Engineers are becoming sorcerers | The future of software development with OpenAI’s Sherwin Wu. OpenAI describes Codex and AgentKit enabling engineers to run fleets of AI agents that shrink code review times and amplify individual productivity. Outcome engineers should treat agent orchestration as a delivery layer and plan for new coordination patterns and guardrails (Principle 09).

Claude Opus 4.6 vs. GPT-5.3 Codex: How I shipped 93,000 lines of code in 5 days. Lenny recounts combining Codex’s review strengths with Opus’s creative coding to produce 93k lines and 44 PRs in five days. That pattern shows hybrid-model pipelines and artifact-centric CI are necessary to verify and ship agent-generated code (Principles 08 and 09).

State Department is gearing up to ‘roll out agentic AI’. The U.S. State Department is preparing to deploy autonomous agents like StateChat across operations to democratize access and cut administrative friction. Large-scale deployments force you to solve governance, onboarding, and reliability at enterprise scale — design for operability and clear human-in-the-loop rules (Principles 04 and 09).

CAISI Issues Request for Information About Securing AI Agent Systems. NIST’s CAISI is soliciting input to define security standards and practices specifically for AI agent systems. Those standards will set expectations for threat modeling, runtime restrictions, and publisher obligations — integrate security and auditing into agent design now (Principles 14 and 10).

Picogrid wins $9M Air Force contract for counter-drone software written by AI. Picogrid used AI to generate translator modules that slashed integration time from weeks to hours and won a $9.3M contract. That’s a concrete example of AI-produced artifacts becoming procurement-grade outputs — prioritize reproducible artifacts, validation tests, and integration contracts (Principles 08 and 07).