Programming still sucks (stvn)

Stvn's reply to Peter Welch's 2014 Programming Sucks (stilldrinking.org/programming-sucks). Same shape — drunk-at-a-party voice, extended metaphor, one anchor character — pointed at a different target. Welch's piece was about software complexity. Stvn's is about what corporate AI adoption did to apprenticeship and institutional knowledge.

The essay's argument is structural, not technical. Two parts move it:

The burning ship metaphor sets up the workplace experience. You're catapulted onto a custom-built ship by a previous captain who left mid-renovation; the sails aren't connected to the mast; an internal-combustion engine is half-bolted to the stern. The "sailor" you ask is an LLM that glitches and offers a better stairs implementation. The navigator is a doll that says "onward and upward" when you press a button. The doll catches fire. Repeat at the end.

The Sara character is the load-bearing rhetorical move. Somewhere below decks, in a cabin nobody knows about, a woman in her mid-50s named Sara keeps a 2016 cron job alive. The job runs at 3am. It does something critical. The comment at the top says # DO NOT CHANGE!!! Ask Ben. Ben left in 2019 and passed a few years back. Sara inherited the job from Ben, who mentored her since 1998. The module the job depends on lives on a USB stick she found in Ben's desk after his funeral. The cron job pays salaries. You don't know this.

What the essay actually argues

The Guy at the party asks the wrong question. Stvn's answer when drunk enough to give it: "AI didn't take our jobs. Greed did. Same greed that moved factories to Bangladesh and keeps slaves in cobalt mines in the Congo, wearing a new mask." AI is the alibi for a decision that had already been made — the runway got cut, the CFO had a spreadsheet, the CEO had been to an offsite where someone had shown him a demo of an agent writing a feature in fourteen minutes, and he believed it the way people believe things when they want to believe them.

The middle section is the indictment of the engineer who signed the list. You knew. You'd watched Goodhart's Law eat velocity, story points, test coverage. You knew the DORA metrics were already telling you what happens to deployment stability when tooling outpaces judgment. You knew what happens to a codebase when the people who'd catch errors get pushed out. You signed off anyway, because the alternative was losing the job.

The Sara section is the consequence. Juniors weren't valuable for what they produced; they were valuable for who they would become — the senior who knows where the bodies are buried. We optimized for output, and abolished apprenticeship. A few years from now, we'll wonder where all the seniors are. We shot them. Nobody will remember.

Sara is the visible proof of what was lost. She came up through the apprenticeship that was abolished — Ben, 1998, the USB stick. She is the pipeline. When she dies the thing that produces people like her is already gone; the cron job dies with her; salaries stop being paid. You won't find her replacement, because you broke the machine that makes her.

What's load-bearing

Three things keep this essay from collapsing into venting.

The Sara character is concrete and falsifiable. You can imagine her, you can count her. She's not a metaphor for institutional knowledge; she's a specific person who exists in plenty of real companies. The essay's force comes from naming what you already half-knew was there. Same trick Welch used with the Toyota leap-second story — pick one specific thing that turns out to be load-bearing for everything.

The CEO-believed-the-demo paragraph is the only direct attack on AI itself. Stvn doesn't argue AI doesn't work. He argues the belief is what changed — "the way people believe things when they want to believe them." This is the same dynamic as ceo-ai-psychosis: the agent-orchestration aesthetic is doing the social work, the underlying productivity is unverified.

The "you signed the list" passage refuses to make the engineer a hero. The narrator is one of the people who signed off. I signed a list too. We're still pointing at each other about whose list was worse. This is what makes it different from most pieces in anti-llm-discourse — it's not a critic-vs-corporation piece; it's a self-implication piece. Same posture as building-syntaqlite-ai (vibe-coding failure as autobiography), simonw-vibe-coding-agentic (Willison admitting his line collapsed in his own practice).

What it shares with the cluster

i-am-an-ai-hater (Moser) and this piece are the two most rhetorical entries in anti-llm-discourse. Different moves: Moser refuses the polite frame and walks out; stvn stays inside the frame but reframes the question. AI didn't take our jobs. Greed did.

The "no more juniors" passage echoes ai-great-leap-forward's institutional-knowledge-loss thread and skill-atrophy-supervision-paradox's mitigation problem (the apprenticeship pipeline that produces capable supervisors is the thing being abolished). It pairs with ceo-ai-psychosis (HandyAI's piece on executive AI-orchestration as performance) — same dynamic from the worker-side instead of the executive-side. And the Goodhart / DORA references put it inside the engineering / craft case alongside no-silver-bullet-llms and peril-of-laziness-lost, even though the rhetoric isn't engineer-coded.

What's not in the essay

No prescriptions. No "what to do instead." No call to organize. The essay's bet is that the reader is already an engineer at one of these companies, has already signed a list or watched someone else sign it, and doesn't need a plan — needs to be told they weren't crazy. I'm not writing this for the Guy at the birthday party. I'm writing it for you.

The closing is the doll catching fire — the same image from the metaphor section. Repetition as the formal device of inevitability.

Sources

• programming-still-sucks — full body
• Author blog: stvn-blog
• Reference: Peter Welch, Programming Sucks (stilldrinking.org/programming-sucks) — this essay's stated dialogue partner