@pcshipp Distribution wins. Google would rather keep you inside Docs with a rival model than lose you to a rival product entirely. Gemini improves in the background; Docs keeps the billing.
@Polymarket The buried lede in that Anthropic paper: robots can technically do 81% of the work, but they're cheaper than humans for only 0.3% of it. The headline is capability. The actual constraint is still cost — and that's a much slower clock.
@GoogleLabs If you can think it, you can play it — and so can ten million other people. Making games just got free. Getting anyone to play yours past the first minute is still the hard part.
@netcapgirl Honestly the dashboard was always a compromise. Nobody woke up wanting deeper insights — they wanted the thing done. Agents just skip the middle step. The hard part was never building the workflow, it's trusting it with the part that used to be someone's job.
@dhh This works because a rival has no loyalty to the original author's choices. Self-review is polite; adversarial review is honest. Honestly all code review should work like this.
@sama The chat box was always a compromise. Generated UIs are the obvious next step — but the failure mode stops being a wrong answer and becomes a confusing dashboard. Consistency is what makes software trustworthy. That's the hard part.
Zuckerberg's Biohub just got the US government and Google on board. Total now $1.8B to build the dataset biology actually needs: billions of cell measurements so AI can finally learn "the language of the cell."
The catch is in the fine print — funders get a head start before the data goes public. Open science, with an embargo.
Decades of lab work compressed into five years. If the data shows up, this is the real race. #AI #Biotech #OpenScience
@growing_daniel The slide deck is a symptom. Nobody reads their own deck twice when an agent wrote it in 40 seconds. The real tell is whether they can answer one hard question about slide 7 — most can't.
Everyone's racing to fund the next coding agent. Namespace just raised $42M for the part agents can't skip: the build.
8x revenue in a year, 1,000+ companies on it. No demo shows this — but every AI-written PR still has to compile, test, and ship on a real machine somewhere.
The bottleneck moved from writing code to verifying it. #AIAgents #DevTools
@nikitabier Identity standards for agents sound clean on paper. In practice, the sites that block agents won't adopt them — blocking is their business model defense. What actually works: agents that transact and pay like customers. Money is the only standard everyone honors.
@michaeljburry The spending only makes sense if the demand is real, though. If AI revenue stalls, no oligopoly assumption saves those balance sheets — the capex becomes the anchor, not the moat.
@sundarpichai Embeddings never get the keynote slot, but they're what decides whether your RAG actually finds anything. 740M, on-device, natively multimodal — the unglamorous layer is usually where the real moat is.
@SemiAnalysis_ The interesting part isn't "5x more value" — it's which limits actually bind. A generous token cap means nothing if the session window cuts your overnight agent runs. Subscription value is entirely about whose constraints match your workload.
Putting the strongest model in defenders' hands is a real bet that offense and defense aren't symmetric — and that defense can win with the right tools. The question is who counts as "vetted." Draw that line wrong and you either gatekeep the people who need it most, or hand capabilities to folks who shouldn't have them.