← The AI Hype Audit — all 645 verdicts

PARTLY

Partly true: Claude Code's advisor and Fable 5.2 are real and Anthropic does recommend Sonnet with a Fable advisor, but the advisor is not limited to three fixed moments and the 'JEV' layer is a third-party add-on, not part of Anthropic's trick

Facebook · Nov 2026

The claimAnthropic shared a Claude Code trick: run Sonnet 5.5 as the main model, Haiku 5.5 subagents for search and tests, and the new Fable 5.2 as an on-call advisor via 'claude --model sonnet --advisor fable'. Fable 5.2 on high effort matched Opus 5.5 on max on FrontierCode at about half the cost. Fable only steps in before a plan locks, when the same error repeats twice, and before the task is called done. Add JEV to route micro-forks in milliseconds.

Most of the core is accurate. Claude Code has an experimental advisor tool, documented by Anthropic, that lets a cheaper main model consult a stronger one, and the advisor reads the full transcript each time. 'claude --advisor fable' is a real (hidden) launch flag, and Anthropic's own docs list Sonnet main plus Fable advisor as a recommended pairing. Fable 5.2 is real too: Anthropic announced it November 9, 2026, said it can act as an advisor to Haiku, Sonnet or Opus, and reported that on FrontierCode 2.0 at High effort it scored 45.3% against Opus 5.5's best of 45.5% at roughly half the cost per task. The overreach is in the details. Anthropic's docs say Claude decides when to call the advisor; it tends to consult before a plan, on repeated errors and before finishing, but it is model-driven, not three fixed triggers, and a CLAUDE.md rule only nudges it. 'JEV' is TypeSafe's Jev, a separate third-party decision model with its own billing; it is not part of Claude Code or of anything Anthropic shared. Fable as advisor needs Fable access and on some plans bills to usage credits, which the caption does mention. The page sells a paid Facebook subscription at the end.

What holds up

  • Contact sheet viewed: an animated terminal mock-up ('Claude Code advisor tip', Sonnet 5.5 writes, Fable 5.2 decides, JEV fork layer) credited to an X handle; no speech in the audio.
  • Anthropic's Claude Code advisor docs (code.claude.com/docs/en/advisor) confirm --advisor, /advisor, full-transcript review, Sonnet + Fable pairing and that Claude, not a rule, decides when to consult.
  • Anthropic's Nov 9, 2026 Fable 5.2 / Mythos 5.2 post confirms the advisor role and the FrontierCode 2.0 result (45.3% vs 45.5% at about half the cost).
  • Jev confirmed as TypeSafe's separate 'System One' decision model, usable as a router around Claude Code only through your own integration.

What doesn’t

  • Paid Facebook subscription pitch at the end of the caption.
  • 'Fable only steps in at three moments' overstates control; consultation timing is model-driven.
  • JEV is a third-party product bundled into an 'Anthropic trick' it is not part of.
  • Fable advisor requires Fable access and can bill to usage credits.

The catch

The Sonnet-plus-Fable-advisor setup is real and Anthropic-endorsed. The precise three-trigger rule and the JEV routing layer are the poster's additions, so you cannot get that exact behavior from Claude Code alone.

How to actually do it

  • Run 'claude update', confirm Fable access with '/model fable' (accept the usage-credits consent if your plan asks), then start with 'claude --model sonnet --advisor fable' or set it with '/advisor fable'.
  • Add a plain instruction in CLAUDE.md such as 'consult the advisor before finalizing a plan and before declaring done'; treat it as a nudge, not a guarantee.
  • Create read-only Haiku subagents for search and test runs if you want them; they are optional and separate from the advisor.
  • Skip JEV unless you already use TypeSafe and want to wire a router yourself; check '/usage' after a few sessions to confirm the savings.

A genuine Anthropic feature and model, dressed up with an extra third-party layer and stricter trigger rules than the product actually offers. The setup command works; the 'only three moments' promise does not.

Confidence
High
Posted by
an AI tips page (about 5,000 plays); reel sent to Buddy Sun Nov 15, 2026, 12:23 AM

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.

© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.