Sam Altman took the stand. Anthropic moved to buy the tool that powers its rivals.

analysis · stratechery · may 13

Why for you: Ben Thompson's frame for OpenAI's new $4B deployment company is the 1970s mainframe service bureau — IBM era firms that charged clients to run batch jobs on hardware the clients couldn't afford to own. OpenAI is doing the same thing at the model layer: clients want AI outcomes but don't want to hire AI teams, so they pay OpenAI to staff them. Thompson walks through the structural economics of why this works for OpenAI right now (it monetizes model capacity without waiting for API adoption), why it creates alignment risk (the consulting arm competes for talent and attention with the research arm), and what the Apple and Intel earnings disclosures tell him about which AI infrastructure bets are compounding. The historical parallel alone is worth the read; the rest is a signal map for the next 12 months of AI business strategy.

article · anthropic / m&a · may 13

Why for you: Stainless makes SDK generators — give it an API spec and it produces idiomatic client libraries in Python, TypeScript, Ruby, Go, and Java. OpenAI uses it for their official SDKs. Google uses it too. Anthropic reportedly wants it for at least $300M. The strategic logic: controlling SDK generation means Anthropic can ship polished developer tooling as fast as it ships models, rather than playing catch-up on the libraries that most developers actually touch. The irony of acquiring a tool that also serves your two biggest competitors is either a power move or a future integration headache — probably both. Paywalled, but the headline tells most of the story.

article · anthropic / smb · may 13

Why for you: Claude for Small Business is not a new product tier — it's a set of pre-built integrations that wire Claude into tools small businesses already run: QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, Microsoft 365. The framing is agentic workflows for operators who have no AI team and no time to build one. The meaningful signal is distribution: Anthropic is now reachable through the SMB SaaS stack rather than only through developers calling the API. That's the same move Stripe made when it went from "API for developers" to "payments inside every tool your business already uses" — and it changes the competitive surface for Claude considerably.

article · anthropic / legal tech · may 12

Why for you: Claude for Legal ships 20+ MCP connectors for legal data sources and 12 practice-area plugins covering contract review, discovery, regulatory research, and litigation support. The MCP connector count is the number to watch: it means Claude becomes the reasoning layer that sits on top of existing legal databases rather than displacing them. Law firms don't want to migrate data — they want their data made smarter. Anthropic's approach (MCP connectors over proprietary data ingestion) is the right architecture for a profession that's deeply suspicious of anything that touches client confidentiality. The piece is written for legal audiences, which means the technology coverage is light but the implications for vertical AI strategy are clear.

article · security / google · may 12

Why for you: Google's Threat Intelligence Group blocked a state-affiliated operation that was using AI to plan mass exploitation of newly discovered zero-days — not to find vulnerabilities but to operationalize them faster once found. The distinction matters: the bottleneck in large-scale attacks has historically been the logistics of turning a known vulnerability into a working exploit across heterogeneous targets at speed. AI compresses that window. Google's own defensive posture relies on the same capability in reverse. The implication for anyone running production software: the patch-fast window is shrinking, and the threshold for "critical severity" is shifting because the time-to-weaponization assumption that informed it no longer holds.

article · openai / legal · may 12

Why for you: Altman took the stand to rebut Musk's claim that he was betrayed and cheated out of a nonprofit he helped found. His testimony: Musk demanded majority equity control and a CEO title before walking away in 2018; what Musk calls a stolen charity was a company Musk wanted to own. The $150B damages claim and the nonprofit conversion dispute are the legal surface. The structural story underneath is about what happens when an AI lab transitions from an idealistic research org to a commercial entity with real power — and who gets to write the historical account of that transition. If the verdict shapes OpenAI's governance structure, it has downstream effects for the whole industry's regulatory conversation.

article · google deepmind / ai interface · may 12

Why for you: The Magic Pointer is a Gemini-powered layer on top of the cursor: point at any element on a webpage and ask Gemini to compare products, explain text, visualize items in augmented reality, or take action. It's ambient — no chat window to open, no context switch. The design philosophy is "AI meets users across all tools without interrupting flow," which is the opposite of the current state where every AI interaction requires opening a sidebar or switching applications. Whether the implementation lands or not, the direction — AI as a layer on top of existing software rather than a replacement for it — is the interface pattern that will define the next generation of productivity tools. Includes live demos.

essay · noahpinion · may 05

Why for you: Noah Smith tracks how AI labs have quietly changed what they emphasize in public: AGI appears far less frequently in OpenAI's official communications than it did in 2018, replaced by job creation, economic growth, and human partnership. His read is that this is strategic repositioning ahead of regulatory scrutiny rather than a genuine change in belief — the labs need to avoid triggering the kind of existential-risk conversation that invites intervention. The interesting follow-on question he raises: if the labs are walking back the AGI framing publicly but still racing privately, what does that mean for the policy frameworks being built on the assumption that they'll be honest about their timelines and intentions? A short piece with a pointed argument.

signal · simon willison · may 13

Why for you: Simon Willison is launching an official blog for Datasette — his open-source tool for exploring SQLite databases — and specifically mentions upcoming announcements. Datasette is the kind of tool that shows up everywhere in a Rails developer's orbit: quick data exploration, lightweight APIs over CSVs, ad-hoc analysis without spinning up a full stack. Willison's track record for shipping useful things (the llm CLI, sqlite-utils, various AI integration experiments) makes a "we have announcements coming" post worth bookmarking. Nothing here yet, but the signal is real.