Ask anyone who lost a weekend to Factorio and they'll tell you about the compulsion—the just-one-more-belt loop that turns midnight into dawn. That's the famous version. The real one is quieter, and it explains why certain engineers are suddenly very good at something nobody taught them in school.
The Spaghetti Is The Point
Every Factorio player's first base is spaghetti by law, not failure. You learn what the game teaches you through building wrong, tearing down, and starting from zero—iteration seven carries forward only lessons, never structure. Fred Brooks wrote about this in 1975: 'Plan to throw one away; you will, anyhow.' The game just turned it into a save file. You cannot design the right factory from the armchair because the problems that matter do not show up on paper—they appear when the belts are running and something backs up two hundred tiles upstream for a reason you could not have predicted until you watched it happen. This is ground-before-you-theorize as mechanics: the tech layer only bites when your factory is large enough to feel its constraints.
The Agent Is the Iron Plate
Here is where most people have the picture backwards. The intuition says the agent is the machine—the assembler standing on the floor doing the work. It is not. In a correctly built factory, the agent is the iron plate: spawned for one transformation, consumed at its station, and dumped. An agent carries no history, no accumulated state, no drift. The durable machinery is the substrate—the dispatch primitives, the chain runners, the supervisors, the belts themselves. That layer persists. The agents do not. This design dissolves exactly the problem the industry keeps paying for: longer context windows, vector stores, summarization layers—everything spent trying to give the plate a memory that it was never supposed to carry.
Two Streams, Opposite Lifespans
Once you see the agent as an item, two streams become visible and they have opposite life spans. The agents are the disposable stream: consumed, dumped, gone, cheap labor that exists for one transformation and never again. The artifacts—code, briefs, reviews, merged PRs—are the other stream. Those persist. Those compound. In Factorio you do not mine iron to make iron; you mine it to build machines that build better machines until your output is more factory. That is the endgame of the game and the real thing: a factory whose product is more and better factory. The workers are cattle. The artifacts are the herd that grows.
Find the Bottleneck or Optimize Nothing
Strip Factorio down to its core and it asks one question forever: where is the constraint right now, and what is it? Every productive hour is spent finding the one belt that is starved, the one machine that is the wall, the single place where the whole line waits. Optimize anything that is not the current bottleneck and you have done precise, satisfying, completely worthless work. Eliyahu Goldratt taught this as Theory of Constraints in 'The Goal' (1984): a system's throughput is set by its binding constraint, and improving anything else is motion without progress. The discipline ports straight to agent factories—your throughput is gated by one thing right now, the credential pool or the review lane, and the work that matters is finding which, not optimizing the six already fast.
Key Takeaways
- Factorio trained a generation in autonomous production orchestration before the job title existed
- Agents should be consumed items (plates), not persistent machines (assemblers)
- State belongs on the belt/artifacts, never inside the worker
- Iteration seven is where you find the right factory, after scrapping six wrong ones
- Throughput is gated by one bottleneck—find it or optimize nothing useful
The Bottom Line
The whole industry is trying to give the iron plate a memory. The factory builder keeps the plate moving and lets the belt carry state. If you've logged Factorio hours and thought it was just gaming, congratulations—you've already been to the apprenticeship that most people are still paying consultants to teach them.