A growing number of engineering teams are ditching Markdown for HTML when writing specs — not because they enjoy writing more verbose documents, but because HTML’s semantic structure gives AI agents significantly richer context when implementing from a spec. Here is where the tradeoff makes sense and how to do it well.
Anthropic’s Mythos can autonomously find and exploit thousands of zero-day vulnerabilities across every major OS and browser. Access is tightly controlled — by a US company, for US-aligned entities. The Atlantic Council calls it more consequential than the Iran war. They’re right. The US just turned AI into a cyberweapon and nobody voted on it.
Microsoft’s Experiences + Devices division is canceling thousands of Claude Code licenses by June 30, forcing engineers onto GitHub Copilot CLI. The headline looks bad for Anthropic. The reality is more complicated — and more instructive.
Claude Code v2.1.129 shipped quietly on May 6. For most users it’s invisible. For Bedrock enterprise shops, MCP server operators, and anyone using third-party model gateways, it changes real behavior — and one change is a deliberate reversal of v2.1.126.
Anthropic has signed a $1.8B, seven-year computing contract with Akamai Technologies — the largest deal in Akamai’s history. Akamai isn’t just a CDN anymore: it launched a global AI inference network across 4,400 edge locations built on NVIDIA Blackwell GPUs in March. The deal is the fourth pillar of Anthropic’s deliberate strategy to never depend on a single compute supplier.
Anthropic is in early talks to raise at least $30B at a pre-money valuation exceeding $900B — nearly triple its February figure of $380B. The leap is backed by real revenue: $44B annualized ARR, 70% gross margins, and Claude Code generating $2.5B on its own. If the round closes, Anthropic would surpass OpenAI’s $852B March valuation.
OpenAI launched a $4B+ PE-backed deployment company on May 11, acquiring AI consultancy Tomoro and embedding 150 engineers into enterprise clients. The structure tells a story: if models alone were enough to win enterprise, you wouldn’t need a 1,000-person professional services arm.
Google I/O 2026 runs May 19–20. Gemini 4 with a 2M+ token context window is the headliner, but the more important story is Firebase Studio becoming an agent-native development platform — Google’s direct answer to Claude Code. Here’s what to watch and why it matters.
Moonshot AI’s Kimi K2.6 landed on April 20 as the most capable open-weight coding model ever released: 1T-parameter MoE, 58.6% SWE-bench Pro, 66.7% Terminal-Bench 2.0, and an Agent Swarm that scales to 300 sub-agents executing 4,000 coordinated steps — at $0.60 per million input tokens.
Outcomes and Multiagent orchestration moved to public beta on May 6. This is the practical guide to deploying self-verifying, multi-agent workflows in production — including how to write rubrics that actually work, the 20-agent coordinator limit, and what Netflix built with it.