In partnership with
News of the day
1. Anthropic's Claude generated proofs of correctness for 123 open-source libraries, catching bugs missed by years of testing. → Read more
2. Banks and regulators debate liability when AI agents execute payments beyond customer instructions, as Visa pilots agent-driven supplier payments. → Read more
3. Blackstone is expanding its role in AI financing, backing data centers, neoclouds, and frontier models despite rising debt concerns. Khaira leads its new AI hub. → Read more
4. Microsoft launches streaming speech models in Foundry, transcribing and voicing responses in near real time. → Read more
Our take
Hi Dotikers!
This week, Anthropic quietly dropped something that might matter more than any flashy chatbot update: Claude now writes mathematical proofs that code is correct. Not tests. Proofs. The company unleashed its model on 123 open-source libraries and generated roughly 180 million lines of proof code, catching bugs along the way that years of human testing had completely missed.
Let that sink in for a second. Formal verification, the gold standard of software safety, has always been the domain of a tiny priesthood of specialists. It is so expensive and so slow that it has mostly been reserved for aerospace systems, cryptography and other places where a bug means catastrophe. The rest of us settle for unit tests, which are basically a polite way of saying "we checked a few things and crossed our fingers."
What Anthropic is proposing is a genuine shift: making formal verification as routine as running a test suite. If that works at scale, it changes the economics of trustworthy software. And there is a delicious irony here, because the loudest criticism of AI-generated code is that nobody can trust it. Anthropic's answer is essentially to have the AI mathematically prove its own homework before handing it in.
This is exactly the direction the industry should be heading. The real risk of AI coding assistants was never that they write code, it is that they write plausible code faster than humans can review it. Pairing generation with verification attacks that problem at the root, rather than adding yet another layer of human eyeballs that will inevitably get tired.
Of course, proofs are only as good as the specifications behind them, and writing good specs remains hard. But between "trust me, it passed the tests" and "here is a mathematical proof," the choice is not difficult. Watch this space closely.
Alex.
Why Most AI Tools Fail at B2B Customer Service
B2B customer issues rarely resolve in a single interaction. They move across teams, systems, and account history that most AI tools were never designed to track.
A new briefing paper from Harvard Business Review Analytic Services, sponsored by Front, looks at why AI built for high-volume consumer support creates new coordination gaps when it meets the multi-team reality of B2B service, and what leaders should ask before investing in their next AI tool.
Get the free briefing paper to see the three questions that separate tools built for your environment from tools that will create more work.
Meme of the day
