Twenty years in, I think this is the endgame for programmers

I've been writing software for twenty years. I've done the late nights chasing memory leaks, the framework migrations, the deployments that went sideways at 2am. So believe me when I say this isn't a hot take from someone who just discovered AI last month. This is what I genuinely see coming, and I think it's the endgame for programming as we've known it.

The best time to start is right now

If you're thinking about becoming a programmer, you've picked the best moment in history to start. Not despite AI agents — because of them. The wall that used to stand between "I have an idea" and "I shipped something" has mostly collapsed. You no longer need years of syntax, design patterns, and build tooling before you can produce something real. You need curiosity, a clear idea of what you want to build, and the willingness to learn the tools.

And that's the skill for a novice today: learn the tools deeply. Learn how to work with coding agents — how to give them context, how to break work into pieces they can verify, how to review what comes back, how to keep your sessions lean and your costs down. The clever tricks now live at the tool level, not the language level. Mastering the agent is the new mastering the editor.

Fundamentals are valid — but no longer essential

I know this is where people push back. "Fundamentals matter." And look — they're not wrong that fundamentals are valid. Knowing how memory works or why an algorithm is slow will never hurt you. But valid and essential are different things.

I learned fundamentals because I had no choice; nothing worked without them. A novice today is in a different world — the age of Mythos. Anthropic's new frontier model scores 93.9% on SWE-Bench and can reason across an entire codebase at once, and its public version, Fable 5, shipped this week. The agent knows the fundamentals — better than most of us, honestly. Insisting that newcomers grind through years of theory before they're allowed to build is like insisting drivers learn to rebuild an engine before touching the wheel. It's noble. It's also no longer the gate.

Prompt engineering is already obsolete

Here's the one that surprises people: I think prompt engineering is dead. All those carefully crafted system prompts, the "you are an expert senior engineer" incantations, the step-by-step instructions on how to structure the code — we can let them go.

Tell the agent what to build, not how to build it. These models are smarter and more insightful than we give them credit for — often more than we are, in their breadth. When I hand-craft a detailed prompt telling the model exactly how to approach a coding problem, I'm not helping it. I'm shrinking it down to the size of my own assumptions. Every time I've stepped back and described the outcome instead — what the feature should do, who it's for, what done looks like — the result has been better than what I would have specified.

We spent two years learning to whisper to these models. The models outgrew the whispering.

The endgame: requirements in, software out

In the near future — years, not decades — I don't think programmers will even look at the code. The loop will be: draft the requirement, watch the agents build → test → loop until it passes, review the behavior, ship. The code becomes an implementation detail, the way assembly became an implementation detail, the way machine code did before that.

That's not a tragedy. It's the same story this industry has always told: we keep moving up the abstraction ladder, and every time, the people who climb fastest win.

What actually matters now

So if it's not fundamentals, and it's not prompting — what is it? It's deep, clever understanding of the product you're building. The features. The users. Why this thing should exist and what makes it good. That's the part no agent can draft for you, because it lives in your judgment, not in the codebase.

How we build, deploy, and scale — those are questions for the agents now. What to build, and why — that's ours. That's the whole job, soon.

Twenty years in, I'm not mourning it. I'm excited. The boring parts are leaving the profession, and the interesting part — the thinking — is all that's left.