What have you built with Fable? Nothing, it's like asking a professor to change lightbulbs

August 17, 2026 · Restorative · 2 min read

Fable is the superior Harvard professor. You don't hire one to build a brick wall. Ask your best model the question you can't answer, then let the cheap one build.

"What have you built with Fable 5 today? Post it in the comments." The question runs under every thread about the model, and it measures the wrong thing.

Fable is the superior Harvard professor. You don't hire one of those to build a brick wall or keep changing lightbulbs. You bring one in when nobody in the room can agree on what the actual problem is.

The work that decides everything else

In an intentional AI sprint with an impact organisation, the expensive mistakes live in the framing. Automating a reporting process nobody reads. Building a dashboard for a number the team already distrusts. Picking the workflow to change because it was the easiest one to describe in the kickoff.

That's the conversation I take to Fable. What's the real constraint here. What happens to the people downstream of this change. Which of these three options survives contact with a team that's already short two roles. It argues back, it holds the whole messy context at once, and it names the part I've been avoiding.

None of that produces anything you can post in the comments.

Then hand the bricks to someone else

Once the shape is settled, building is a different job with different requirements. Sonnet writes the code. It's fast, it's cheap, it doesn't get tired at hour six, and by then the work is specified well enough that speed is the thing worth buying.

Point your strongest reasoning model at that and you're paying professor rates for lightbulbs. The lightbulb gets changed. You've also spent the one resource in your stack that could have told you the wiring was the problem.

The excitement behind the daily build logs is real. The scoreboard it runs on counts only what you can screenshot, and the highest-leverage thing a model does for you often leaves no artifact at all.

Open your best model tomorrow and ask it something you can't answer yourself. Let the cheap one build whatever it tells you.