Dear Friend,
After reading yet another headline quoting a leading economist warning that AI will fundamentally reshape the labor market (and yet another rebuttal that we will, as we always have, create new jobs and all will be fine), I thought back to a conversation I just had with my dear friend and radical team member Jeffrey about the fallacy of the official future vs the ever-widening cone of uncertainty: We need to do more serious scenario planning around AI’s impact on jobs. I’m sure that work exists – but most of what I read is the oversimplified headline version of “what if,” with very little “and then what.” The “and then what” is what we all should pay attention to. If I have learned one thing from studying (and living through) disruption, it is that to weather change, you need to be prepared. And you won’t be prepared if you don’t do the work of – well – actually preparing.
And now, this…
Headlines from the Future
Social Media Surveillance: Personal AI agents are all the rage right now. Inside their Tamagotchi-like visuals, they keep track of your calendar, email, and to-dos, and happily lend a hand with your daily chores. But of course, when something is free, it’s not free – it is just that you become the product. Meta’s Muse AI pact with the devil includes Muse building a detailed profile of every person in your life. As security researcher Karan Joshi, who got Muse to hand over its own instructions, put it: “They’re trying to know you like a friend, which is honestly pretty creepy.”
Don’t Trust That “Human”: AI company Tavus says that their AI video model Tavus says its Griffin is the first AI model to pass the video Turing test. Nearly half of all people who talked to Griffin through a Zoom-like video chat were convinced they were talking to a human. As impressive as this is, the misuse potential is huge. You can’t trust that the “person” on the other end of your video call isn’t an AI anymore.
AI-Triggered Bank Run: Apollo’s Chief Economist Torsten Sløk warns that AI agents like Meta’s Muse could trigger an “agentic bank run.” The argument goes like this: A lot of people’s money is in low-yielding checking accounts; AI agents would sweep it into higher-yielding savings accounts and CDs. Banks, which lend out those cheap deposits, lose their funding and their spread. I think this is more hypothetical than real – it requires a lot of trust in your agent, and neither Meta’s Muse nor OpenAI’s Dots can move your money at the moment. But it does point toward a very weird future in which any business built on customer inertia should be on high alert.
The End/Beginning of Software: A single developer, vibe-coding with Claude, is recreating the whole Adobe suite of photo and video editing tools (Photoshop, Illustrator, Premiere Pro, etc.). It’s as ambitious as it is nuts – but only a few months later, he has some apps that (somewhat) work. It aims to be a faithful recreation of the original software, written from scratch in Rust and in a full clean room (i.e., no code is being copied from the original, making this legally mostly fine). If – and it’s a big if – this works, it has huge implications for the future of software: On one hand, it’s amazing that you can point an LLM at an app and get a rough clone; on the other hand, it raises the question of who will pay for software innovation. Adobe won’t be amused.
The Few People Holding up the Internet: Much of the internet runs on open-source software, which is largely created and maintained by volunteers. Aside from the financial asymmetry between those who build these tools and those who benefit (often the world’s largest corporations – and every human on the internet), it comes with a real risk: The moment a volunteer decides to stop working on their project, the ripple effects can be massive. Eleven of the 23 critical projects the piece tracks have only one or two regular maintainers. Take the time zone database, which runs on 4 billion phones and most internet servers – UCLA lecturer Paul Eggert has maintained it in his spare time since 2012, with no grant funding. We (collectively) ought to pay for what has become critical infrastructure – sudo’s maintainer (an app installed on pretty much every server worldwide) just got $61,700 a year after three decades, which shows both that it can be done and how late we are.
What Would Happen If AI Became Really Fast: Here is an interesting thought experiment – what would happen if AI were to become fast. AI today has four limiting factors: model capabilities, cost, context windows (how much data an AI can ingest), and speed. Most of our attention goes to the first three – speed has improved, but gets far less attention. EchoHive imagines AI answering at a million tokens per second (today’s AIs are operating at around 100-200 tokens/sec). With this speed, LLMs could code 40 app drafts or write 1,000 story endings in a single second. And it would take AI only three seconds to review 10,000 job candidates. With this amount of raw output, evaluation will quickly become the ultimate bottleneck – add one minute of testing to those 40 app drafts, and 10,000× faster writing makes the whole job only about 132× faster.
What We Are Reading
OpenAI Just Carpet-Bombed Mathematics AI didn’t just beat mathematicians at their own game; it may have made the game not worth playing anymore. @Jane
Norway Considers Most Drastic Smart Glasses Ban yet Norway is moving to temporarily ban smart glasses in public spaces, from beaches to schools to changing rooms, as privacy concerns over covert recording push governments worldwide toward tighter regulation around these wearables. @Mafe
A Marketer Built a Fake Deodorant Brand. Weeks Later, It Was Being Recommended Over Real Products The near future of brand visibility and product discovery is going to be a wild, messy ride. @Jeffrey
What Will Our Economic Future Look Like? Anthropic’s scenario explorer lets you test your assumptions about AI and see how growth, wages, and jobs could move in very different directions. @Kacee
Our Minds Aren’t Equipped to Handle AI Here is another (science-based) argument that the overconsumption of AI is just not good for us. @Pascal
…and now for some fun!
🌬️ Ceiling fans have become very, very weird. Here is a wonderful deep dive into what is going on and why.
🏙️ This is insane – Mike Tomlin spent years building a hyper-detailed and massive city inside of Minecraft.
🧑💻 Felix Rieseberg made a truly bonkers personal website – and explains how he did it.
What We Do When We’re Not Writing
Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.
↗ Dive into the deep end: Access our complete collection of 3,100+ radical links.