What’s happening in the world of AI:

Highlights

Dario Tells The Pentagon To Shove It

Discussed at [00:36]

Dario Amodei just called the Pentagon’s bluff.

@iamkylebalmer

Anthropic tell the Pentagon to shove it. The fact that an AI company can tell the Department of War NO is telling of the power of these co... See more

After a week of escalating threats, ultimatums, and what looked like capitulation, the CEO of Anthropic has published a statement telling the Department of War that their threats do not change Anthropic's position. The full statement has 6.5 million views and counting. Here's the breakdown.

It opens with patriotism. Dario positions Anthropic as more American than the Pentagon itself. Claude is extensively deployed across the Department of War for intelligence analysis, operational planning, cyber operations, and more. He reminds them that Anthropic was the first frontier AI company to deploy on classified government networks. He points out they voluntarily gave up several hundred million dollars in revenue by cutting off Chinese Communist Party-linked firms. The message is clear: we are the patriots here.

Then the two red lines.

Mass domestic surveillance: Anthropic supports foreign intelligence and counterintelligence. Spy on the Chinese, spy on the Russians, that's fine. But mass surveillance of American citizens is incompatible with democratic values. Dario makes a crucial point about AI's ability to de-anonymise data: powerful AI can now assemble scattered, individually innocuous data into a comprehensible picture of any person's life, automatically and at massive scale. Your "anonymous" data isn't anonymous anymore.

Fully autonomous weapons: This is more nuanced than the headlines suggest. Dario is not saying no to kill bots. He's saying not yet. He explicitly supports partially autonomous weapons like those used in Ukraine. He even says fully autonomous weapons "may prove critical for our national defence." But today's frontier AI systems simply aren't reliable enough. Claude hallucinates. It gets things wrong. When you rig that to a missile, people die. Anthropic offered to work with the Department of War on R&D to improve reliability. The Pentagon refused.

Then comes the line we'll remember: "These threats do not change our position. We cannot in good conscience accede to their request."

He closes by reminding the Pentagon that Claude is essential to their operations and that removing Anthropic would hurt America more than it hurts Anthropic. The ball is in their court.

Kyle's take: The cojones on Dario. He's just told the Department of War to go and shove it. On the day of their deadline. This could be a massive embarrassment for the US government if they back down, or Anthropic could cease to exist by Monday.

Update: Trump got involved and has censured Anthropic. He’s given them an out though in the 6 month wind-down. This feels like the TikTok ban all over again. More in tomorrow’s newsletter.

We need to be clear about what's actually being said here. Dario is not anti-war. He is not saying we will never build killing machines. He's saying we're not against autonomous weapons, we just need to make sure they don't kill Americans. The loss of non-American lives is apparently acceptable. So the future is still kill bots. It's just kill bots with better targeting.

The positioning is masterful. Somebody in the PR department deserves a raise, if Anthropic still exists next week… He's framed it so that the Department of War looks unpatriotic for pushing back on safety that protects American lives. He starts with patriotism, ends with patriotism, and puts the uncomfortable stuff in the middle.

What's been bizarre is how few people in the AI space have been talking about this all week. I've been covering it daily and I've barely seen anyone else touch it. I wonder if American creators with sponsorship deals from companies like Microsoft and Google, companies with their own government ties, find it too sensitive to discuss. I'm in the UK with no sponsorship deals, so I'll praise and talk shit about anyone I want!

Source: Dario Amodei's full statement | Anthropic tweet | BBC coverage

Jack Dorsey Cuts 40% of Block: The First Real AI Layoff?

Discussed at [27:25]

Jack Dorsey, the man who built Twitter, has fired 4,000 people from Block. That's nearly half the company. Block owns Square, Cash App, and Tidal, and they're profitable and growing. This isn't a struggling company trimming fat. This is a healthy business deciding it doesn't need as many humans.

@iamkylebalmer

Jack Dorsey just eliminated 40% of his workforce overnight. Why? AI. Is this the first mass layoff because of AI? #ainews #learnai #jackdorsey

From his public statement: "We're already seeing that the intelligence tools we're creating and using, paired with smaller and flatter teams, are enabling a new way of working which fundamentally changes what it means to build and run a company."

He was explicit about why he did it all at once rather than gradually: repeated rounds of cuts are destructive to morale. He'd rather take the hard, clear action now. The severance is generous by any standard: 20 weeks salary, one week per year of tenure, equity vested through May, six months healthcare, $5,000 transition support, and you keep your devices.

The market's response was telling. Block's stock jumped 25%. Investors didn't see a company in crisis apparently.

The final line of his statement is the one that should make the remaining 6,000 employees nervous: "Our customers will be able to build their own features directly, composed of our capabilities and served through our interfaces." That's the SaaS death spiral in one sentence. Your customers don't need your product if they can build their own….

Kyle's take: Previous AI layoff stories were mostly companies in trouble using AI as cover. Klarna was haemorrhaging money and trying to force an IPO. They rehired a year later and looked silly. Salesforce also dropped staff and regretted it. But they are a legacy tech company who is directly threatened by AI.

This is different. Block is profitable. Revenues are growing. Jack just decided he doesn't need 10,000 people to do the work that 6,000 people plus AI can do.

I can confirm this (purely!) anecdotally. I now do the work of five to ten people. That's how I produce this much content, run a business, do daily live streams, and maintain courses all simultaneously. If you scale that from one person to an organisation, the maths gets uncomfortable very quickly.

Even if the speed and extent is wrong here the direction is VERY clear.

But the real damage won't be dramatic layoffs like this. It'll be the non-creation of new jobs. When the next economic upturn comes, companies that would normally hire won't. They'll augment existing staff with AI or just not open positions at all. Entry-level roles are where this bites first because that's what AI is already good at. The jobs won't be destroyed. They just won't be created. Subtle.

My advice: if you have a job, keep it. Keep getting paid. But on the side, start building contingencies. Build something. Build optionality. Because we do not know what's coming.

Source: Jack Dorsey's statement

 Nano Banana Pro 2 Is Here

Discussed at [47:22]

Google DeepMind has released Nano Banana Pro 2 inside Gemini. The big upgrade is consistency: the model can now maintain the likeness of people, including faces, and up to 14 objects across different frames in a single workflow. Previously, if you were generating multiple images in a conversation, it would forget what it had done. You'd end up with images you couldn't use together. Storyboards, carousels, anything that needed visual continuity required generating everything in one image and cropping. That's no longer the case.

It also does the clever thing the original did: instead of just guessing what your image should look like, it sends your request to the language model first. So if you ask for an infographic about the water cycle, Gemini thinks about what should actually be included, writes its own optimised prompt, then generates the image. That's why it's so good at infographics and text-heavy visuals.

Kyle's take: I chatted to David Sharon, the product manager, a couple of weeks ago on YouTube. He didn't give away any secrets about v2 of course, but here it is. The consistency upgrade is the big one. Being able to maintain character and object consistency across frames is what makes this actually useful for real workflows rather than just one-off image generation. More NanoBanana is always welcome!

Source: Google DeepMind announcement

 Claude's Triple Feature Week: Memory, Remote Control, Scheduled Tasks

Discussed at [50:08]

Despite potentially ceasing to exist by Monday, Anthropic shipped three major features this week. All three look like they studied Open Claw, identified what people actually used it for, and built safer versions into Claude.

Auto Memory (the big one): Claude now remembers what it learns across sessions. Your project context, debugging patterns, preferred approaches, it carries all of that forward without you having to write anything down. This was the only reason Kyle still used ChatGPT. When you've used ChatGPT for years, it knows your businesses, your interests, your preferences. Moving to Claude meant starting fresh every time. No longer. Think of Claude MD as your instructions to Claude and Memory MD as Claude's own scratch pad.

Remote Control (released earlier this week): Start a Claude Code session on your computer, pick it up on your phone. Claude keeps running on your machine. You steer it from anywhere.

Scheduled Tasks (via Cowork): Set up recurring tasks that run automatically. Morning briefings, weekly updates, automated reports. This was 80% of what Open Claw users were actually doing, just cron jobs with a friendly interface.

Kyle's take: Open Claw was a proof of concept that showed the public had enormous appetite for a persistent, proactive assistant you can access anywhere. Every company noticed. Manus released their agent. Perplexity launched Perplexity Computer. MiniMax built Max Claw. And now Anthropic is rebuilding Open Claw inside Claude, piece by piece, but safer. Oh and presumably OpenAI will release their own version of OpenClaw at some point now they have hired Peter Steinberger.=

If you watched my lives weeks ago, I said: don't touch Open Claw if the terminal freaks you out. Wait for the professional versions. Here they are.

Streaming on YouTube (with full 4k screen share) and TikTok (follow and turn on notifications for Live Notification).

Audio Podcast on iTunes and Spotify.