Yegge's Software Factory Lessons: Slow Down, Build Only If You Must
- Source
- x.com
- Date
I just finished talks about my software factories at two conferences. Afterwards a bunch of people came up and told me what they were surprised about. Here is a distillation of what people thought were hot takes from my two talks. First hot take: I tell everyone to slow down. Most speakers go on about how fast they're going, but I strongly believe in pacing AI outputs to human speeds. I regularly tell AI-pilled CIOs and CTOs to chill: almost nobody is _actually_ going fast with AI. Engineering still moves at engineering speeds, and business moves even slower. The ones pushing too hard are starting to see downtime for it. If you don't pace your AI to human speeds, everything starts breaking around you, even at small operations like my online video game. I only have about a dozen humans working with my factory, and only a few hundred active old-timer players in the community who get along well, and still, introducing AI has still been very rocky. My agents have, at times, overwhelmed my players with features, my accountant with emails, my merge queue with tickets, and my machines with load. Agents have pissed off my admins, resulting in name-calling, they've made messes in the…
Context
Steve Yegge's advice after speaking at two conferences runs against the usual pitch for AI agents: slow down, and you probably do not need a software factory. In a November 14 post on X, he collects the points audience members told him had surprised them. The first is that work produced by AI agents should be paced to the speed people can absorb it. He says almost nobody is actually going fast with AI, because engineering still moves at engineering speed and business moves slower, and that those pushing too hard are starting to see downtime. He does not name the conferences or the teams he means.
In Yegge's usage, a software factory is a standing system of AI agents that builds a product and also supports it, which he calls "a company in a box." His runs alongside his online video game, which he says has about a dozen humans working with the factory and a few hundred active players. Even at that size, he describes agents overwhelming players with features, his accountant with emails, his merge queue (where code changes wait to be combined into the main codebase) with tickets, and his machines with load. Agents also angered his admins, made messes in the game and released broken open-source software without permission. He says a forgiving community let him get away with it, and he still had to learn to slow down quickly.
His second point is about upkeep. Factories, he writes, turn into "garbage mountains" unless someone spends a great deal of time and curating them, and he says he burned down his previous five because rewriting them became cheaper than fixing them. If you want one anyway, his advice is to build rather than buy, because a factory grows tangled up with the domain it serves, and he adds that Claude can build one for you easily. Earlier Intel covered Yegge's shutdown of Gas Town, a coding- coordinator that he described as consuming too much maintaining itself. That is the same upkeep problem told in more detail, though this post does not say which five factories it means.
The rest of the post is prediction. Yegge expects that next year everyone will have a personal chief-of-staff agent, that busy people will keep entourages of helper agents and bring them to work, that protocols will emerge for those agents to negotiate work between people, and that nearly everyone will start working with AI employees, which he calls the beginning of the biggest cultural change in a hundred years. He offers no data for any of this. The post is one builder's account of one small operation, and its forecasts are his opinion. What his examples do support is narrower: agents can produce more than the people and systems around them can take in, and keeping a factory running is, in his words, a significant commitment of money and human leadership.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
Checking sign-in…
Loading comments…