← The AI Hype Audit — all 645 verdicts

REAL

Holds up: this Vietnamese-language explainer describes Claude Code's experimental advisor tool accurately, including its full-context cost and the fact that Claude, not a rule, decides when to consult

Facebook · Nov 2026

The claimClaude Code has an Advisor Tool: a main model (for example Sonnet) does the coding while a stronger advisor model (for example Opus) is consulted at key moments, such as before choosing an architecture, when the same bug repeats, and before calling a task done. Claude decides when to consult; the advisor sees the full conversation including tool calls, which adds token cost. It differs from subagents, is experimental, needs the Anthropic API and may cost extra.

This caption is a careful, accurate summary of Anthropic's own documentation. Claude Code's advisor tool is real and marked experimental. It pairs a main model with a stronger advisor, Claude decides when to consult it (typically before committing to an approach, when an error keeps recurring, and before declaring done), the advisor receives the full conversation including every tool call and result, and Claude generally follows the advice but pushes back when the code or test results contradict it. All of that matches code.claude.com/docs/en/advisor almost line for line. The caption also gets the tradeoffs right: each consultation re-reads the whole transcript, which costs extra tokens and is not cached, and the advisor is different from a subagent, which does a separate task rather than advising the main agent. One small nuance: 'requires the Anthropic API' means it runs on Anthropic's own platform (plus Claude Platform on AWS and Anthropic-managed Google Cloud), not Bedrock or Foundry; subscription users can use it too, it is not API-key only. The animated video is a stylized mock-up (Sonnet 5.5 builds, Opus 5.5 on call, plus a 'JEV' fork panel from a third-party tool) rather than a real screen recording. Nothing is sold; it is an educational post with hashtags.

What holds up

  • Contact sheet viewed: animated terminal mock-up 'Claude Code advisor tip', Sonnet 5.5 main session, Opus 5.5 advisor on call, Haiku 5.5 subagents, a JEV panel; audio is filler only.
  • Caption translated from Vietnamese and compared point by point with Anthropic's advisor docs; every technical claim matches.
  • Docs confirm experimental status, provider limits (not Bedrock, Google Cloud's Agent Platform or Foundry), subscription availability and uncached full-transcript reads.
  • x-ref: same advisor pattern graded partly in r18 this batch, where extra claims were added.

What doesn’t

  • Video is a stylized mock-up, not a real session recording.
  • The JEV panel in the video is a third-party tool, not part of Claude Code.
  • 'Requires the Anthropic API' is slightly imprecise; Pro and Max subscribers can use it too.

The catch

Accurate and appropriately cautious. The only thing to watch is that the animation shows a third-party JEV layer that Claude Code does not include.

How to actually do it

  • Update Claude Code, then run '/advisor opus' (or start with 'claude --model sonnet --advisor opus') on a long refactor or debugging task.
  • Check '/usage' after the session to see what the advisor calls added, and turn it off with '/advisor off' for short tasks.
  • Keep your normal tests and review; the advisor is a second opinion, not proof the code is correct.

A rare tech explainer that stays inside what the vendor documents and names the costs. The video adds a decorative third-party layer, but the text is reliable.

Confidence
High
Posted by
a Vietnamese web-development and AI page (about 16,000 plays); reel sent to Buddy Sun Nov 15, 2026, 12:25 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.