The month AI stopped being just a tool and started using tools.

The battle for the agent integration layer

Before: January focused on whether AI agents could handle work on their own, mostly one tool or one agent at a time.

Now: February introduced the structure above the agent. Perplexity Computer coordinated 19 models in parallel, and competition moved into the orchestration layer.

Models orchestrated through one interface 19

The model race entered a price-collapse phase

Before: In January, model competition was still mainly framed around capability. Cost mattered, but it was not the center of the story.

Now: In February, collapsing cost became the real headline. Once near-frontier performance is available at one-twentieth the price, single-model dependency starts looking irresponsible.

MiniMax cost relative to Claude 1/20

AI governance moved from principle to implementation

Before: AI ethics had often been discussed abstractly. Direct confrontation between state demand and model safety policy had not broken into the open in this way.

Now: Anthropic publicly said no to the Pentagon, turning safety guardrails into a visible point of strategic differentiation rather than a hidden internal policy.

xAI co-founders who had departed 6/12