Source item

1/ 🔥 @NoPriorsPod x @LatentSpacePod chat with @SatyaNadella at @Microsoft Build. He has the sharpest mental models of any public company CEO I've interviewed. $MSFT is at its heart still a tools company! Big focus on agentic coding, harness & AI evals. Takeaways:👇 https://t.co/iAAjHdCVnK 2/ Satya started with a reminder: to be a platform, more value has to be created on top than is captured first party. Microsoft wants a healthy ecosystem! Their goal: making any company, AI-native or legacy enterprise, a first-class participant with its own AI. 3/ Top quote: "Can everyone operate at the frontier with *their* frontier intelligence? That is so important, because otherwise I don't know how you achieve stable equilibrium. How do I go and say, well, my company is gonna have a terminal value because I now know how to continuously compound on top of a platform. Without it, why have a developer conference? I can just come and have you all sort of just worship at the altar of one model." 4/ On @MicrosoftAI -- @mustafasuleyman and team aren't chasing benchmarks. This was an opinionated model release: clean-lineage base plus a hill-climbing scaffold for customers. Proof of success: a 5B(!) reasoning model beat GPT-5 for certain tasks on the Land O'Lakes demo. 5/ Thus, the definition of "frontier" is increasingly task-specific, and speed and efficiency matter. The new plan for winning with AI: take a small cognitive core, learn from your own traces => beat bigger models on *your* tasks. 6/ Your private evals may be the biggest IP your company owns. Satya's acid test: run an eval on Model A. Swap in Model B. If you can hill-climb, you control your stack. If you can't, you don't. 7/ One disarmingly honest admission about revisited assumptions: the last 2 years showed "intelligence is log of compute" was directionally right, but he (and researchers) underestimated how hard deploying tokens into real-world value would be! It will take the work of many. 8/ Coding AI has worked so well it broke the IDE. 100 parallel agent sessions create cognitive load that no chatbox (or human) can hold. The next UI is a canvas for managing delegated agents. (Human) Attention is all you need. 9/ Token capital and human capital compound inside a company. Agents trained on company traces become "company veterans" that should sit on the balance sheet. Tacit knowledge finally becomes a capturable asset. 10/ SaaS isn't dead. It's getting unbundled. Data models and business logic under existing apps are still valuable, but packaging changes. Work IQ exposing the M365 data layer as an agent substrate is one example to study. (also, pricing models are unstable, and path-dependent) 11/ One fun story: Azure's networking team didn't ask for more headcount. They asked for more tokens. They reconceptualized their job from "doing Azure networking" to "building the agentic system that does Azure networking." Meta-work is the work. 12/ The implied alternative to Satya's terminal value argument is the labs...eat the economy. Every company becomes a temporary wrapper, all margin routes through a frontier API, and no one underwrites any LT investment because the ground keeps moving. Satya's pitch is that he'd rather be the floor under your terminal value than another model fighting for the ceiling. 13/13 all in all, super impressed on the agility of their AI strategy. Microsoft will microsoft! Full interview posted tomorrow. Full interview here and anywhere you get podcasts https://t.co/LxCRLjrs0K