Five practical shifts in agentic engineering

GitHub Issues become an agent coordination protocol. They provide a persistent, auditable layer for agents working across models, machines, and sessions. That makes Principle 09 — Agentic Coordination is a New Org concrete: coordination artifacts matter as much as model calls.

Advanced evals expose hidden AI failures. The approach turns vague product-quality judgments into repeatable tests that guide iteration. Outcome engineers need this Principle 16 — Audit the Outcomes discipline to distinguish useful behavior from plausible demos.

VS Code 1.138 brings agent sessions to Dev Containers. Agents can run in isolated development environments with Codex interoperability and reusable workflow automation. This advances Principle 07 — Build the Island by making repeatable, disposable execution infrastructure part of the coding loop.

Fixing agent memory. Persistent memory can preserve failures and malicious instructions, so inspection, provenance, and safety checks must precede reuse. That is Principle 02 — Ground Truth and Principle 14 — The Immune System applied to context rather than just model output.

Markdown in /src. Treating specifications as versioned source artifacts gives agents a reviewable basis from which to derive code and tests. It strengthens Principle 13 — The Documentation by making intent, implementation, and verification easier to connect.