Dots are remarkably capable, always-on agents built to handle everything. They’re a whole new way to work with AI—one that gets to know what matters to you, is always working on your behalf, and takes important work off your plate so you get more of your time and attention back. — Read More
Monthly Archives: September 2026
Nvidia’s scale-in play: Controlling agents is the next infrastructure priority
Nvidia Corp. is extending the data processing unit from infrastructure offload to a broader security role across the artificial intelligence factory. The opportunity is to make agentic AI safer to operate at scale.
…Having built out its scale-up, scale-out and scale-across network architecture, the company is introducing another category it is calling scale-in. Its purpose is to connect more of the AI factory’s resources while extending security controls across the infrastructure. In Nvidia positioning, the DPU extends beyond a device that secures access to a server and becomes a separate place to monitor and control activity across the factory. — Read More
‘Vibe Coding’ Has Become ‘Doomcoding’
We’re not dumbing down. We’re dumbing from the middle out. Let’s talk about “doomcoding.”
I am going to uncover what I think the biggest problem is right now with “vibe coding” … or AI-assisted coding or agentic coding, basically any flavor or skill level of creating something with these fun new tools that’s going to get you ahead and make you a billionaire.
Because the big new problem with vibe coding is big. And new. And I don’t think anyone is addressing it yet.
It’s … doomcoding. I’ll explain. — Read More
How do you certify AGI?
We are racing toward AGI without agreeing on what would prove we’ve reached it. There is no stable definition, no benchmark that survives the pace of progress, and no reliable way to inspect what happens inside frontier models. As researchers increasingly warn about the jump from AGI to ASI, that Certification Gap is becoming a problem we may not have the luxury of solving after the fact.
Every major AI lab now treats artificial general intelligence as a delivery date, not a philosophy seminar. OpenAI’s founding charter is organized around it. Google DeepMind publishes formal frameworks for measuring progress toward it. And the money follows the conviction: by Stuart Russell’s count, the world is spending ten times what the Manhattan Project cost and a hundred times the price of the Large Hadron Collider, roughly 500 billion dollars so far, to build machines that match or exceed human intelligence across the board. — Read More
Trump signs executive order to launch AI-powered ‘America.gov’
President Donald Trump signed an executive order on Tuesday establishing America.gov, an AI-powered government portal through which users can access public services and search thousands of federal databases.
The order, “Streamlining access to government services through America.gov,” establishes the website and directs all agencies to integrate their public-facing services into the platform within 90 days. — Read More
Apps, Agents, and Aggregation
Three revolutionary products — you know the line. A wide-screen iPod with touch controls, a revolutionary mobile phone, and a breakthrough Internet communications device. That was how Steve Jobs introduced the iPhone.
I’ve linked to this snippet before, usually to note how the audience didn’t really understand what an “Internet communications device” was, even though that was the iPhone’s most revolutionary capability.
The phrase I’m thinking of right now, however, is “Are you getting it?” Are you getting that messaging — chatbots now, natural interfaces later — is how we will communicate with AI? Are you getting that not only will we not program computers, we won’t use them — AI will? Are you getting that pre-built UI — write once, run everywhere, for everyone — is dead? These are not three separate predictions: this is reality, right now, in 2026. The future is here, even if it’s not widely distributed. Or is it? — Read More
AI existential risk probabilities are (still) too unreliable to inform policy
Over two years ago we wrote a detailed deconstruction of p(doom) and argued that its primary effect is to launder vague intuitions and fears through a facade of quantification. The essay has fresh relevance today as p(doom) rhetoric is driving public discourse and policy attention to an unprecedented degree. Unsurprisingly, today’s probability estimates of AI existential risk are no more rigorous than the ones from 2024. Hence this repost. — Read More
Rogue OpenAI agents targeted three separate US government websites
OpenAI said Friday that some of its AI agents went rogue and probed US government websites this summer — the latest revelation of the artificial intelligence company’s technology.
… OpenAI said Saturday that its agents accessed publicly available data from the Commerce Department’s Census Bureau using login credentials it found online, and separately shared public data from the SEC website on another website. OpenAI’s agents attempted but failed to gain access to the Education Department and gather data from its civil rights office, according to the report. — Read More
Should Investors Be Concerned About The Recent AI Safety Incidents?
Yesterday Madison Mills from Axios wrote that OpenAI and other frontier labs are probing way more security incidents than we were previously aware of. The media is awash in AI doomsday forecasts but, as someone who is a little more optimistic that AI won’t kill us all, I want to discuss why. It helps to start by thinking about [three] of the assumptions in the AI doomsday mindset.
— There is just one dominant AI and it kills us
— We can’t figure out how to set guardrails to control AI
— AI implementations will look similar to the way they look today
— Read More
How to Get 10x Better AI Answers Without Writing Better Prompts
I spent three months rewriting the same prompt.
Not literally the same words. But every time an AI answer came back generic, my fix was always the same instinct. Add another adjective. Add “be specific.” Add “think step by step.” Add a fake expert persona at the top. I had a whole folder of “prompt templates” that I copy pasted into every new chat like a magic spell.
Then one day I gave Claude a two line prompt with zero clever phrasing and got the best output I had gotten in weeks. The only thing different was that I had pasted in my actual project files first.
That was the moment I stopped optimizing my prompts and started optimizing what the model actually knew before I asked it anything. The wording barely mattered. The information around the wording is what changed everything. — Read More
