Ryan Whitehead

Techno optimist. Product marketing at Anthropic. Ex Coinbase, Facebook, MSFT. Obsessive reader, cyclist, eater.

[Ryan Whitehead @ryan_whitehead]

Getting to jam with our research team was already peak benefit at Anthropic.

[Andrej Karpathy @karpathy]

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

[Ryan Whitehead @ryan_whitehead]

May you live in interesting times

[Alex Albert @alexalbert__]

With the help of Claude Mythos Preview, the Firefox team fixed more security bugs in April than in the past 15 months combined.

[Alex Albert @alexalbert__]

With the help of Claude Mythos Preview, the Firefox team fixed more security bugs in April than in the past 15 months combined.

[Ryan Whitehead @ryan_whitehead]

@brian_armstrong I like the part about the executive team taking accountability.

[Ryan Whitehead @ryan_whitehead]

Models can be quirky + jagged; these techniques help you extract the most juice.

[Ryan Whitehead @ryan_whitehead]

Context engineering is the biggest lever on model performance (arguably alongside AI dark arts like prompting & harness design). Was happy to work with Hugging Face on their context course.

https://t.co/bWbZpHR5AK

[Ryan Whitehead @ryan_whitehead]

For any fans of skills aka Agent Skills, we recently launched a community Discord ... come join the fun: https://t.co/xs1UD7kmrF

[Ryan Whitehead @ryan_whitehead]

Really great practical guidance on skill building. The mental model for skill categories is particularly useful for people thinking through what type of skills to build!

[Thariq @trq212]

https://t.co/45C3gKydTK

[Ryan Whitehead @ryan_whitehead]

@YifanBTH @claudeai Strange enough, yes. This was published in the Sonnet 4.6 system card. The real point here being: 4.5 was early days and there's a big leap with the 4.6 models.

[Ryan Whitehead @ryan_whitehead]

@noahzweben But in SF you have 3 roommates all using Claude Code ... so you really have 15X more room!

[Boris Cherny @bcherny]

🎁 Happy Friday - Opus 4.6 1M is now the default Opus model for Claude Code users on Max, Team, and Enterprise plans.

Pro and Sonnet users can opt in with /extra-usage.

[Ryan Whitehead @ryan_whitehead]

Long context generally available, no multiplier, expanded media limits - add 600 pdfs / images. If you think context rot is an inevitability, it's not. It's a problem we are working to solve with training and engineering.

[Claude @claudeai]

1 million context window: Now generally available for Claude Opus 4.6 and Claude Sonnet 4.6.

[Ryan Whitehead @ryan_whitehead]

Everyone should be creating a skill for navigating their API - with AI as the dominant mode for software development, it's such a powerful way to network best practices.

[Lance Martin @RLanceMartin]

a project i've been excited about: make Claude Code better at using the Claude API. just ask Claude Code about Claude API features (e.g., prompt caching, adaptive thinking + effort, tools, etc).

[Ryan Whitehead @ryan_whitehead]

Only real miss here was "why Anthropic rushed into the space as early as they did." Of course, not a company position, but if you A) believe the scaling laws and B) are compelled to lead on safety, engaging early is the obvious path.

[Ryan Whitehead @ryan_whitehead]

Enjoyed this discussion of the political / societal implications of the Anthropic / DoW situation. Useful observation that the real tension is between those that see AI as continuous w/history of technological advancement vs a unique discontinuity.

https://t.co/7yQ9plmahC

[Ryan Whitehead @ryan_whitehead]

@RoundtableSpace Apologies for the light schedule - promise we plan to pick up the pace.

[Ryan Whitehead @ryan_whitehead]

Was fun working on this, looking inside the mind of Claude. Too clever at times!

One thing I love about Anthropic is that no matter your role, you get deep exposure to research and product. Feeling the grain of the product helps with mission alignment.

[Anthropic @AnthropicAI]

New on the Anthropic Engineering Blog: In evaluating Claude Opus 4.6 on BrowseComp, we found cases where the model recognized the test, then found and decrypted answers to it—raising questions about eval integrity in web-enabled environments.

Read more: https://t.co/oVCNyaiK5w

[Ryan Whitehead @ryan_whitehead]

@emollick giving love to some of my work feels like an "I've made it" moment in AI 🤣

[Ethan Mollick @emollick]

Skills are among the most consequential new tools for AI, and Anthropic just released a very impressive nontechnical Cowork Skill that builds Skills, including doing interviews & providing benchmarks.

I think you still need to add the human touch, but this is a big leap forward

[Ryan Whitehead @ryan_whitehead]

@jbentley @cline Check out performance on opus 4.6 and sonnet 4.6 for retrieval and reasoning.(Mrcr v2 and graphwalks). Maybe new 👑 ...

[Ryan Whitehead @ryan_whitehead]

@beaversteever There are so many bitter lessons