The Missing Ingredient in Loop Engineering: Observability
Everyone's talking about loop engineering: automations, worktrees, sub-agents. They all skip the ingredient that makes letting agents run for days safely: observability.
More issues
GitHub Launches Stacked PRs as a First-Party Workflow
GitHub shipped Stacked PRs in private preview this week — a first-party CLI and UI for managing stacked pull requests. The workflow formalizes a pattern that was previously only practical for teams with dedicated internal tooling.
Help Agents Help Themselves: Orchestration Patterns That Make Failure Cheap
Niels Bantilan's talk at Coding Agents 2026 reframes the agentic reliability problem: we obsess over semantic correctness, but production failures often live in the infrastructure layer. Here's what durable agent design actually looks like.
The Claude Code Source Leak Revealed More About Us Than Anthropic
Anthropic's entire Claude Code source leaked via npm source maps, and I argue they should open-source it. Meanwhile, the industry's cries of entitlement to horror at "AI slop" under the hood reveal more about our perceptions of crafting software than they do about Anthropic's engineering practices.
Sub-Agents, Not Swarms: What Pinterest Learned Productionizing LLM Training With Coding Agents
The question of how to architect multi-agent systems for production workloads doesn't have a settled answer yet. Pinterest's Growth AI team tried both sub-agents and swarms for LLM post-training, and their results challenge some common assumptions.
Evals Are How Serious AI Practitioners Ship Without Vibes
There's a gap between teams that ship AI features based on gut feel and teams that actually know whether their system is working. Most teams are on the wrong side of it and don't realize it yet.
Why I Formed My New Consultancy, Limbic Systems
Last year, I helped a boutique consulting agency save 20+ hours per project and fundamentally change how they sell. I realized our approach and nearly everything we built over that 9-month period could help any service business become AI-native.
The Reports of MCP's Death Are Greatly Exaggerated
Perplexity CTO says they dropped MCP for "traditional" APIs and CLIs, and many claim to be following suit. Meanwhile, the protocol surpassed 97M monthly SDK downloads, over 17,000+ registered servers, and wide adoption from OpenAI, Google, Microsoft, and AWS. One of these narratives is wrong.
Anthropic Announces Agentic Code Reviews for Claude Code
Anthropic's new team-based review system that dispatches targeted agents on every PR to catch the bugs that skims miss, built for depth, not speed.
The Necessary Evolution of "Research, Plan, Implement" as an Agentic Practice in 2026
"Vibe coding" gave way to RPI, and now RPI is evolving into something more rigorous. Dexter Horthy's QRSPI breaks down where the original Research, Plan, Implement pattern fails in the real world and shows what a more disciplined agentic workflow looks like in 2026.