Anthropic’s Boris Cherny, creator of Claude Code, says there are days he manages tens of thousands of AI agents at once. Cherny describes Claude Code running nested agent hierarchies that automate coding and pursue recursive improvement, revealing the operational patterns and safety trade-offs of massive agent fleets. Outcome engineers must treat orchestration, telemetry, and safety as core infra problems rather than research curiosities — Principle 09.

Datadog launches 100+ features at DASH to push autonomous AI ops. Datadog expands Bits AI agents to run ops across the SDLC and embeds observability controls, turning agent behavior into first-class signals you can monitor and govern. If you build agentic systems, integrate observability and circuit breakers early so operators can read outcomes and intervene — Principle 09.

Tiger Data launches PostgreSQL extension designed for AI agents. Ghost exposes a Postgres extension and managed service that lets autonomous agents read, write, and coordinate state safely at scale, shifting where you hold context and truth. Design your agent state model, audit trails, and consistency guarantees around these primitives to keep outcomes legible and auditable — Principles 06 and 11.

Zscaler launches AI Broker and Endpoint AI Security for AI agents. Zscaler positions zero-trust controls and runtime enforcement as foundational for protecting autonomous agents across networks and endpoints, including policy, TLS, and identity for agent traffic. Treat agent identity and policy enforcement as part of your security architecture so agents can be both autonomous and constrained — Principles 10 and 14.

Harness-1: Open-source 20B search agent outperforms GPT-5.4 on recall. Harness-1 is an Apache-2.0 20B search agent that beats GPT-5.4 on factual retrieval, proving smaller open models can win on recall with the right retrieval and context engineering. Outcome engineers should re-evaluate where closed large models are necessary versus tuned open models for retrieval-heavy agent tasks and design your vector and retrieval stack accordingly — Principles 06 and 11.