Perplexity Ships the "Agentic Cloud"

Perplexity Ships the "Agentic Cloud"

Perplexity's new Agent API, launched August 13, is a bet that the next platform war isn't about better search or better chat — it's about who owns the plumbing for agents that act. The API bundles web search, URL fetching, code execution, MCP servers, and connectors for GitHub and Slack into a single programmable endpoint, with presets for building agents and multi-provider orchestration across OpenAI, Anthropic, and others [2]. It also plugs into Public for agentic trading, which tells you exactly how far "agent does things with real consequences" has already gone [2].

Aravind Srinivas framed it as infrastructure for the "Agentic Cloud," and the developer reaction has been notably upbeat — this reads less like a chatbot feature and more like AWS-for-agents positioning [2].

What's interesting is the model-agnostic angle: Perplexity isn't betting on one lab's models winning, it's betting on being the orchestration layer regardless of who wins. That's the smart move right now, and it's exactly the kind of infrastructure bet we'd expect to see more of as the "which model is best" question becomes less important than "who's wiring it all together."

Salesforce Puts Its Entire CRM Inside Claude

This is the one to watch. On August 26, Salesforce and Anthropic announced "Claudeforce" — a plugin that pushes Salesforce's CRM data, workflows, and governance directly into Claude, with 37 prebuilt sales skills like meeting prep and pipeline review baked in [3]. Pilot customers get it now; open beta lands in September [3]. VentureBeat's framing was blunt: Salesforce just put its entire CRM inside Claude and is telling customers they'll never need the app again [3].

Marc Benioff and Dario Amodei pitched it as fusing probabilistic AI reasoning with deterministic CRM systems — which is a polite way of saying the UI is moving from the app layer to the model layer, and Salesforce is fine becoming the backend database as long as it stays the system of record [3]. TechTarget's coverage picked up on exactly that shift [3].

If this works, it's a template every enterprise SaaS vendor will have to answer within a year: either become the intelligence layer, or become someone else's backend. There's no comfortable middle path left.

Nvidia Reportedly Closing In on a $13B Hugging Face Acquisition

As of August 26-27, multiple outlets — Forbes, Reuters, TechCrunch — report Nvidia has agreed to or is close to acquiring Hugging Face for somewhere between $12.9B and $13B+ [4]. For context, Hugging Face was valued at $4.5B in 2023 and currently generates roughly $150M in annualized revenue — so this is a valuation multiple built almost entirely on strategic position, not current cash flow [4].

The strategic logic is straightforward: Hugging Face is the default home for open-source models, and Nvidia wants to own the on-ramp to that ecosystem as closed-source labs (OpenAI, Anthropic, Google) pull further ahead on frontier capability. Owning the open-source distribution layer is a hedge against a world where the frontier gets locked down — Nvidia gets to sell picks and shovels either way [4].

The buzz online is squarely about consolidation risk: the platform that's supposed to be the neutral home for open AI potentially becoming a subsidiary of the biggest hardware vendor in the space. Worth watching how the open-source community reacts once (if) this closes.