How I Became an Apocaloptimist Engineer

This week I watched The AI Doc: Or How I Became an Apocaloptimist on Netflix. Daniel Roher, expecting his first child, goes around asking the people building AI whether the world his kid inherits will be wonderful or over. He gets both answers, sometimes from the same person, and lands on a word for holding both at once: apocaloptimist.

I recognised the two camps immediately. Software engineering has the same argument running in every Slack channel and every conference hallway.

The two camps #

The doomers say the craft is dying. Nobody writes code anymore, so nobody understands it. Juniors will never learn what a stack trace means. And the point that comes up most often, from people I respect: the fun is gone. They got into this job to write code, and now they watch a machine do it.

The boosters say engineers are a transitional species. Describe the product, get the product. Ten times the output, no more teams, and if you are still typing you are already behind.

Like every argument with two loud ends, the interesting part is in the middle. Here is a slider. Go find your spot.

What actually changed in my day #

I use Claude Code in the terminal for most of the day, and Codex from OpenAI for some of it. Here is where AI fits into my workflow:

• Features and implementation logic. I use AI to implement features, business rules and data flows. We work through the approach, the code and the edge cases together.
• Boilerplate and scaffolding. New endpoints, models, config files, CLI wiring. The shape is known, the typing was never the interesting bit.
• Tests and fixtures. I say what the behaviour is and I get the test file, the mocks and the fixture data. When we were strangling a 15-year-old monolith, deciding what parity meant was the job. Writing the hundredth parity test was not.
• Infra and CI. GitHub Actions, Dockerfiles, Amplify config. I read the diff. I do not want to remember the YAML.
• Docs, my CV, this blog. The opinions are mine. The first draft and the rewriting rounds are shared.

The thing the doomers did not predict, at least for me: I have more energy for the job, not less. The boring parts used to eat the afternoon. Now I have more time for the parts I actually like: deciding what to build and getting a feature to work.

Still in the room #

Skipping the typing is not skipping the thinking. Every line still gets read. When generated code is wrong, and it is wrong often enough, I work through the problem with the tool until the behaviour is right. I need to understand what the code does and why it works.

If you let the tool write and never read, you get the doomer scenario: code nobody understands. The middle takes discipline: use the tools, stay involved and take responsibility for what ships.

That is the apocaloptimist position, translated to a keyboard. The world might end, and also I shipped three things before lunch and enjoyed it.

About this post #

This post was written with AI help. I brought the documentary, the idea and the opinion. Claude Code drafted the text, built the slider above and opened the pull request that published it. I reviewed it like any other PR, which is the whole point.