The New Stack published a comprehensive roundup of what it called Anthropic's most chaotic and productive month in the company's history. March 2026 saw at least 14 major product launches alongside five significant service outages, capped by two separate data leaks that exposed both an unannounced model and the tool's entire source code.
On the product side, the pace was relentless. Claude Code shipped auto mode (letting the agent decide what is safe to run autonomously), Channels (connecting to Telegram, Discord, and iMessage), computer use (point-and-click screen control), voice mode, and the general availability of the 1M token context window at standard pricing. Claude Code also landed on web and mobile for the first time, letting developers kick off parallel agentic sessions from their phones.
The leaks stole the headlines. First, a misconfigured CMS exposed draft blog posts describing Claude Mythos — the most powerful model Anthropic has ever built, sitting in a new tier above Opus. Then, on the last day of the month, a source map file in the npm registry revealed Claude Code's full 512,000-line TypeScript codebase, including unreleased features like BUDDY (an AI pet) and KAIROS (an always-on persistent agent).
For the context engineering community, March 2026 demonstrated both the extraordinary velocity of Anthropic's product team and the growing pains of scaling infrastructure to match explosive demand. The five outages — including multiple periods where Claude Code was partially or fully unavailable — underscored that reliability remains the company's biggest challenge as it races to ship features.
