Agents: Secrets, Memory, Search, Telemetry, and Lightning

Can Agents Keep a Secret? 1Password CTO Nancy Wang on Agentic Security argues just-in-time secrets let agents access infrastructure securely, replacing custody with “access without custody” and minimizing long-lived credentials. Outcome engineers must bake ephemeral credential issuance, tight audit hooks, and automated rotation into agent pipelines to reduce blast radius and satisfy gating and compliance.

Anthropic gives chat and Cowork one memory unifies Claude’s memories across chat and Cowork, syncing context in real time while excluding sensitive data by default. This changes how you design context propagation and consent — treat shared memory as a first-class context layer and build explicit opt-outs, sanitizers, and validators.

Keenable raises $26M seed to build web search index for AI agents launches an agent‑optimized web index and API to deliver retrieval tailored for agent workflows. Outcome engineers should evaluate agent-native search as a context source and integrate it into your graph and retrieval pipelines to lower latency and improve grounding.

How telemetry pipelines keep AI agent costs under control explains rising agent telemetry is exploding observability bills and urges redesign of telemetry pipelines. Practitioners must adopt sampling, rollups, schema-driven logs, and retention policies so traces scale with your agents without bankrupting production.

Microsoft releases Agent Lightning v1.0 — why it matters for platform engineers ships a harnessed‑RL platform that controls agent-environment loops and stabilizes training across service boundaries. Platform teams can adopt the harness pattern to build repeatable CI, safe exploration, and reproducible evaluation for agent deployments.