AI Learns. We Forget.

Voices stolen at scale, employees monitored to train models, AI agents deleting production databases – and somewhere underneath it all, the collective pace of human thought is quietly slowing down.

Dear Friend,

A quick personal update before we dive in: OUTLEARN launched on Tuesday and – I’ll be honest – the response has been beyond what I expected. The book hit #1 New Release in Amazon’s Strategy & Competition category on launch day and is currently sitting at #2 on the bestseller list in that same category.

But what’s even more awesome is your feedback. One early reader wrote that he bought three copies on launch day to hand out to colleagues at a startup they’re building together. Another described it as a “fully actionable page-turner” and said the reframing of postmortems as harvest meetings would – his words – “turbocharge how you extract practicable learnings from any project outcome.”

That’s exactly what I was hoping this book would do. Not sit on a shelf. Get used on a Monday morning.

If you haven’t grabbed a copy yet: Get it here. It’s 150 pages, every chapter ends with something deployable, and it’s cheaper than lunch. And if you’ve already read it – an honest Amazon review in these first two weeks genuinely makes the difference between a book that reaches people and one that disappears.

And now, on to our usual programming…

Here is an interesting argument: “AI doesn’t really ’think.’ Rather, it remembers how we thought together. And we’re about to stop giving it anything worth remembering.” This is from a provocative article by Bright Simons. I might not fully buy into all aspects of his argument, but his essay is very well worth reading (and pondering over). Let me leave you with just one more quote from the article: “The result is a world in which individual productivity rises while the collective pace of human thought starts to fall.”

Read the thing. And then, read this…

---

Headlines from the Future

Why (Conventional) Software Truly Is under Attack. Friend of radical Martin Alderson is back with a deep analysis of why many conventional software systems are under attack by AI – exemplified by the example of Figma, the once-darling Adobe-killer (and failed acquisition target of said company). Anthropic launched their new Design tool, automating 80% of what Figma does – as part of their Claude app, no design skills required.

But the structural point is harder to wriggle out of. Figma has ~2,000 employees. Anthropic has ~2,500 total and I doubt Claude Design took more than a handful to build. Figma now needs to out-execute a competitor whose inference is ~free to them, whose marginal cost to ship is roughly zero, and who employs fewer people on the competing product than Figma has on a single pod. That’s a very hard position to pivot out of. This feels like a preview of where SaaS economics are heading. The companies that built big orgs on the assumption of steady seat expansion are going to find themselves competing with products built by tiny teams inside the frontier labs. Figma just happens to be the first big public name where one of their primary inference suppliers has started competing against them

━━━━━

Software Brain is Eating The World. If you do not read anything else this week, do yourself a favor and read this article by Nilay Patel on “Software Brain.” It’s a thoughtful piece about the disconnect between what the makers of AI think they are building, and what many of us experiences.

The entire human experience cannot be captured in a database. That’s the limit of software brain. That’s why people hate AI. It flattens them.

━━━━━

Who Watches the Watchmen? This famous question, first posed by the Roman poet Juvenal in his Satires, dating back to the 1st century AD, is increasingly answered by: As long as you are feeding the machine, nobody… Meta, reportedly, is running surveillance software on work PCs of their employees:

Newswire Reuters reports that Meta management sent staff a memo informing them that they’ll soon run a new tool called “Model Capability Initiative” that will record their keystrokes, mouse movements, and even take occasional screenshots – all in the name of gathering data the social networking giant can use to build better AI models.

The staff doesn’t seem to be too happy about it: “Meta employees are up in arms over a mandatory program to train AI on their mouse movements and keystrokes.”

━━━━━

4TB of Voice Samples Were Just Stolen from 40,000 AI Contractors. Voice cloning has become incredibly easy, good, and, in some cases, useful. Advanced AI models only need a few seconds of your voice to create a convincing clone – which also means that with those few seconds of your voice, I can create a convincing spoof of you. Which, in turn, means that when your voice data is stolen, you might be deep in the s#!%.

On April 4, 2026, the extortion group Lapsus$ posted Mercor on its leak site. The dump is reported at roughly four terabytes and bundles a payload that breach analysts have been warning about for two years: voice biometrics paired with the same person’s government-issued identity document. According to the leaked sample index, the archive covers more than 40,000 contractors who signed up to label data, record reading passages, and run through verification calls for AI training.

---

What We Are Reading

AI’s Next Frontier: People Skills AI models now beat humans on emotional intelligence tests, which only goes to prove that acing a test and understanding a feeling are completely different things. @Jane

Paypal Says AI Shopping Agents Are Creating an Invisible Storefront Economy Nearly all merchants are already seeing AI agent traffic, but fewer than 25% have the machine-readable catalogs, APIs, or the agent-compatible checkout systems needed to act in real time and convert on the spot. @Mafe

Scarce Assets: The Abundance-Driven Scarcity Supercycle Sticking with my theme from last week: Another angle on the idea (and value) of finding what remains scarce in markets where some things seem newly and wildly abundant. @Jeffrey

Chatgpt Traffic Analysis: Insights from 17 Months of Clickstream Data Most people assume AI search = more discovery, but this data suggests the opposite: that in fact distribution is compressing. If you’re not in the answer set, you’re effectively invisible. @Kacee

When the Internet Was a Place If you are fortunate enough to have experienced the Internet pre-Web 2.0, you know that it was a very, very different place. It’s high time to claim back some of those properties. @Pascal

---

Down the Rabbit Hole

🤔 Finally want to understand how LLMs actually work? Here is a wonderfully designed and easy-to-follow primer.

🤑 We mentioned it in the briefing last week. The era of heavily subsidized AI models might come to an end quicker than many of us thought or had hoped for. GitHub Copilot just announced that they will be moving solely to a usage-based billing model. Meanwhile, tokens get more expensive, and AI agents can now cost as much (and more) than human workers. No more free beer!

🏬 Someone set up a store and let an AI agent run it. Here is the story of how it’s going (hint: not great, but also not a complete disaster).

🧑‍🏫 A Catholic scholar argues that GenAI threatens authentic education by replacing the process of learning with the production of polished output, emphasizing the need for pedagogical redesign to restore the formation of thoughtful, responsible individuals: Repairing the ruins – Why AI can’t replace education.

💭 Your friendly public service announcement: A.I. should elevate your thinking, not replace it.

🕴️ Somehow I am genuinely surprised that this wasn’t the case before (and how this is news to begin with): Accenture to roll out Copilot to all 743,000 employees in boost for Microsoft

🙀 This story is making the rounds at the moment – an AI agent goes rogue, deletes a company’s entire production database, and then apologizes for it. The deeper cut is that it’s not just the AI agents fault, but the database system itself didn’t have any safeguards in place to prevent this from happening in the first place.

📰 The (sad) future of journalism – an OpenAI-linked news outlet appears to be entirely AI-generated. And the bigger picture: AI swarms could hijack democracy without anyone noticing.

🥳 XOXO, the Portland-based answer to SxSW, is no more. But its legacy lives on in the form of this wonderful (and wonderful-looking) XOXOFest website.

👻 Now we (finally) know: Spooky feelings in old houses may be caused by boiler sounds, study suggests.

↗ Dive into the deep end: Access our complete collection of 2,700+ radical links.

---

Should We Work Together?

Hi! I’m Pascal from radical. This newsletter is our labor of love. When we’re not writing, we run radical, a firm that helps organizations navigate the future without the “innovation theater.” Most leaders want to seize new opportunities, but they hate endless strategy decks that go nowhere. At radical, we don’t run “projects”; we build your organization’s internal capacity to handle disruption and change. Our goal is to make you future-proof so you can stop reacting to the world and start shaping it. If you’re interested, let’s jump on a call to see if we’re a good fit. Click here to speak with us.