1. Anthropic Raises $65B at $965B — The Agent Infrastructure Bet

Anthropic just closed a $65B Series H at a $965B post-money valuation — the largest single raise in AI history. The capital isn't earmarked for model training. It goes to data center buildout and agent runtime infrastructure, the plumbing that would make Anthropic's stack the substrate every agent runs on.

This is the moment the infrastructure arms race goes explicit. The question isn't whether Anthropic builds good models — it's whether the company aims to be the AWS of agentic compute. Every builder now has a binary decision: build on Anthropic's emerging runtime layer, or build alongside it and stay stack-independent.

Why it matters: Anthropic's $65B isn't a valuation milestone — it's a forcing function for every builder to decide whether their stack runs on or around Anthropic's infrastructure.

Read more →

2. Ktx: An Open-Source Context Layer for Data Agents

Ktx is a new open-source executable context layer for data agents — persistent state, tool integration, and memory management in a single drop-in package. The library is designed to solve the "agent forgets what it was doing" problem that kills long-running data pipelines in production.

Persistent state is the unsolved problem most builders paper over with hacks. Ktx offers a clean abstraction you can wire up in an afternoon, not a framework that requires weeks of onboarding. It works today, which matters more than the framework debate.

Why it matters: Ktx's drop-in persistent-state layer closes the gap between demo agents that work once and production agents that work reliably — wire it in before the next pipeline restart breaks your long-running workflow.

Read more →

3. ITBench-AA: Frontier Models Score Below 50% on Enterprise IT Tasks

IBM and Artificial Analysis released ITBench-AA, the first benchmark targeting agentic enterprise IT tasks — incident response, system configuration, real-world IT workflows. Frontier models scored below 50%. Every major model tested fell short of what enterprise IT actually demands.

The benchmark is uncomfortable data for anyone selling agents to enterprise IT teams. The gap between demos and production workflows is still wide. Wide gaps are also where the valuable work — and the durable companies — get built.

Why it matters: ITBench-AA's sub-50% scores are the honest baseline every enterprise agent builder needs — use them to set client expectations now rather than let a production incident set them later.

Read more →

The three stories this edition share a common thread: the agentic stack is being defined in real time, and the winners will be those who solve infrastructure, memory, and production readiness. Anthropic's raise signals a land grab for the runtime layer; Ktx offers a practical fix for the memory gap; ITBench-AA reminds us that benchmarks must reflect real-world demands. Builders should watch for consolidation around infrastructure providers and invest in solutions that bridge the demo-to-production chasm.

Ktx is an open-source executable context layer for data agents. Drop it into a data pipeline and get persistent state, tool integration, and memory management out of the box — no new framework to learn. The #1 production failure mode for agents is context loss; Ktx fixes that in an afternoon. https://github.com/Kaelio/ktx

Today's edition: 58 sources scanned by Atlas (DeepSeek) → Curator (Claude) selected the stories → Scribe (Claude) wrote the draft → Mercury (DeepSeek) formats for delivery. Atlas: $0.003 | Claude agents: ~$0 (Max subscription). The curation call today was keeping three agent-supply-chain stories (SQLite, protestware, banning PRs) as Radar rather than surfacing one as a Top 3 — the stack sovereignty through-line held tighter with the raise, the memory fix, and the enterprise benchmark as the lead trio.