Code Is Cheap Now. Working Systems Are Not.

Anthropic re-released Claude Fable 5 on July 1. It is the most capable AI model ever made generally available. State of the art on nearly every benchmark. Your feed is full of benchmark charts right now.

The benchmarks are the wrong story.

The right story is one number from Stripe's early testing. Fable 5 migrated a 50-million-line Ruby codebase in one day. The same project was estimated at more than two months for a full engineering team. That is not a productivity gain. That is a repricing of what code costs.

Zoom out and the data says the same thing. Developers pushed roughly 986 million commits to GitHub in 2025, up 25% year over year. Pull request volume grew 29%. An estimated 29% of new Python functions in the US are now written by AI. And frontier capability rents for $10 per million input tokens. Anyone with a credit card has access to the same model Stripe used.

When something gets cheap, the constraint moves. For 60 years the constraint in software was writing the code. That era ended somewhere in the last 18 months. The new constraints are everything around the code: picking the right workflow to automate, wiring the system into the tools your business already runs on, getting your data clean enough to use, reviewing and testing what the model produces, getting your team to actually adopt it, and keeping it running after launch.

Fable 5's own launch proved the last point. The model shipped June 9. The US government put it under export controls on June 12. It came back July 1. That is 19 days offline. If your operations depended on that one model with no fallback, you were down for nearly three weeks through no fault of your own. A working system has fallbacks, monitoring, and an owner. A model has none of those things.

Here is what this means if you run an operations-heavy business. The old excuse for living in spreadsheets and generic SaaS was that custom software cost too much. That excuse is dying. A system built around your actual workflow now costs a fraction of what it did three years ago, and the price keeps falling. The companies that convert that into shipped, adopted systems will pull away from the companies still running pilots. Fast.

Most businesses do not need a smarter model. The models are already good enough, and they get better every quarter without you doing anything. What businesses need is the system around the model: scoped to a real bottleneck, built, integrated, tested, adopted by the team, and operated after launch.

The code got cheap. The work moved. That is the whole story.