I Haven't Coded in 3 Months

March 14, 2026

i need to say something out loud because it still sounds weird even to me: i haven't written a line of code by hand in three months. not a function, not a variable, not even a css property. and i've shipped more in those three months than i did in all of 2024.

i just... talk now. i describe what i want, what i'm thinking, what the thing should feel like. and it gets built. not a rough draft i have to fix. the actual thing. production-ready, deployed, done.

the moment it clicked

there wasn't a single moment honestly. it crept up on me. i started using claude code for small stuff. "fix this bug." "add a loading state here." normal stuff. and then one day i realized i was describing entire features in plain english and they were just... appearing. correctly. with tests. with edge cases handled.

if i'm being specific, claude opus 4.5 was the real turning point. that was the first time it felt like i was working with something that actually understood what i meant, not just what i said. and now with opus 4.6 and a million-token context window, it's not even close. it holds the entire codebase in its head. i can say "you know that thing we did three hours ago? do that but for the payments module" and it just knows. no re-explaining, no pasting files in, no context lost. it remembers everything. that changes the whole dynamic from giving instructions to having a conversation.

and i thought okay maybe this is a toy and i'm getting lucky. but then i built an entire infrastructure layer like that. then an integration. then a full frontend. and at some point i stopped opening my editor entirely.

that was three months ago. now my days look completely different. i spend most of my time on two things: planning what to build, and reviewing what got built. the actual building happens in between, almost on its own. i went from being the person writing the code to being the person deciding what the code should do and making sure it did it right. architect and reviewer. the middle part just... filled itself in.

it feels like cheating

i'm not going to pretend this doesn't feel weird. i spent years learning to code. actual years. debugging segfaults at 2am, reading docs that made no sense, building things the slow painful way. that work mattered. it's why i can describe what i want precisely enough for it to come out right. you can't skip the understanding part.

but the typing part? the syntax? the semicolons and brackets and remembering which arguments go where? turns out that was never the hard part. the hard part was always knowing what to build and why. the code was just the transcription.

the skeptics

i get it. i was one of them like a year ago. "ai-generated code is sloppy." "you're building technical debt." "you don't really understand what it's producing." "real engineers write their own code."

and look, some of that is true in the wrong hands. i wrote a whole post about deleting ai slop. the difference is i know what good code looks like. i can read what comes out, i can evaluate it, i can push back when it's wrong. i'm not blindly accepting whatever gets generated. i'm directing it.

it's the difference between someone who can't drive using autopilot, and a professional driver using autopilot. same tool. completely different outcome.

unstoppable is not an exaggeration

this is the part where i'm going to sound like i'm overselling it. but i genuinely feel like i can build anything now. not anything well necessarily. not anything at scale on day one. but anything. the barrier between idea and working software has basically collapsed.

i had an idea for a feature on tuesday. it was in production by wednesday afternoon. not a hack. proper implementation, tested, reviewed, shipped. that used to be a sprint. now it's a conversation.

and it compounds. the faster you can build, the faster you can learn what works. the faster you learn, the better your next idea. i'm iterating at a speed that would have been physically impossible two years ago. it's not 10x. it's a different game entirely.

the barrier between idea and working software has basically collapsed.

so what actually matters now

here's the thing nobody talks about. when building is this easy, code stops being the moat. anyone can build anything. the person with the best prompt can ship the same feature as the person with ten years of experience. maybe not as clean, but functional. out the door. in users' hands.

so if everyone can build everything, what separates you?

taste. judgment. knowing what to build.

the companies that win from here aren't the ones with the best engineers (sorry). they're the ones who understand their users so deeply that they build the right thing on the first try. the ones with opinions. the ones who can look at infinite possibilities and pick the one that actually matters.

i wrote about the ai hangover in january. about how slop accumulated and the industry was going to course-correct. this is the course correction. not less ai. more intentional ai. the tool got so good that the tool is no longer the bottleneck. you are. what you choose to point it at is everything.

i haven't coded in three months. but i've never felt more like an engineer. because it turns out engineering was never about the code. it was about solving the right problems. and now there's nothing between me and the problem except a conversation.

that's the superpower. not that you can build faster. that you finally have the space to think about what's worth building at all.