project felt like pushing a boulder uphill. But I kept going. Once the tools got good enough (and I got good enough with them), I built BuildBot with my daughter Katie: nine integrated applications for small businesses, then rebuilt again as an application factory that handles architecture, deployment, testing, and documentation end to end. What used to take teams weeks took under an hour.
That taught me the thing I now build around. Making software stopped being the hard part. Deciding what deserves to exist, and what should never be handed to a machine, became the whole job. Which is why my work now sits at the intersection of human experience and AI capability rather than on either side of it.
Coming into AI development without a traditional coding background taught me something different. I learned how to harness the AI itself. How to prompt it for good results, how to prep its context with the right information, how to auto-document and update its knowledge files along the way, how to give it skills. I think about it the way Anthropic describes it: a narrow bridge with cliffs needs exact guardrails and precise instructions, but an open field with no hazards just needs a general direction and trust. Knowing which one you're on is the whole game.
What I'm working through now is where the intelligence should actually live. AI is cheap right up until it isn't. One workflow costs pennies to run, which is exactly why it's so easy to say yes to, but multiply that across a company, running all day, every day, and the bill stops being a rounding error. Routing everything to a frontier model is the easy answer and often the wrong one, because a smaller model running locally can do a specific job just as well for a fraction of the cost, and it keeps working even when someone else's terms change. So the question I bring to every solution now is which parts genuinely need the frontier, which belong on hardware you own, and what each one actually costs once it is running at scale.
I used to be the architect of the solution. Now in 2026, I'm the input to the architect and the checker at the end. My 10-80-10 rule: humans do the first 10% (finding the information, prepping good input, shaping the plan), AI does 80% of the work, and humans check the last 10% and tweak the result. The hard part over time is pushing on those two 10% ends. That's what creates full automation.
I pay attention to a lot of people on the forefront. Eventually I start seeing glimpses of where I can engage in my own way. That's not genius. It's pattern recognition from paying attention.