BREAKING
09:34 UTC
AI IDE shake-up: Windsurf vs Cursor heat up, migration guides land, user churn visible
trend pattern
medium
The AI IDE race is unsettled—pilot Windsurf and Cursor on your code before you pick a team standard.
anthropic
09:35 UTC
Claude Code 2.1.88 ships big reliability fixes, a retry hook, and smoother terminal UX
new feature deep dive
medium
Upgrade to Claude Code 2.1.88 for steadier long sessions, reliable multi-schema output, cleaner terminal UX, and a new retry hook for safer automation.
github
09:38 UTC
Copilot shifts to default training on interaction data (non‑enterprise) as Copilot CLI 1.0.14 lands with stability and BYOM fixes
policy legal enterprise
medium
Lock down Copilot data-sharing defaults for non-enterprise seats and standardize on Copilot CLI 1.0.14 for smoother BYOM and MCP workflows.
openai
09:39 UTC
Codex command injection let attackers steal GitHub tokens; fixes shipped—teams should rotate and harden now
release problems outages controversies
high
Now-fixed Codex injection exposed GitHub tokens; rotate credentials and harden agent–GitHub boundaries to cut blast radius.
model-context-protocol-mcp
09:41 UTC
Make your MCP registry the agent control plane (and sanity‑check v0.8.2)
trend pattern
medium
Treat the MCP registry as your agent control plane and pilot it now, while validating agents-js v0.8.2 changes in canary.
github
09:42 UTC
Agentic QE v3.8.13 ships code-intel CLI, incremental indexing, and a command-injection fix
new feature deep dive
medium
Upgrade to agentic-qe v3.8.13, turn on the CLI for fast complexity and incremental scans, and pin versions to stay clear of the patched injection bug.
datasette
09:44 UTC
Local LLMs for engineering: promise, pitfalls, and the guardrails you need
trend pattern
medium
Local LLMs can pay off, but only if you engineer the harness and enforce model guardrails like any other production dependency.
massgen
09:46 UTC
Multi-agent coding is getting a real playbook: when to verify, how to evaluate
trend pattern
medium
Use multi-agent roles with opinionated evaluation, and only add a verifier when upstream feedback isn’t yet reliable.