Claude Fable 5 just dropped and I'm speechless...
TL;DR
Claude Fable 5 changes the core question from correctness to direction: Alex says earlier models made you ask "is Claude doing the work right," but Fable 5 is strong enough that the new bottleneck is whether it is doing the right work toward the right goal.
Anthropic's playbook is to treat the model like a peer, not a junior assistant: The three tips he highlights are to use Claude as a thought partner, give it high-level goals instead of micro-steps, and be more ambitious about what you ask it to build.
His recommended workflow is "advanced plan mode" followed by /goal loops: He first has Claude ask as many clarifying questions as needed, then pastes the resulting spec into /goal so the agent keeps checking itself against success criteria until the app is done.
The demo app is intentionally big and still ships in one pass: Using Claude Code in Ghosty with Fable 5 on auto mode and high effort, he asks for a web app with local storage, deep integrations, a calendar, notes, Pomodoro stats, kanban board, and an arcade break game.
The finished product actually works across all the requested features: Alex shows tasks, calendar events, week view, journal-style notes, Pomodoro streaks, drag-and-drop kanban, and a Snake game that records scores, and says it shipped with zero bugs.
The pricing may signal a broader industry shift away from cheap subscriptions: Fable 5 is only in subscriptions until June 22, after which it moves to API-only access, costs double Opus, and Alex frames that as a possible sign that frontier models may stop being heavily subsidized.
The Breakdown
Alex Finn calls Claude Fable 5 "the biggest model release of the year" and then stress-tests the hype by having it build an entire productivity app in one autonomous loop, complete with calendar, notes, Pomodoro timer, kanban board, and a working Snake game, with "zero bugs." His bigger point is the mindset shift: stop babysitting AI step by step and start giving it ambitious end goals, because the real question is no longer whether it can code correctly, but whether it is doing the right work.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
The Art of Tasteful Prompting
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.

Playbook
The Codex /goal Playbook
OpenAI shipped /goal for the Codex CLI. It turns a prompt into a persisted, self-continuing contract.