The Future of AI

In the age of AI, where is software heading?

This presentation by Andrej Karpathy (former head of AI at Tesla) is making the rounds and with good reason. I'm putting this first in this week's newsletter because I think it's the best 40 minutes you could spend this week to understand ideas that are foundational to AI, where software is ultimately heading, the role we have as humans, and how we prepare for the future that's ahead. (Bonus: proof you don't need fancy slides for an effective presentation.)

Andrej Karpathy: Software is changing (again)
Watch Time: 40 minutes

AI and Product Development

How AI is compressing product strategy

In his thoughtful post, David Hoang (Head of Design/AI at Atlassian) captures the impact AI is having on product strategy:

"Traditionally, strategy lived upstream—something you nailed down before building. Now, it moves with the product. Living strategies evolve in real time, informed by what prototypes, user feedback, and data are telling us. Strategy is no longer fixed; it’s dynamic.

...A compressed strategy doesn’t mean the strategy itself is less important. What’s happening is that the noise of the bloated processes, the meetings-about-meetings, the performance masquerading as planning, is what’s getting squeezed out. What remains are the parts of the strategy that matter most. ⁠⁠In some ways, they’ve become even more vital because clarity is the new constraint. When building becomes cheap and fast, judgment, intention, and alignment become the true competitive advantage.⁠⁠"

This post is an excellent complement to the Andrej Karpathy presentation referenced above.

Strategy is compressing
⚡ Quick Read (5 minutes)

Product Announcements

Your browser as AI workspace: Meet Dia

What if your browser became your primary AI workspace? The Browser Company, makers of the Arc browser, is betting that's exactly where we're headed with their AI-first browser, Dia (currently in beta).

The Verge has a great review:

"Within Dia, every tab and window starts with an omnibox. If you type the name of a website, it should just take you there. If you type something that sounds like web search, you should get web search results. And if you ask for something an AI assistant can handle, it should bring up not just the assistant, but the right version of the assistant with the right data and skills required to help you get stuff done.

...if Miller and The Browser Company are right about where AI is headed, your web browser could become much more than just a web browser. It could become the app that is with you everywhere, that knows you best, that can help you with anything."

As noted in the Verge article, Perplexity and OpenAI are also rumored to be developing a browser.

The Verge: Dia is now available in beta
☕ Medium Read (8 minutes)

The Browser Company: Getting started with Dia
Watch time: 4 minutes

AI in the Organization

The secret to getting teams excited about AI? Start with a hackathon

Ethan Mollick recently observed something remarkable at Intuit's Global Engineering Days: 8,500 people working in cross-functional teams for a week on AI projects of their choice, producing 900 demos—some of which are being turned into actual products. But here's the real insight: "just giving employees time to work with AI tools on real projects is a key to adoption, even if a particular project itself doesn't advance."

Ethan Mollick later asked Intuit CTO Alex Balazs and his team to summarize what they learned to help others stage their own hackathons.

Rules and takeaways for holding an AI hackathon
⚡ Quick Read (2 minutes)

AI Foundational Concepts

Understanding RAG vs. MCP

If you've been hearing about RAG and MCP in AI discussions but aren't quite sure what they mean, Vitaly Friedman's latest LinkedIn post succinctly breaks it down.

• Think of RAG (Retrieval-Augmented Generation) as AI's way of accessing additional information to answer your questions—for example, when you upload a document to ChatGPT and it can then answer questions about that specific content.

• While RAG just fetches information, MCP (Model Context Protocol) lets AI actually do things—update your calendar, send emails, or access your Figma files.

For UX designers, this distinction is crucial because MCP represents a fundamental shift toward AI-powered workflows that may entirely bypass traditional interfaces. Instead of users having to click through forms, they could ask AI to "book that meeting and update the project status," with MCP handling the entire process behind the scenes.

MCP: What It Means For UX Designers
⚡ 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