01What is the Claude Advisor tool and how it works
The Claude Advisor tool lets you run a cheaper model like Sonnet or Haiku as your main worker, then call on Opus as an advisor in the background. Opus gives short plans when the task gets tricky, so you get near-Opus intelligence without paying Opus prices for every step.
Think of it like a new employee calling a senior colleague for help. The junior handles 90% of the work. When they hit a complicated situation, they check in with the senior person for a quick plan, then keep going.
The senior person doesn't take over—they just point the way.
In this episode, Julian Goldie explains that the Advisor tool removes the old choice between paying for the smart model or settling for the cheap one. You get the best of both.
02Main features of the Advisor tool
The Advisor tool packs several features that make it useful for business owners running AI automations.
- Automatic escalation: Sonnet decides when to call Opus. You don't have to micromanage it.
- Full context reading: Opus sees every message, tool result, and decision so far before giving guidance.
- Short plans: Opus generates only 400 to 700 text tokens of advice. The executor model does the full output at its lower rate.
- Single API request: The whole advisor interaction happens in one request, so no extra round trips or context management.
- Cost controls: You can cap how many times Opus gets called in a session. The agent uses its best judgment on when calls are worth it.
- Caching system: In long agent loops, the system caches the advisor's context. Repeat calls cost less, starting from the third advisor call in a conversation.
These features make it easy to keep costs low while improving reliability on complex tasks.
03How to set up the Advisor tool
Getting started with the Advisor tool is simple, even if you're not a coding expert. Julian Goldie walks through the setup in the episode.
- Make sure you have access to the Claude Platform API. The Advisor tool is live and in beta there.
- Add the Advisor tool to your tools array in the API. It's a one-line change.
- Include the beta header in your API request. This tells Anthropic you're using the beta version.
- Choose your executor model—Sonnet or Haiku. This is the model that does the main work.
- Set a max limit for how many times Opus can be called per session if you want to control costs.
- Test your workflow. The model handles the rest—it escalates to Opus on its own when needed.
You don't need to redesign your whole setup. The Advisor becomes available with that one line, and you stay in control.
04How to use the Advisor tool in your business
The Advisor tool shines when you have complex, multi-step tasks where decisions build on each other. Julian shares examples from real business scenarios.
For a content agency: Claude builds client blog posts, pulls research, formats outlines, and writes drafts. Sonnet handles straightforward parts. But when the brief is vague, the brand voice is tricky, or the research conflicts, Sonnet pauses and calls Opus.
Opus reads everything and says, "Here's how you handle this." Sonnet continues with better direction. The client gets a better draft, and you fix fewer problems.
For outreach automation: An agent researches leads, writes personalized first lines, and drafts emails. Most leads are simple, but some need nuanced handling. The Advisor tool helps the agent make smarter decisions on those edge cases without you babysitting it.
For e-commerce: Opus in the background catches weird cases like returns requests that need careful tone or products with complicated specs that need accurate framing.
- Content creation and client work
- Lead generation and outreach
- Customer support workflows
- Research tasks
The tool takes the unreliability out of complex agent tasks, which is where real time savings come from.
05Results and benchmarks from the episode
Anthropic's own evaluations show measurable improvements. In the episode, Julian highlights key numbers.
- Sonnet with Opus advisor: 2.7 percentage point improvement on a key coding benchmark.
- Cost reduction: Costs per agent task dropped by 11.9% while using the Advisor tool.
- Haiku with Opus advisor: On one benchmark, it scored more than double its solo score.
- Haiku cost advantage: It cost 85% less per task than Sonnet alone.
These results aren't a coincidence. The design is intentional: Opus only generates a short plan of 400 to 700 tokens, while the executor handles the full output at its lower rate. So overall costs stay well below running Opus from start to finish.
If you're running high-volume automations with lots of small tasks, Haiku plus Opus advisor could be the most cost-effective setup available right now.
06Limits, mistakes, and when to skip the Advisor tool
The Advisor tool is not for everything. Julian gives an honest reality check.
It's a weaker fit for single-turn questions and simple tasks. If every turn requires the advisor's full capability, adding Opus as an advisor won't help. A one-question prompt doesn't need an advisor at all.
For simple automations—grabbing data, summarizing it, posting it—your existing Sonnet setup is fine.
The value comes on complex, multi-step tasks where decisions compound. Examples include handling exceptions, responding to varied input, and producing client-ready work.
- Don't use it for trivial tasks.
- Don't expect it to fix a badly designed workflow.
- Be ready for beta quirks—updates are coming.
The tool is in beta on the Claude Platform API. It's live, but as with all betas, it may be buggy at first. Keep that in mind when building critical automations.
07Who should use the Advisor tool and why it matters
The Advisor tool matters for anyone running AI agents to automate client work, lead generation, content pipelines, outreach, customer support, or research. If you've had cheaper models drift, miss context, or make bad decisions mid-task, this tool addresses that directly.
It's especially valuable for business owners who want to scale without hiring more staff. Long horizon tasks—like researching, writing, reviewing, formatting, and sending—become affordable because Opus consults at key moments instead of running the whole way. Affordable tasks get run more often, which means more automation and more done without you touching it.
Julian sees a broader signal too: Anthropic wants to be the place you build your AI workflows. They recently launched Claude Managed Agents and moved to charge separately for third-party tools. The Advisor tool fits into that ecosystem.
Business owners who build these workflows now, while most people are still manual, will have a real edge.
If you're running high-volume automations on a tight budget, test the Haiku plus Opus advisor setup. Double the intelligence for a fraction of the cost is a hard combination to ignore.
