Creating your own replacement

Welcome to your weekly immersion into the cutting edge of AI.

The first month and a half of my year has been characterized by spending plenty of time coding – hacking away at ways of solving my own problems. I have noticed a significant uptick in people vibe coding on Twitter and Reddit, and it’s easy to understand why. Building apps, to me, has become the single most leveraged activity I can think of. Knowing what can be built, and how easy it can be, is mind-bending and irreversible.

For those wanting to dive deeper into AI coding, I've recently created a dedicated group in our WhatsApp community aptly named Coding Club. Join us to learn more and share what you’re working on. I am also hosting a first AI Coding Club meetup later this week on Zoom for those wanting to strengthen their skills and discuss specifics. Would love to see you there.

Finally, the WhatsApp group now has a Foresight Exchange, a space for innovation experts everywhere to trade ideas and research. We are actively building tools in this space - especially around generative approaches to research - so make sure you’re in the group to try our new tools soon.

Until next week,
MZ

---

Code is the primary use case for Claude

Unsurprising learnings from Anthropic.

Today, usage is concentrated in software development and technical writing tasks. Over one-third of occupations (roughly 36%) see AI use in at least a quarter of their associated tasks, while approximately 4% of occupations use it across three-quarters of their associated tasks.

---

AI Action Summit 2025

Yann LeCun on world models and the limitations of LLMs.

---

AI Boyfriends Trending

From one of my favorite newsletters, Exploding Topics.

---

Acts Not Facts (35 min)

Brilliant presentation by the always interesting Matt Webb from last year's TH/NGS here in Amsterdam, all about agents and building apps with AI. Great mix of hands-on experience with big picture thinking. With our very own Monique van Dusseldorp hosting the session.

Matt is always super interesting indeed - his newsletter is always full of new thoughts and ideas. Here the last one, on ''Extending AI chat with Model Context Protocol (and why it matters)'' - which may interest some of you.

---

Future Agents (30 min)

Spectacular Q&A from the WSJ with Bret Taylor, chairman at OpenAI. They focus on agents but cover much of the buzz that's on the agenda: Musk, China, etc.

In 2025, I think every company will have an AI agent with its brand at the top—just like having a website in 1995 or a mobile app a decade ago.

---

Building Effective AI Agents (20 min)

Great continuation video from Anthropic for their Building Effective Agents blog post a few weeks ago.

An agent is where you’re letting the LLM decide how many times to run—continuing to loop until it finds a resolution.

---

Language models are obviously not good at math, but here is an interesting representation of exactly how good or bad they are at multiplication.

---

You are receiving this newsletter because you signed up on envisioning.io or Substack.