Developer Tools 32
- Agent Skill Registries Need Dependency Governance
- why your ai agent fails: it is a harness problem, not a prompt problem
- Herdr is the throughput layer for coding agents
- codex stopped being only a developer tool
- choosing the right agent surface: Codex, Claude Code, Copilot, and when not to use an agent
- review and verification: making agent code mergeable
- subagents and specialized agents: parallelism without chaos
- instructions as code: AGENTS.md, CLAUDE.md, and copilot-instructions.md
- the senior engineer's agent loop: context, plan, patch, proof
- how senior engineers should code with agents without becoming reviewers of chaos
- coding benchmarks are becoming production dependencies
- agent kits are turning developer docs into executable guardrails
- github copilot is becoming an ai budget control plane
- copilot browser tools make the frontend reviewable
- the kubernetes dashboard died and the platform workbench took its place
- local AI runtimes are part of your attack surface now
- agents need shared memory, not another context window
- AI code governance is the new code review bottleneck
- plugin marketplaces are the new endpoint policy for coding agents
- Bring-your-own-model is a control plane problem
- AI credits are the new lines of code metric
- AGENTS.md is becoming the new code review contract
- agentic workflows are being domesticated by actions
- the developer laptop is the first production environment for agents
- agent canvases are the end of chat-only coding tools
- copilot cloud agent is becoming an automation api
- How to safely inspect a repository before running anything on your machine
- replit chased the ai hype and left a simple coding gap
- mcp catalogs are becoming the new internal developer portal
- agentic terminals are eating the IDE (and why that is not entirely good)
- Your Terminal Is Becoming a Governed AI Runtime
- AI Tools Have Shorter Half-Lives Than the Workflows They Automate