Building got easy. Deciding didn’t.
AI tools have gotten good enough that writing the code is rarely the hard part of a side project anymore. Notes from three recent builds.
For most of my career, the distance between an idea and a working thing was measured in people and quarters. You wrote the vision, aligned the stakeholders, staffed the team, and then waited for the first demo.
That distance has collapsed. The AI tools have gotten really good, and I don’t mean autocomplete-good. I mean you describe what you want, they ask the right follow-up questions, and you end the evening with something that runs.
Three recent builds
This site. My personal site used to be my training business. I wanted the two separated: training on its own domain, and this one about me and what I build. That meant moving a site to new hosting, a new domain, DNS records, certificates, cleaning up an old deploy script, and then designing something new. I did it in a conversation. When the certificate failed to generate, I pasted the error and got back the three likely causes in order, with the exact commands to check each one.
TennisCut. I record my tennis sessions, and a 40-minute recording is mostly walking to pick up balls. TennisCut takes the raw video, cuts it down to just the rallies, builds a report with rally stats and where I move on court, and uploads the reel to YouTube with a chapter for every rally. An optional AI coach reads the stats and suggests drills, for about ten cents a session.
prfrmrr. I follow a handful of DJs, and their tour dates are scattered across sites that often disagree with each other. prfrmrr runs a research agent every day that collects every claim, keeps the source for each one, and flags conflicts for review instead of guessing.
None of these would have been worth a team and a quarter. All of them were worth an evening or a weekend.
What actually got easier
The hard part moved from how do I build this to what exactly do I want.
The code, the infrastructure, the glue between services: that’s now the fast part. The slow part is the part that was always the real work:
- Being precise about the problem. “Cut my tennis videos” is a wish. “Keep serves, groundstrokes and volleys; drop everything between points” is a spec.
- Making decisions. Should old URLs redirect or disappear? Should conflicting data be resolved automatically or reviewed? The tools will happily ask you, and you have to answer.
- Knowing when it’s done. Verification still matters. I check the build, click through the pages, and look at the result on my phone before anything goes live.
Why this matters beyond side projects
I spend my working life helping enterprises adopt cloud and AI. Building small things myself keeps that honest. It’s one thing to put “agentic workflows” on a slide; it’s another to watch an agent research tour dates every day, and to design for the times it gets them wrong.
If you’ve been sitting on an idea because it felt like too much work, it probably isn’t anymore. Write down what you actually want, as precisely as you can, and start.