I was wrong about AI and development. Here's what changed my mind.

A year ago, if you'd told me AI would touch a meaningful chunk of my actual paid work, I'd have pushed back. I thought of it as a toy for writing boilerplate — useful, but nowhere near the real craft of building something pixel-accurate, something that actually holds up in production.

I don't think that anymore.

I ship landing pages in 1-2 days now that used to take a week. Not because I cut corners — because AI removes the boring 60% (debugging syntax, Googling the same CSS trick for the hundredth time, writing the same n8n node logic from scratch) so I can spend more time on the 40% that actually makes a project good: animation feel, structure that won't break in six months, thinking about what the client actually needs.

Here's the part that surprised me most: the developers who are getting replaced aren't losing to AI directly. They're losing to other developers who figured out how to use AI as leverage, faster than they did.

I used to think "using AI" meant admitting I couldn't do it myself. Now I think refusing to use it is the actual risk.

Still think craft matters more than ever — AI can't tell you what "feels right" for a client's brand, and it definitely can't build trust with a design studio on a call. But the tools changed. Pretending otherwise isn't purity, it's just slower.

What's something you changed your mind about this year?

The shift from seeing AI as boilerplate help to using it in paid work is a relatable arc. What was the first real project where it genuinely changed your mind?

Built an automotive demo for YUKTAAI — an AI Vehicle Advisor for used-car marketplaces.

The demo shows how a shopper can go from:

“I’m looking for a used car”

to a confident vehicle decision through smart qualification, relevant car recommendations, vehicle-history signals, comparison support, and dealer lead capture.

Instead of just showing hundreds of listings, YUKTAAI helps shoppers understand which vehicle fits their needs, which one they can trust, and what the next best step should be.

From browsing to buying confidence.

The journey from a vague used-car request to a confident recommendation sounds genuinely useful. Which vehicle-history signal was hardest to integrate into the advisor?