Fontier Models

Claude Opus 4.5: Better at messy, real-world work

Anthropic released Claude Opus 4.5, its new flagship model, alongside a fix for a long-standing frustration with Claude: chats suddenly ending when they get too long. In Claude, long conversations now use an “infinite chat” mode—rather than hitting a hard stop at the context-window limit, Claude automatically summarizes earlier parts of the conversation so you can keep going. On the developer side, API pricing for Opus dropped sharply (from $15/$75 to $5/$25 per million input/output tokens).

What stands out from early testing is Opus 4.5’s ability to handle messy, real-world tasks that other models struggle with. In one example from Nate Jones ($), a Christmas-tree business owner used it to reconcile handwritten shipping manifests with receipt sheets. This task involved OCR of pencil tally marks, multiple conflicting data sources, and calculations across hundreds of items. Jones tested the same workflow across Gemini 3, ChatGPT 5.1 Pro, Grok 4.1, and Opus 4.5. Opus 4.5 was the only model that produced a result close enough to use in practice, explicitly flagging discrepancies instead of forcing them into a false sense of consistency. His takeaway: Opus 4.5 is the model to “hire” when the job is specific, but the information is messy, which describes a lot of real work.

Ars Technica | Anthropic introduces cheaper, more powerful, more efficient Opus 4.5 model
⚡ Quick Read (3 minutes)

Nate Jones | I Tested Opus 4.5 Early ($)
Watch Time: 15 minutes

Designing for AI

Emily Campbell on designing AI experiences

Emily Campbell, VP of Design at HackerRank and creator of the Shape of AI pattern library, joined the Deep Dive podcast to discuss how designing for AI fundamentally differs from traditional software. She maintains that designers are no longer just building interfaces—they're creating a meeting place between humans and something synthetic.

One example she shares is Cofounder.ai, which during onboarding, immediately seeks to understand you by connecting to your email and generating a sample message in your tone and voice. This mirrors how you'd build trust with a human assistant: show me some of your work first, then I'll give you more responsibility.

In her Shape of AI pattern library, Emily organizes emerging patterns into categories such as "wayfinders" (helping users get started), "tuners" (prompt enhancers that help refine input), and "governors" (trust-building patterns, such as showing the AI's thought process). This pattern library is an invaluable resource for anyone designing AI experiences.

Dive Club Podcast | Emily Campbell - AI UX deep dive
Watch Time: 43 minutes

AI in the Organization

The rise of the "equipped employee"

Cody Schneider argues that the most effective employees in an AI-driven future won’t just use off-the-shelf tools: they’ll build their own “stack” of agents and personal software.

Cody’s approach: treat every repetitive action as a “bug” in your own system, then fix it by creating automations, research agents, monitoring systems, and custom interfaces tailored to your job. Over time, that stack stops feeling like tools and starts feeling like “infrastructure”—a personal backend that removes friction from your work. The open question: as this becomes easier to do, will companies start expecting people to show up with their own personal backend already in their toolkit?

Cody Schneider | The most effective employees will have custom agents and personal software they bring to their jobs
⚡ Quick Read (2 minutes)

AI and Product Development

AI and the reshaping of product team roles

In a recent post, Felix Haas and Mathias Klenk offer a clear-eyed take on how AI is changing the way product development is actually done, with AI tools like Cursor, Lovable, and v0 blurring traditional role boundaries. Designers can now move beyond mockups to generate and test working code without waiting for handoffs. Engineers, with AI handling more routine coding, are shifting their value toward product intuition and design thinking. The result is people working across disciplines and using AI to ship features from idea to execution.

The piece also tackles what this means for product managers: not extinction, but evolution. When everyone can build faster, PMs become less about coordinating handoffs and more about strategy, synthesis, and making sure the team is solving the right problems. It's a useful framework for understanding why the old assembly-line model of "design → handoff → implementation" is breaking down—and what is replacing it.

Felix Haas and Mathias Klenk | Everyone is a builder now
⚡ Quick Read (2 minutes)

That’s it for this week.

Thanks for reading, and see you next Wednesday with more curated AI/UX news and insights. 👋

All the best, Heidi