← All daily reads · part of Radar
Radar — 4 Jul 2026
🛠️ Development
- Engineering Discipline: guardrails for AI coding agentsOpen-source skills/plugin repo that adds engineering guardrails — clarify, plan, verify, then clean up AI-written code — to Claude Code, Codex, Cursor, Gemini CLI and OpenCode.
- Codified Context: durable memory for coding agentsA companion repo and arXiv paper on giving coding agents durable memory — a three-tier hot/cold context setup so they keep a project's conventions across sessions instead of forgetting them.
- ASE: fusing agentic AI coding with software engineeringApache-2.0 toolkit from Ralf Engelschall that pairs agentic AI coding with traditional software engineering — agent hooks, skills, an MCP service and a CLI, shipped as a Claude Code plugin.
🎨 Design & UX
- A clear map of the Figma-to-code AI messA practical breakdown of today's Figma-to-code AI tools for product designers — what each does, where the gaps are, and how the design-to-code handoff actually fits together.
- An AI agent that reviews pages for a11y and design-system driftOpen-source multi-agent tool that audits a live URL for accessibility, UX heuristics and design-token drift in parallel, then posts a severity-ranked report — including as GitHub PR comments.
🧭 Delivery & PM
- The PM job market is splitting: orchestrators vs. the restA widely-shared read on how AI is splitting the PM job market — the people who can orchestrate whole systems of work and agents are pulling away from everyone else.