After BetterCast, I was cooked. No team, no salary, a lot of restless energy. Then ChatGPT landed. I fell into it hard. I used it all day, every day, to the point I’d hit the usage limits and that became my natural break. My wife joked that every time I came downstairs it was because I’d maxed out the quota.
I’ve always had this picture of an internal agency: a room full of developers and designers I could point at an idea and get an MVP moving, quickly. The finances never lined up for that. ChatGPT did something close. I’m not a developer, I’ve worked as a self-taught product manager for years, I understand code, I can read it, I can reason about it—but I’ve never had the mental bandwidth to write production code. When I got my first usable code out of ChatGPT, that’s when it clicked. I could take its output, drop it into an IDE, iterate, and actually ship something.
The first bits were basic. I barely understood local environments beyond the concept. But with a few weeks of back and forth getting a server running on my machine, fixing errors, tightening prompts I realised I’d effectively found that “room full of developers” I always wanted. For twenty bucks a month.
The Project: A Website That Writes and Optimises Itself
I was tired of WordPress and I wanted to try something different: a site that would write and optimise itself. Not “AI as magic,” just a system that could research, draft, title, add images, and keep improving.
I’d used Django before, so that’s where I started. Python gave me room to bolt on little tools in the backend, not just “a blog.” I built small pieces first: autogenerate images and drop them into the post, write titles and meta descriptions automatically, wire up a content writer that would:
- take a topic,
- go do research,
- distill that research into usable sections,
- write the article with the structure I wanted.
Alongside that, I had a simple plan to buy older, niche domains and aim for hyper-specialised content with affiliate and ad links. The idea was: these sites would write their own content, optimise themselves, and rewrite when needed.
That was BongoCat. And, yes, it was a terrible idea. It didn’t go anywhere commercially. I ended up keeping it as my personal site for a while. Today BongoCat is dead just code sitting on my NAS. If you’re reading this, you’re on a small Hugo site I put together with AI’s help. BongoCat was the start.
What Was Hard (and What Surprised Me)
I jumped in at the deep end:
- I didn’t know what I was doing.
- I didn’t have a real plan.
- I didn’t understand how software is actually constructed end-to-end.
So I was trying to learn, all at once: Python, JavaScript, how files and modules fit together, how Django structures a project, how to set up Docker locally, how to deploy the thing. Frontend, back-end, environments, networking, production vs development. All of it.
The other half of the challenge was coaxing ChatGPT. Early models would lose the thread, invent functions, rewrite parts of a stack that didn’t need touching. Useful, but erratic. A lot of that has improved over the last couple of years, and the benefit of starting when it was rough is that I got pretty good at steering it into better outputs especially for code by being explicit, keeping context tight, and iterating in smaller steps.
Why I Tried the Affiliate Route
Simple: I needed revenue. I was out of work, living off savings. So I looked at Amazon affiliates, dropshipping, and other programs as a way to validate whether a self writing, self optimising site could pay its way.
The workflow was straightforward on paper: find the questions people ask in a niche, collect sources, distill them to solid answers, and push clean, structured posts with the right metadata. On the surface, that was about income. Deeper down, it was about learning: Python, JavaScript, Django, Docker, data flow, how to stitch systems together. That education stuck.
What I Actually Learned
-
Technical literacy that compounds.
I’m far more comfortable reading code, understanding how functions work, and seeing how pieces fit. I still lean very heavily on AI for code. I still couldn’t write a function end to end on my own full stop. But the direction of travel in the industry makes that less of a blocker. I understand Docker, environments, databases, and how to wire things together. That changed what I’m willing to attempt. -
Build vs buy has shifted.
For years I wanted to build software. Now it’s easier than ever to build just enough for my own needs. The market is saturated with subscriptions solving most generic problems. The interesting space for me is building small tools that solve my specific problems, not chasing a SaaS dream. -
The market is crowded and getting tighter.
We’ve had years of high developer salaries, a flood of new engineers, then layoffs, then AI accelerating output. Competition is heavy. Shipping yet another general purpose tool and expecting it to break out is harder. That reinforced the point above: I’m better off building for myself.
The Future I See
AI isn’t going anywhere. Local models, hosted models, hybrids the tooling is here and getting better. More of the work is going to look like natural-language specification: I describe what I want, and the system interprets, writes the code, scaffolds structure, generates tests, and plugs into CI.
Maybe not a single new “AI language,” but a layer that takes human language seriously and turns it into working software.
I don’t think this replaces big, established SaaS any time soon. I do think it lets more people step away from subscriptions for certain tasks and run small tools themselves. That’s where my head is now—very homelab: build the thing I need, host it, keep it simple.
Where BongoCat Landed
BongoCat didn’t become a business. It did reset how I work. I got the technical literacy I was missing. I proved to myself that I can direct AI to build useful software and ship it.
And I still believe the core of this: coding is natural language. The room full of developers I always wanted is now a prompt away.