Coding with LLMs: we can talk to computers and we're complaining about it

Updated Oct 18, 2025

October 17, 2025

The title references a classic Louis C.K. bit about how we have incredible technology but everyone complains.^[1] "You're flying! You're sitting in a chair in the sky!" And people complain because the WiFi is down.

When I read discussions about coding with LLMs, this keeps coming to mind. We can talk to computers. We can describe what we want in our own words and get working code back. It's far from perfect but only three years ago, this was science fiction. Now it's Tuesday, and we're upset about it.

No idea.

Nobody does.

Nobody knows if these tools will get significantly better in the next few years or if we're already near the plateau.^[2] Nobody knows if comprehension debt will make codebases much harder to maintain or if we'll develop better practices for managing LLM-generated code. Nobody knows if the job market will contract or expand into roles we haven't imagined yet.

What I do know

Now I can prototype in weekends what used to take months. That side project I've been putting off because the setup alone would eat my entire Saturday? Let's try it now. Yes, I spend more time on review of generated code, checking it actually does what I think, catching subtle bugs. But even with that extra review time, I'm still net-faster than before.

Now I can get unstuck instantly. Remember spending 3 hours debugging a cryptic error? Paste it, get context-aware suggestions, verify they work, keep building. Sometimes the suggestions are wrong. Sometimes they're verbose when simple would do. But the friction is still way lower.

Now I can stay in flow longer. That friction of stopping mid-thought to search for something, translating your specific problem into generic search terms, skimming results, adapting someone else's solution back to your context? It's shorter now. Ask in context, get answers in context. And yes, you double-check the answers, refactor to fit your style, make sure it's not confidently leading you somewhere wrong. But the interruption is smaller.

Now I can experiment freely. Want to try a different architecture? Refactor that messy component I've been avoiding? The cost of experimentation dropped. I'm reviewing and refining instead of typing everything from scratch.

Now I can focus on non-trivial stuff. The mechanical parts (boilerplate, syntax, the repetitive stuff I could do but would rather not) get handled. Much more portion of my development time is now devoted to architecture, trade-offs, figuring out when to keep things simple and when complexity is justified. The parts that LLMs are not good at.

Right Now

I'm learning faster than I have in years. Not because I got smarter, but because the friction dropped. The extra time I spend on code review? Still less than the time I save everywhere else.

Are there downsides? Oh yeah. Will some of the fears turn out to be real? Most probably.

But right now, today, I can talk to a computer and it talks back. It helps me learn. It helps me build. It gives me access to knowledge that used to take hours to find.

We're all sitting in chairs in the sky. The WiFi is flaky, the code isn't perfect, and nobody knows what happens next. But we're still flying.