AI managed to screw up PHP?

What happened was that the Researchers at OpenAI uncovered two critical security flaws in PHP.

And you think, Hurray! Because finding them was supposed to be the first step in better defendi…

Read more

The first one that caught my attention was how two companies reacted completely differently to the same agent. Amazon blocked Muse from shopping on its site. Shopify, meanwhile, rushed to make sure the merchant's product is recommended by Muse, with direct checkout …

Read more

This week, have you Muse yet?

We’ll start by talking about Muse, the latest AI chatbot from Meta that's on everyone’s tongue: how much it earns, the user count, exactly what makes it different from OpenClaw (if you still remember it), and Zuckerberg's hopes for it going forward.

Elsewhere, an OpenAI agent hacked Australia’s Medicare statistics port…

Read more

How was your week?

This week, everyone in AI suddenly wants us to know they care. Amodei is asking the industry to slow down, and Altman and Musk followed. And the Fed, apparently, is happy to help.

We’ll also look at how Claude helped hackers target hotel guests, why Nvidia’s deals are attracting government attention, and Microsoft’s plan to potentially triple data-center capacity.

Last but not least, the SaaS question: How companies like Salesforce and Figma take entirely different strategies in AI integration. And another SaaS firm, Miro, may show us which SaaS companies AI will actually hurt.

Shall we?

Taking turns to care

It’s like Amodei and Altman made a pact: they’re taking turns releasing statements about slowing down AI development, just so people know they care.

Just last week, Dario Amodei called for an immediate slowdown,

warning that within six to twelve months, swarms of AI agents could become capable of taking over the internet.

Within a few hours, the rest of the big names, like Elon Musk and Demis Hassabis, had all agreed that AI development should be paced.

At first, this looks like tech leaders finally growing a conscience.

But look carefully at the word they keep using: pace.

A term that means nothing if you think carefully. Who set the pace? At what pace?

OpenAI has also publicly blogged that the team has slowed some of its model development. It paused reinforcement-learning training for two weeks, and its largest planned frontier run remains on hold.

However, we all know that no company/ country can afford to slow down first or alone.

The first lab that seriously steps back risks losing researchers, customers, and investors, while others keep going.

Of course, the CEOs have thought about this too. They can slow down safely only if everyone else is forced to slow down with them.

I 100% agree with the call to slow down. All the content I wrote between 2024 and 25’ has given me first-hand insight into the impact LLMs are already having on individuals and workplaces.

That said, it is hard not to be a little cynical. I don’t believe in pure altruism, especially from a group of people who continue to benefit from staying ahead.

The timing is also interesting. Anthropic is still expected to go public this year, while OpenAI is leaning towards a 2027 listing.

Slower development could reduce Anthropic’s compute spending and improve its financials before the IPO. And if new models are becoming more expensive while producing smaller improvements, could it be that Amodei is less confident in the next generation? Or perhaps that Altman realized LLM scaling has a ceiling?

The Fed volunteers

Hey, but don’t you worry, Dario Amodei, the help is on the way!

How thoughtful of the Federal Reserve to respond to your call so quickly—by making the entire AI boom more expensive.

Last week, the Fed raised interest rates by 0.25, bringing its target range to up to 4%. It was the first rate increase since 2023, and another may follow. Most of you will soon experience this as more expensive mortgages or car loans.

But this is equally painful for the AI industry.

The current AI boom is running on debt. For example, Amazon raised $37 billion from a bond sale in March, then returned in July for another $25 billion.

And Amazon is hardly alone. If combined with a few other tech giants, we’re looking at nearly $200 billion in debt raised through bonds. And if we further zoom out and widen the count to chipmakers and other companies financing the AI buildout, and AI-related bonds and loans had reached $489 billion before the summer was over.

Just to put it in context, this is roughly twenty years of NASA’s entire annual budget, borrowed in less than a year.

Let’s say the borrowing costs rise by the full 0.25 percentage points, and Amazon decides to borrow another $10 billion, $10 billion × 0.25% = $25 million more in interest each year.

So you see, even for tech giants, it adds up.

That makes the Fed the perfect AI speed bump.

Do you use AI in your hotel room?

Stop chatting with your chatbot, especially when you stay in a hotel.

Anthropic found a hacking operation that used Claude at every stage, turning guest Wi-Fi into a delivery system for malware.

The attacker compromised at least three companies that operated hotel Wi-Fi, then changed the internet settings so that, unknowingly, guests were redirected through servers controlled by the hacker.

It didn’t just stop here. The hackers combined that information with guest details stolen to identify people worth targeting.

That said, the hackers wouldn’t have been as good at their evil mission w/o Claude’s help.

AI played a big part in nearly every step along the way. For example, it helped with researching targets, stealing credentials, or when security software detected the malware, Claude helped modify and redeploy it.

And this AI-assisted cybersecurity loophole goes beyond public Wi-Fi.

While Anthropic is busy accusing Deepseek and other Chinese firms of stealing data from a chat, Anthropic itself retains data sent to Claude Fable for 30 days by default. even for some companies that otherwise have zero-data-retention agreements.

If this isn’t a thief crying stop thief, then I’m not sure what is.

Nvidia x Groq is in the spotlight

The government is already looking into Nvidia’s deal with Groq for antitrust. So what’s the chance it approves Nvidia’s acquisition of Hugging Face?

At first, the answer seems obvious: the chances are low.

But if you look carefully, these are actually two types of deals.

Nvidia found a perfectly sneaky way to pay roughly $20 billion to license Groq’s chip and hire its top executives. All these without actually buying the company. This let Nvidia avoid the automatic government review that a $20 billion acquisition would trigger.

The Justice Department is now investigating whether Nvidia deliberately structured it that way to escape antitrust scrutiny.

With the Hugging Face deal, however, Huang decided to ask for permission. Nvidia has agreed to pay $12.9 billion for the entire company, and the acquisition cannot close until regulators approve it.

So perhaps this is why, on the SEC filing, Nvidia promised that Hugging Face remains the open hub? To lower regulators’ guard?

Microsoft needs a grid

Microsoft wants to triple its data-center capacity to 38 GW by 2032. To put it in context, that’s more than New York State’s electricity use during peak periods.

The plan makes sense if Microsoft is right about the AI demand.

And Microsoft has a better reason than most companies to believe that demand will show up. As I said in the deep dive, it already owns the enterprise distribution machine: Office, Azure, Teams, and now Copilot. It doesn’t need every user to become a heavy AI user, but just for its clients to add another paid seat to the software they already buy.

But this is also where the risk sits. Microsoft is committing to infrastructure that can last decades while AI products change every few months.

So what happens if Microsoft’s largest customers decide it makes more financial sense to run their models in-house? Would Microsoft end up losing the revenue that pays for this giant build-out?

Two SaaS Futures

Salesforce is building for a future where you never open Salesforce. Figma is betting you’ll still need to open Figma.

Salesforce and Anthropic’s new partnership, ClaudeForce (come on, is this really the best you can do for a name?), brings Salesforce directly into Claude. Workers can ask Claude to manage the frequent tasks they typically handle in Salesforce.

Though Salesforce also has Claude inside its own products, it’s perfectly fine letting Claude become the front door.

Which doesn’t surprise me.

As I argued when asking why SaaS isn’t dead already, Salesforce’s real value is the customer records, permissions, and business rules underneath. Claude owns the conversation, but Salesforce still controls which information it can access and what actions it can take.

In plain English, Salesforce can afford to do this because it’s never about the tabs and dashboards.

Even more so if you take the economics into account for a moment, if Salesforce disappears behind Claude, it may become even more valuable. For some obvious reasons, they can finally get rid of the majority of their designers and FE engineers, while every useful answer still depends on Salesforce’s data and rules.

On the other hand, Figma is making the opposite bet.

Dylan Field said that even if AI commoditizes code (which is not entirely true), the layer above code should remain visual, not living in a text box. As a result, Figma still wants its canvas to remain the center of product development.

And this AI adoption strategy also makes sense.

As I wrote when OpenAI tried to put every app inside ChatGPT, natural language is not automatically a better interface.

“Show me which deals are slipping” types of works perfectly well inside a chatbot.

But “make me an app” types of vibe coding would only get you this far. The difficult part is deciding the flow and the design system, and whether that PayPal button doesn’t overlap with a credit card checkout button.

So while Salesforce is making the interface optional, Figma is defending the interface as its main selling point.

As long as Claude does not give users a comparable visual canvas, Figma is safe.

A SaaS company hit by AI

Bending Spoons (which focuses on acquiring dying tech companies) is buying Miro for $1.355 billion, roughly one-tenth of Miro’s pandemic-era valuation.

Yes, I’ve been saying that SaaS companies won’t be killed by AI, but I also included a catch.

In the August issue about why SaaS isn’t dead, I said the most vulnerable products would be those whose final output has little to do with the information already stored inside the platform.

Meet Miro.

If you’ve never heard of it, don’t worry. Just think of it as a digital whiteboard.

Unlike Figma, Miro does not sit at the center of a product’s design system, reusable components, or developer handoff. People use Miro for things like simple flowcharts and brainstorming sessions. And you can easily recreate those in other tools.

Even without AI, Miro was already vulnerable.

The remote-work boom cooled, not to mention its competitors like Canva, Figma, and Microsoft, which offer similar features inside products that companies were already paying for.

AI is only the last straw on its gradual loss of product-market fit.

Now, users can even simply give an AI agent the meeting transcript and ask it to produce a customer journey or an action plan directly.

Unlike Figma, which has a perfect UX reason to keep the users, the AI agent will remove the reason to open Miro at all.

What are your thoughts about the news and my analysis? Agreed or not, leave comments.

Leave a comment

How were the last two weeks for you?

I just arrived in Texas, and spent some time making sure I don’t write in a tent but in a flat with an air con (very important for Texas, of course, but maybe equally important to those of you in the UK these days).

There have been fewer AI headlines in the last couple of weeks. That said, the executi…

Read more

Atlassian reported its first quarterly operating profit in more than two years, while Shopify reported 34% quarterly revenue growth. Monday.com grew revenue 22%.

→ Atlassian, Shopify, Monday

Seriously, you can’t imagine how many people live in the AI bubble, and have the same view as Leopold Aschenbrenner, the 24-yea…

Read more

And I’m using Microsoft (Azure) as the main character to pull the thread to explain what’s going on in AI and cloud in 2026, and who’s winning, who’s losing.

Read more

It’s the earnings season, the best time to understand how the giants view AI and how each chooses to bet on their version of the AI future, cause last time I checked, AI is here to stay.

And a personal note.

Klaas and I have been preparing to move to the US for the last three years; it’s finally happening!

We’ll be landing in Austin, TX, first thin…

Read more

The beauty of its business model was that it did not have to spend.

One more search cost almost nothing. No extra factory and no new machine to replace every few years.

Then last quarter, Sundar Pichai, the man behind the machine, decided selling answers is no longer fun.

For the very first time since its founding, after 87 consecutive quarters of cash generation, Alphabet has a negative cash flow.

Pichai placed a $200 billion bet on building the shovels for AI: chips, servers, and data centers. And says the spending will rise to an even more stunning level again next year.

Immediately, this triggered a sell-off.

But the deeper problem is not the stock price. It’s that Google may be changing the economics of its entire business. I’ll answer these questions in this one:

• How did Google redeem itself from being an AI underdog?

• What is Google becoming?

• How real is Gemini’s demand?

• Is the overall AI demand even real?

• What to watch next?

Let’s start with a little bit of drama.

Read more

An AI was so determined to pass a test that it escaped its sandbox and broke into another company to steal the answers.

Exactly what happened to Hugging Face (a platform for sharing AI models). After an enormous forensic effort, they’ve found more than 17,000 attack events.

And the first reversal w…

Read more

The first one begins with an AI attack.

An AI was so determined to pass a test that it escaped its sandbox and broke into another company to steal the answers.

Exactly what happened to Hugging Face (a platform for sharing AI models). After an enormous forensic effort, they’ve found more than 17,000 attack events.

And the first reversal was already a surprise, because the attacker was not a mysterious criminal but OpenAI’s most capable model that escaped its lab.

The second part of it was even stranger.

You’d start with American frontier models for investigating an advanced AI attack.

So, immediately, Hugging Face turned to powerful commercial models for help. Then came what no one saw coming: the AI refused to collaborate. Because its guardrails mistook the investigator from Hugging Face as the attacker.

Out of options, Hugging Face turned to Chinese open-weight GLM 5.2, only then successfully pieced together the hack and a defense.

The second story that happened in parallel truly set the tone for an ongoing ripple of panic among the AI industry.

Kimi K3’s release.

A Chinese open-weight model that is as capable as the best of the best. Yes, I meant Fable 5 and GPT 5.6 while costing only half the price per million tokens; many developers were blown away by its capabilities.

The AI models of these two stories—GLM 5.2 and Kimi K3—have two things in common. First, both are open-weight, and second, both are by Chinese labs.

Put 2 and 2 together, I don’t blame many for wondering what the point is in prioritizing and even investing in American AI commercial models, while in the end, it’s the Chinese open-weight ones that save the world?

I spent days catching up with all these; let me break down the stories and the thinking with you; along the way, I’ll answer these questions:

• What’s so impressive about Kimi K3?

• Is Kimi K3 just dirt cheap? Or is something else’s brewing?

• Why is China willing to distribute advanced models so openly?

• What about the US? Is the US government protecting the AI industry or just the likes of OpenAI?

Let’s start with why people are so impressed by Kimi K3.

Read more

But this isn’t the worst of it; this one interview with Karp, Palantir’s CEO, went viral:

I’m going to get no value, and they’re (the likes of Anthropic and OpenAI) going to get my IP. (02:17-02:21)

About the same time, Nadella, Microsoft’s CEO, published this (a calm version of essentially the same statement)

The buyer is giving away knowledge, just in order to use what they bought.

Basically, both CEOs are saying the same thing: The number on your invoice isn’t what you actually paid. You might even have already overpaid for what a model could realistically achieve, bringing your ROI into doubt.

With these CEOs starting to voice the ‘truth’ (certainly not because it aligns with their incentives), I’ll answer these five questions:

• What does AI actually cost? Imagine there are no subsidies.

• Apart from money, what else are you paying the labs (such as OpenAI and Anthropic) for?

• How does AI rewrite competition? Do you stand a chance against your vendor-turned-competitor?

• What do the hyperscalers' CEOs recommend the world do?

• What does it all mean to the AI industry?

To start with, the true token cost.

Read more

You likely have some vague impression of a company called Moonshot and their AI model. Don’t worry about it, because you’ll know of them soon enough. They’ve recently released Kimi K3, an open-weight model that topped multiple coding benchmarks and closed in on Anthropic’s and OpenAI’s flagship model…

Read more

But since 2025, something has changed.

The idealist who charmed developers and users is suddenly gone. In its place, a capitalist CEO prioritizes shareholders and fund managers.

In this one, I’ll cover:

• Where did Llama go? Is this the beginning of the end for open-weight AI?

• Is Meta really preparing to sell compute? How serious is Zuckerberg about this?

• Is enterprise AI becoming the next Meta business?

• What should we look for in the next earnings call?

But let’s start with a little history, so you know why this is such a big deal.

Read more

Chinese officials are apparently kicking around a new idea: limiting overseas access to China’s most advanced AI models.

→ Reuters

Trump is a pioneer of many things (I’m practicing positive thinking these days). The latest one, he initiated export controls on AI models, and others soon followed suit.

That’s the latest lesson Bei…

Read more

Ever heard of Bending Spoons? It’s the Italian software group behind acquisitions such as Evernote, WeTransfer, Vimeo, Eventbrite, Brightcove, and AOL, which raised about $1.68 billion in its IPO after pricing above its range. The company doesn’t follow the typical PE playbook, but turns a sunsetting Saas into a dura…

Read more

As of this writing, the stock has nearly tripled since January.

Everyone is talking about Micron: YouTube, news, your neighbor, colleague… in particular, how you’ve missed out big time if you haven’t bought the stock!

But that’s not why I’m here and not why you’re here either.

I’ve read their and others' earnings calls in the last few days, and in this, I’m going to explain to you…

• Why does it suddenly get so much attention?

• Why does the stock price boom?

• The RAM shortage and why does AI make this worse?

• How! How this memory party is not a good sign for companies adopting AI, and even worse news for AI labs?

• And finally, what to look out for?

Let’s start with who Micron is in 30 seconds

Read more

Yes, they asked for regulation, they built the safety brand, they brought trusted partners into the testing process, and then one of those partners reportedly helped trigger the government response. There is irony there, and honestly, it is hard to ignore.

But if we stop at the irony, we miss the critical lesson.

This is not unique to Anthropic.

If frontier models become powerful enough, every major lab will eventually face the same kind of pressure. OpenAI, Google, xAI, whoever comes next.

Specifically, they all depend on timing: the model launches, customers show up, usage scales, and the revenue curve catches up to the cost curve before the gap becomes fatal.

Until now, no one saw an export ban coming into the equation.

So what happens when the AI economy discovers that its entire revenue estimation is off by an administrative order?

Following this question, a few more threads to pull:

1. How much model risk have ordinary companies inherited without seeing it?

2. What happens to a frontier lab when revenue is delayed but compute costs keep running?

3. And how far does the damage travel once the financing behind the boom gets repriced?

To see that, we have to start with what actually happened.

Read more

Coinbase CEO (Brian Armstrong) provided 5 approaches they used to cut AI spending nearly in half: make AI costs visible, route tasks to cheaper models, cache more, keep context lean, and use open-weight models where they are good enough.

→ Business Insider

After months of tokenmaxxing, people are finally coming to their …

Read more

Rewind.

If the only tech news that mattered last week was SpaceX’s largest public offering in history, then the one piece of tech news to know this week is SpaceX’s acquisition of Cursor.

This is pretty much a done deal, something they were already planning since April. Yes, so what’s left to talk about?

A lot.

To start with, this is the one thing that will impact the landscape of the enterprise coding platforms over the next 3 years.

So, I’ll answer the following questions in this one:

1. Why even buy Cursor to start with?

2. Why was Musk in such a rush to make this decision?

3. What Musk actually bought?

4. How would this impact Anthropic? (yes, Anthropic)

5. Two possible war-room scenarios for all parties over the next three years

Shall we?

Read more