Nvidia proved the boom is broadening | Thursday, August 27, 2026

Salesforce and Anthropic announced Claudeforce, and the headline feature is Salesforce in Claude: a plugin with 37 pre-built sales skills that lets a seller prepare for a meeting, review deal health, and update a pipeline by talking to the assistant, without opening Salesforce at all. It runs over MCP through a Salesforce harness that governs what the agent can touch. Pilot now, open beta in September. The enterprise question this raises is larger than one vendor. If the system of record becomes something you reach through an agent rather than a screen, the interface layer, and the pricing power that sits on it, is up for grabs. Benioff is already answering "SaaSpocalypse" worries, which tells you the incumbents feel the shift too.

Emerald AI raised $150 million in a Series A at a $1.05 billion valuation, co-led by Energize Capital and DCVC, to scale software that flexes a data center's power draw in response to grid conditions. The pitch is that flexibility could unlock more than 100 gigawatts of capacity on the existing US grid without building new plants, and the company is working with Digital Realty and Nvidia on a roughly 100-megawatt power-flexible AI factory in Manassas, Virginia. Nvidia can guide to $108 billion, but that revenue only lands if someone can plug the racks in. Watch power as the binding constraint on your own AI timeline, not chips.

Gatik, which runs autonomous box trucks on fixed middle-mile routes, raised $200 million led by the Qatar Investment Authority and Koch Disruptive Technologies, two months after signing a multiyear commercial agreement with PepsiCo. The company frames the round as its move from pilots to driverless operations for multiple customers across several cities. The read for enterprise leaders outside logistics is the pattern: the autonomy that attracts checks this size is the kind already booked against a named customer and a repeatable route. Constrained scope and a real contract beat open-ended capability, which is the same lesson agent programs are learning one workflow at a time.

AWS expanded Web Search on Amazon Bedrock AgentCore, adding per-request domain and published-date filtering and extending the managed tool to Europe and Asia Pacific alongside the US. The design point matters more than the feature list: agents fetch current, cited web knowledge with no data leaving the customer's own AWS environment. That is a direct answer to the reason many enterprises kept live web access away from their agents, namely that grounding a model on the open internet usually meant routing sensitive context through someone else's service. Region-scoped, logged, and inside your boundary is the shape enterprise-grade agent tooling is settling into.

The controls for autonomous agents are being written by standards bodies faster than by legislatures. NIST is developing standards specifically for autonomous agents around identity, action logging, and containment, and CISA's agentic-AI guidance calls for cryptographically verified agent identities, short-lived credentials, encrypted agent-to-agent communication, and mandatory human sign-off for irreversible actions. China went further and made an agent-specific framework enforceable in July, requiring each agent's authority to be tiered before deployment. None of this is a single national law yet. The practical instruction is the same one the technology keeps sending: give every agent an identity, a scope, and a logged trail, before a regulator or an incident makes you.