.png)
I spend my days closer to how software ships than what it does, and the loudest thing in my world right now is AI. It’s not new, though. Machine learning’s been around since the 1950s. ChatGPT just made it mainstream, and people act like it appeared out of nowhere.
Everybody is writing thought leader posts about it. Some are thinking out loud and trying to make sense of things. Others are doing it to get famous overnight. This is one more of those posts. I’ll let you decide which kind it is.
I’ve been chewing on this for a year or two, mostly trying to tell what’s real here from what’s just loud.
The good
- People feel more productive, and for the grunt work they are.
- It’s the fastest way I’ve found to learn something new, a search that explains things back to you.
- Anyone with an idea can stand up a rough proof of concept and make it feel real, even when it’s nowhere near production ready.
- Writing the code stopped being the slow part. Everything around the code is.
The bad
- It’s easy to trust what the model hands back without checking it, right up until it breaks something.
- Laziness is setting in. Not the good kind where you automate the boring tasks. The kind where you outsource your thinking and slowly forget how to do it yourself.
- There’s slop everywhere. Ugly generated images, walls of text nobody can get through.
- We’re shipping more code than we can understand, without the guardrails to catch it.
What I actually think
People are focused on the wrong things. The energy goes into moving fast and skipping steps to get features out. The questions I see most are which skills or MCP servers someone should install.
In my opinion, that’s backwards. What’s worth your time is understanding how LLMs and the agent loop actually work. That knowledge lives further down the stack and it’s admittedly less flashy. Once you have it, the questions about skills and MCP servers mostly answer themselves.
So before you chase the new shiny thing for clout, ask a different question. What do you want to get out of using AI? Don’t use it just to use it. Use it when it solves a real problem.
And “I can go faster and ship more” isn’t a good enough answer. Faster toward what? Ship what, exactly? Sort out what you’re trying to accomplish first, then do the thing. Don’t skip step one to get to step two.
The unglamorous stuff matters more now
Writing code got cheap. Everything around it didn’t, and that’s the part AI can’t do for you.
Fundamentals are one unglamorous thing people skip for the shiny stuff. How you ship is another. Most teams focus on application features. That makes sense, since it’s why you’re building in the first place. The trouble starts when that’s all you watch.
I once spent a few hours cleaning up after vibe coded changes someone shipped without understanding them. Nothing stopped it, because the focus and the investment had gone to features instead. Solid automated tests and a sturdy pipeline would have caught that automatically, before it ever reached production. That’s confidence you can’t get any other way today.
Get that right and you might not even need to read every line. The guardrails buy you confidence it works, not an excuse to stop thinking. Without them, that’s just shipping code nobody understands. With them, you can step back and let the pipeline do its job.
None of this is glamorous, and nobody notices it when it’s working. It’s like roads. Nobody thinks about them until they’re full of potholes. Putting time in here feels slow at the start. It pays off later. I’ve watched it happen.
AI is a tool
AI is a tool like any other. We should use it with purpose and have it serve us, not the other way around.
This is just what good engineers do anyway. The ground shifts constantly, so we keep learning. We take what helps and we leave the hype behind.
Use AI as a thinking partner to sharpen your own ideas. Don’t hand them over to it instead. Remember what it’s like to be human and bring that into the work. Very smart autocomplete will flatten everything into the same generic shape if you let it.
The loudest voices will move on to whatever’s next. I’d rather sort the signal from the noise and keep what holds up. There’s real value in here, and I’m betting on it.