OpenAI says “chat is dead” and plans to rebuild ChatGPT as a full-blown agent app. OpenAI announces rebuilding ChatGPT into an agent-driven superapp that runs autonomous agents and integrates partner apps. Outcome engineers should treat this as a signal to design orchestration layers, capability contracts, and partner integration patterns now — this is Agentic Coordination made product (Principle 09).

Perplexity’s “Search as Code” lets AI models write their own search pipelines instead of calling fixed APIs. Perplexity ships a sandbox where models generate Python search pipelines, improving accuracy and cutting token costs. That changes retrieval engineering: expect agents to author and evolve their own pipelines, so build safe sandboxes, reproducible pipelines, and tooling to test retrieval behaviors (Principles 06 & 07).

Her · हेर — a detective for your Claude Code sessions. Her reconstructs Claude Code session traces, flags risky actions, and explains agent behavior deterministically without leaving your private runtime. Outcome engineers gain a template for session forensics and local observability that supports incident investigation and automated auditing of agent decisions (Principles 06 & 14).

A Case for Simulation-Driven Resilience in Agentic Data Systems. The authors argue simulation-driven testing uncovers metastable failures from agent workloads and prevents production incidents. Treat simulation suites as first-class testing for agentic stacks — you need models of workload interactions, failure modes, and recovery drills to keep agent systems stable (Principles 07 & 14).

My automated doubt development process. The author describes a multi-agent ‘doubt’ audit that scrutinizes specs and code early to rebuild trust in AI-assisted development. This pattern points to embedding automated adversarial review and human checkpoints into CI/CD pipelines so agents deliver verifiable, auditable artifacts rather than opaque outputs (Principles 03 & 14).