Tech & AI Twitter Unpacked: The Smartest Daily Chatter You Missed

Liquid AI just released a 350-million-parameter model that outperforms models twice its size. This isn't just another benchmark win... it's a signal that the entire AI race is changing lanes. For years, the game was scale. More data, more parameters, bigger models. But the performance gap between frontier models like GPT-4o and Claude 3.7 has shrunk to just a few percentage points. The model itself is becoming a commodity. The new competitive moat isn't the engine... it's the driver. It's what the community is calling Agent Experience, or AX. Users don't care if your model is one percent better on a leaderboard.

They care if the agent understands their intent, handles errors gracefully, and actually gets the job done. This is why the most talked-about founders aren't building another general-purpose assistant. They're building what one analyst called "the most boring, domain-specific, deeply embedded software the world has ever seen." Because that’s where a superior agent experience can actually be built—on a foundation of deep, specific domain data. This shift from raw power to applied purpose is forcing a much harder conversation. And that conversation just exploded between Anthropic and the Pentagon.

CEO Dario Amodei is publicly resisting Pentagon demands for unrestricted use of Claude in military applications... including mass surveillance and autonomous weapons. The Pentagon’s response was to label Anthropic a "supply chain risk," and the Trump administration followed by halting federal use of their tech. Amodei is holding a line, insisting on guardrails. The Pentagon insists existing laws are enough. This is the conflict that defines the next era. It's not about if the AI can do something, but if it should. And we just got a field report on that exact question. A career diplomat, cited by Gary Marcus, reflected on the recent US-Iran conflict.

The consensus? AI was useful for operations... but it was not good on strategy. Generative AI can process logistics and manage tactical support. It cannot, however, project into novel situations or build the kind of robust world model needed for strategic foresight. It lacks judgment. So as the industry pivots from bigger models to smarter agents, the highest-stakes customer is discovering the technology’s fundamental limits. The real story isn't about building a better brain; it's about deciding what kind of work we trust it to do.