The Agent Stack #039 — Friday Signal

AWS, Cloudflare, and Google are quietly redesigning cloud infrastructure for a future where machines generate most internet traffic. This isn’t about serving humans anymore.

The shift is already happening. Cloudflare launched a unified data platform this week specifically designed for AI agents to query and act on real-time data. AWS rolled out new edge computing services optimised for agent-to-agent communication. The old request-response web model breaks down when you have thousands of agents making millions of API calls per second.

This matters because your agents need infrastructure that can handle their behaviour patterns. Human users browse, read, click. Agents hammer APIs, process streams, and coordinate with other agents in tight loops. Traditional CDNs and load balancers weren’t built for this.

The economics are shifting too. AI tokens are being treated as tradable commodities like electricity or bandwidth. Major exchanges are designing derivative products around AI compute tokens. When your agent’s reasoning becomes as measurable as kilowatt hours, the entire pricing model for cloud services changes.

Meanwhile, enterprise deals are stalling on safety concerns rather than capability questions. Databricks reports that security and governance kill more AI deals than technical limitations. Companies want agents that can act autonomously but within strict boundaries.

Quick Hits

• Anthropic raised £51 billion at a £762 billion valuation, potentially their final private round before IPO. Their new Opus 4.8 includes “Dynamic Workflows” for coordinating agent swarms.

• Asana acquired StackAI for undisclosed terms, adding no-code agent building to their workflow platform. Another sign that traditional productivity companies are absorbing AI-native startups.

• CNN sued Perplexity for generating “verbatim” copies of paywalled articles. The first major copyright case targeting AI agents that browse and summarise content rather than just training on it.

One Thing to Try

Set up automated regression testing for your agents. That HN thread about catching agent failures after prompt changes hits home. Create a test suite that runs your agent through key scenarios after every model or prompt update. Even basic evals beat discovering regressions through user complaints.

The infrastructure is changing. Make sure your agents are ready for it.