We've been building software since 2016. For the past year, AI coding tools like Claude Code, Cursor, and GitHub Copilot have been part of our daily workflow. Here's what changed and what didn't.
What AI actually does for us
It handles the repetitive parts faster. Boilerplate, CRUD operations, test scaffolding, migrations. Tasks that used to eat hours now take minutes. Our developers spend less time typing and more time thinking.
It catches mistakes sooner. AI is a decent first-pass reviewer. It spots missing edge cases, suggests simpler approaches, flags potential issues. It doesn't replace code review, but fewer problems make it to the pull request.
It makes research faster. Instead of digging through documentation for 20 minutes, we describe what we need and get a working starting point. We still verify everything because AI gets things wrong regularly. But the starting point saves real time.
What AI doesn't do
It doesn't know your business. AI can write a function, but it can't tell you whether that function should exist. It can't say "this feature will confuse your users" or "you should solve this with a process change, not code." That's product thinking, and it comes from years of building products with real users.
It doesn't own anything. Your product isn't a one-time build. It evolves over months and years. When something breaks at 2am, or a new feature conflicts with something built 18 months ago, you need someone who knows the system inside out. AI generates code. It doesn't care if that code is maintainable, secure, or will scale. Someone has to own those decisions, and that requires context no model has.
It doesn't say no. This is the big one. AI will happily build whatever you ask for. A good development partner will push back. "You don't need a custom solution for this." "This feature isn't worth the complexity." "Let's validate the idea before we build it." The most valuable thing a developer can do is stop you from building the wrong thing.
What this means for you
If you're evaluating development partners, ask how they use AI. If they don't use it at all, they're leaving speed on the table. If they position themselves as "just developers," AI is coming for that value proposition fast.
Look for teams that use AI as a tool but sell thinking as the service. Teams where the senior developers spend their time on architecture, product strategy, and code quality instead of churning out more features.
The agencies that were selling hours of code output are in trouble. AI made that a commodity. The ones that were already doing more than writing code just got faster.
Where we stand
AI made us faster. It didn't change what we do.
We still challenge requirements, still push back on features that don't make sense, still architect systems that hold up years later. We just spend less time on the parts that didn't require judgement in the first place.
