We've published Build an autonomous agent with Fable 5 — a complete, runnable field guide to running the model unattended and being able to trust what it tells you. It's the longest thing on this site, and the most opinionated.
The premise: with Fable 5, intelligence stopped being the bottleneck. The model holds a million tokens of context and runs for hours on its own — so the hard part is now running it like an organization, with delegation, verification, budgets, and earned trust, instead of like a chat window. That layer of files, scripts, and ledgers is what the guide calls an agentic OS. The model supplies the intelligence; the OS supplies the honesty.
It's built in nine layers, in order, and every one ends in a checkpoint you can run against your own setup: the engine (why the same request can cost 10–50× more depending on how you route it), a constitution a script can enforce, separation of powers so nothing grades its own work, a daily heartbeat where Fable decides and cheaper models execute, an earned-trust ledger, standing goals that re-verify finished work daily, a three-way budget, injection defense, a runbook, and a 30-day rollout.
There's a narrated walk-through video on the page (about two minutes), a machine-readable twin for agents at /api/agent.json, and it's free to read. If you've worked through the cookbook, this is where the recipes come together into one system.