“It’s been three months since I last wrote a line of code.”
That’s the opening line of a piece from Between The Prompts that’s been making the rounds on Hacker News, and it hits different from the usual “AI will replace developers” hot take. Because the author isn’t unemployed. He’s still delivering features. Still solving architecture problems. Still reviewing pull requests. He just… doesn’t type the code anymore.
The gradient, not the cliff
The piece describes a gradual shift that most developers using AI tools will recognise. It starts with Copilot autocomplete — nice, helpful, a productivity boost. Then agent mode. Then Claude Code. And then one day you look up and realise you’re not writing code. You’re directing code.
The author’s team built what they call “the Conductor” — a set of Claude Code skills covering their entire workflow:
- Research — Claude pulls from the codebase, Notion, Slack, the web, and produces a research document with open questions
- Shaping — Defining the problem and solution clearly enough that a non-technical person could understand
- Planning — Breaking the shaped feature into tasks with completion criteria
- Building — Claude executes the plan
- Compounding — Capturing lessons learned for future development
The phase that surprised the author most was research. Even when he had no idea where to start, the research phase got him oriented. “It’s a cure for the ‘blank page’ syndrome,” he writes.
This isn’t one developer’s quirky experiment. He says most of his teammates are going through the same transition. “This is a team transition, not a personal experiment.”
Why this matters for every developer
We’ve covered the prompt engineer narrative and the career panic around AI reskilling. But this piece sits somewhere more interesting than both. It’s not about a new job title. It’s not about panic. It’s about identity.
If you’re a developer and you don’t write code, are you still a developer?
The traditional answer was: no. Developers write code. That’s the job. That’s the identity. It’s literally in the name.
But the Between The Prompts piece suggests a different framing. Developer value is migrating from how to build to what to build and why. From syntax to judgment. From implementation to architecture. From typing to deciding.
That’s a hard identity shift. Developers build their sense of self around craft — clean code, elegant solutions, the satisfaction of making something work. When an AI can produce working code faster than you can type it, that craft identity gets wobbly.
The NZ developer angle
New Zealand’s tech sector is small but concentrated. Most NZ developers work in teams of 5-20, not the massive engineering orgs of Silicon Valley. That actually makes this transition harder in some ways.
In a small team, there’s less room for specialisation. The person who shapes the feature and the person who builds it are often the same person. If AI takes over the building, the shaping has to get much better — and most developers have never been trained to shape. We’ve been trained to implement.
NZ’s tech education system — bootcamps, polytech courses, university CS programs — is still overwhelmingly focused on teaching people to write code. Not to direct AI agents. Not to think in systems. Not to shape problems. The AI education gap we’ve written about before isn’t abstract. It’s this, right now, in every developer’s career.
The uncomfortable middle ground
Here’s what I think the Between The Prompts piece gets right that most AI commentary misses: this isn’t a cliff. You don’t wake up one day replaced. It’s a gradient. And most developers are somewhere on it already, whether they’ve acknowledged it or not.
Copilot autocomplete? You’re on the gradient. Cursor agent mode? Further along. Claude Code running your whole pipeline? You’re where the author is.
The question isn’t whether the gradient leads to “developer” becoming obsolete. It’s whether the profession can expand its identity fast enough to stay meaningful as the implementation part gets automated away.
My read: it can, but it requires a cultural shift that’s going to be painful. Developers love code. Losing the daily practice of writing it — even if you’re still “a developer” — is a kind of grief. And pretending it isn’t doesn’t make the transition easier.
What to actually do
If you’re a developer reading this, some practical advice from watching this space:
- Learn to shape. The skill of defining a problem clearly enough that an AI can solve it is not the same as the skill of solving it. Practice it deliberately.
- Learn to evaluate. Code review becomes the core skill when you’re not writing the code. Can you tell good AI output from bad? Fast?
- Learn to compound. The author’s “compounding” phase — capturing lessons for future work — is the meta-skill that makes the whole loop work.
- Don’t panic, but don’t coast. The gradient is real. Being further along it isn’t a failure. Not noticing you’re on it is.
Sources
- If Claude Writes the Code, What Makes Me Still a Developer? — Between The Prompts
- The Prompt Engineer Is the New Code — Singularity.Kiwi
🔍 THE BOTTOM LINE: The developer identity crisis isn’t about losing your job to AI. It’s about keeping your sense of purpose when the thing you were best at — writing code — isn’t the thing that matters most anymore. The developers who figure out what does matter most will be fine. The ones who refuse to ask the question won’t.
