Three weekend developments — the Fable retraction, Satya's ecosystem thesis & Salesforce's $3.6B Fin acquisition — reveal why the moat has shifted from models to harnesses.
Guardrails on frontier AI were inevitable given the pace of adoption. But the capabilities available to corporations remain tremendous.
Frontier model prices keep rising while open-source crosses the good enough line. Coinbase, Lindy, Harvey & Cursor are substituting — & the savings go straight back into more tokens.
A small router model on a single Mac now handles roughly four in five background tasks, cutting cloud spend & wait times without changing the work itself.
Microsoft's MAI-Code-1-Flash matches Claude Haiku 4.5 on SWE-Bench Verified using a third of the tokens. The right metric for AI in production is intelligence per dollar, value per token spent.
OpenRouter token data shows open-weight models overtaking closed models as provider leadership fragments across Anthropic, DeepSeek, Google, OpenAI, & a long tail.
Short interest in public AI stocks has risen modestly overall, but the pressure is concentrated in AI cloud and smaller software names.
How a personal AI agent built on markdown skills lets a frontier model teach smaller, local models to do real work, without retraining.
Software is no longer about UX & data. It is about the harness, the layer that turns an LLM into a reliable agent. Seven components define the new stack.
If data gravity defined the last decade, agent gravity will define this one. As AI agents become the primary interface to enterprise data, the platforms that run them will capture an outsized share of value — & they're already locking the doors.
