AgentsAnthropic 2026 Agentic Coding Trends ReportAnthropic·high signal60% of dev work uses AI but only 0-20% fully delegated. Rakuten tested Claude Code on 12.5M-line codebase with 99.9% accuracy in 7 hours.SourceSource pageAnthropic↳ Follow the threadShared entity / Policy dependencyKastra Launches Policy Enforcement for Claude Code, Cursor, and CodexHacker NewsShared entity / Stack layerClaude Code v2.1.203 and v2.1.204 Add Login-Expiry Warnings and Fix Headless Hook StreamingClaude Code DocsShared entity / Policy dependencyAnthropic Restores Global Access to Fable 5 and Mythos 5, Launches Claude for Government BetaAnthropicShared entity / Stack layerClaude Code v2.1.206 Adds a /doctor Check That Proposes Trimming Your CLAUDE.mdClaude Code DocsShared entity / Stack layerUnderstanding AI: Fable 5's Filters Are Strict Because It's Built on Mythos, a Model Anthropic Refused to ShipUnderstanding AI (Timothy B. Lee)Shared entity / Stack layerClaude Code v2.1.205 Blocks Auto-Mode Transcript Tampering and Prompts Before rm -rf on Unresolved VariablesClaude Code ChangelogShared entity / Policy dependencyponytail: A 'Laziest Senior Dev' Skill That Cuts Agent-Written Code by ~54% Is the Fastest-Rising Repo on GitHubGitHubShared entity / Stack layerClaude Cowork Sandbox-Escape Chain Reportedly Escalates Local Code Execution to RootCyberdesserts