Anthropic's new expertise test

Thanks for reading my monthly newsletter for leading in the era of AI. If you want to read more from me, I share ideas, frameworks, and more twice weekly on my Substack here.

Anthropic's new 'Computer use' capability

Yesterday, Anthropic's announced their new 'Computer use' capability, and the internet is buzzing with hype about it. In case you missed it, here's one of the demo videos:

How this is the perfect test for expertise:

To any expert in AI, automation, and software engineering, this capability is an inefficient, error-prone, less secure way to use AI for automation.

The demo for copy-pasting data (a second video you can watch here) is less efficient and less reliable than SharePoint workflows we experts built back in 2014.

To anyone who is not an expert in AI, automation, and software engineering, this is the first time they're seeing AI + automation visualized in a way they can instantly understand, so they believe this is game-changing.

Automation normally requires layers of abstraction to understand, which most people don't have the patience or bandwidth to slow down and think through (and just as often, it's difficult for automation and AI experts to explain these concepts well).

Why it's a brilliant move by Anthropic

1. It leverages mob mentality. There's a prevailing narrative to "Keep up with AI or be left behind"—so by demoing this capability broadly, anyone who displays skepticism will be labelled a luddite or "not with it."
2. It scales Reinforcement Learning from Human Feedback (RLHF). When users interact with this capability, they are supervising the model's work and providing real-time feedback, training the model to improve over time. This means customers will be paying Anthropic to train its model on millions of tasks simultaneously.

Six questions that hint at why I am bearish on this as the form factor for AI + automation

1. What will the compute cost be if millions of automated workflows are created on the fly by non-experts?
2. What happens if the model hallucinates on a workflow involving something important?
3. How does this scale across an enterprise?
4. How will IT ensure safe use?
5. Why wouldn't we want to use natural language to infer intent and then leverage APIs to achieve more reliable, secure outcomes?

Why I'm so excited about this update

No other update from the AI pure plays has provided this subtle of an opportunity to see if the people you're following online or paying to advise you actually know what they're talking about. In other words, are they your AI Indiana Jones ready to lead you into the jungle or just someone in costume?

If you want to assess their credibility, just ask them their take on the new Anthropic Computer use capability.

If they say it's game-changing, you've seen what you need to see.

Reach out if you need recommendations for real experts.

Thanks for reading,

Brian

Whenever you're ready, here are 3 ways I can help you:

1. Keynote Speaking: I've briefed dozens of Fortune 500 C-suite executive teams, spoken to live, in-person audiences of more than 10,000 attendees, am a guest lecturer at Kellogg School of Management, and led panels of distinguished guests ranging from academia to public sector leaders to Fortune 500 C-suite executives.

2. Future Solving Advising: Join hundreds FORTUNE 500 C-level executives and startup founders who have leveraged my advisement on AI, the future of technology, and how to position yourself for the future in the era of AI.

3. Future Solving Workshops: Join 25+ of the FORTUNE 500 and NASA, who have positioned themselves for the future in the era of AI by leveraging new frameworks from the Future Solving Method I introduced in my book, Autonomous Transformation, to set a vision and strategy and spark action.