AI Changed How I Build Software. Here's What That Actually Looks Like.
Every tech blog in 2026 has an AI take. Most of them fall into two camps: either AI is going to replace all developers by Thursday, or it's an overhyped bubble that can't write a for-loop. Both are wrong, and both are boring.
Here's what I can tell you from actually using AI tools every day to build and ship software: it's a genuine transformation, but not the one most people are talking about.
What changed
The biggest shift isn't that AI writes code for me. It's that the bottleneck moved.
For twenty-plus years, the hard constraint on what I could build was time — specifically, the time it takes to go from "I know exactly what this should do" to "it's built and working." That gap used to be days or weeks. Now it's often hours.
That sounds like a productivity story, and it is, but the second-order effect is more interesting. When building gets faster, you can try more things. You can prototype an idea on Monday, test it with real users on Tuesday, and decide by Wednesday whether it's worth finishing. The cost of experimentation collapsed.
For a solo developer, that's massive. I used to pick one idea and commit to it for months because the investment was too high to spread around. Now I can explore three directions in the time it used to take to explore one.
What didn't change
AI didn't replace taste. It didn't replace knowing what to build. It didn't replace the ability to look at a feature and say "this is unnecessary, cut it."
If anything, those skills got more important. When you can generate code faster, the quality of your decisions becomes the differentiator — not the speed of your typing. A developer with bad judgment and AI tools will just ship bad software faster.
I still spend most of my time thinking, not coding. Sketching out how something should work. Deciding what to leave out. Testing whether the thing I built actually solves the problem I thought it would. AI accelerated the middle part — the construction — but the beginning and the end are still entirely human.
The tools I actually use
I'm not going to write a "top 10 AI tools" listicle. But broadly, here's where AI fits into my workflow:
Writing code. AI pair programming is real and useful. Not for generating entire applications from a prompt — that still produces garbage for anything non-trivial — but for accelerating the parts where you know what you want and just need to get there faster. Boilerplate, refactoring, translating logic from one language to another, writing tests for code you've already written. It's like having a very fast, somewhat careless junior developer sitting next to you.
Writing words. Drafts, documentation, support replies. AI gets you 60-70% of the way there and you edit the rest. The key is editing aggressively — if you publish AI output without rewriting it, people can tell. They can always tell.
Research and exploration. This is maybe the most underrated use. When I'm working in an unfamiliar framework or exploring an API I haven't used before, AI is genuinely better than documentation for getting oriented quickly. Not for the details — you still read the docs — but for the "give me the big picture in two minutes" phase.
What I worry about
Two things.
First, quality erosion. It's tempting to accept "good enough" output because the alternative is spending another hour on something. The entire internet is already filling up with AI-generated content that's technically correct and completely soulless. I don't want my apps or this blog to feel like that.
Second, dependency. I've been writing code for long enough that I can still build things without AI tools. But developers who started in the last two years might not have that foundation. When the tool goes down or gives you confidently wrong advice, you need to know enough to catch it. The skill underneath the tool still matters.
The honest version
AI made me faster. It made running a one-person studio more viable than it's ever been. It did not make the hard parts easier — the hard parts were never about typing speed.
If you're an indie developer or a solo founder wondering whether AI tools are worth integrating into your workflow: yes, obviously. But treat them like power tools, not autopilot. You still need to know where you're going.