19 models, one conductor

The conductor has arrived. Perplexity just launched Computer—an AI agent that doesn’t pick one model. It orchestrates 19 of them simultaneously. Claude for reasoning, Gemini for research, GPT-5.2 for context, Grok for speed. $200/month, month-long autonomous workflows, persistent memory across sessions. We broke down the architecture, the pricing trap, and the dependency paradox—because the orchestra metaphor cuts both ways when you don’t own any of the instruments.

The agentic AI race is now a three-way war. Anthropic’s Claude Cowork just expanded with enterprise plugins that triggered a $285 billion SaaS selloff—and the hardware underneath keeps reshuffling: AMD locked in a $60 billion deal with Meta to challenge Nvidia’s chip monopoly. Nvidia, for its part, just posted $68.1 billion in Q4 revenue—up 73% year-over-year. The infrastructure buildout is accelerating faster than the platforms fighting over it.

Here’s the twist: even as SaaS stocks cratered, Salesforce reported Agentforce hit $800 million ARR with 2.4 billion agentic work units completed. Enterprise software isn’t dying—it’s being forced to go AI-native. And for developers building on this wave, we published a hands-on guide to building Google Analytics and Search Console AI skills—the kind of practical tooling that separates reading about AI agents from actually deploying them.

Latest from PulseMark