Honeycomb ships Personal and Team AI SRE agents (early access)

What changed: Honeycomb announced Personal and Team Agents that run goal-oriented investigations against high-cardinality telemetry, remember findings across runs, and can act only within team-set limits; the product went into early access on November 3, 2026.

Why it matters: Honeycomb’s agents are explicitly built to use production traces and observability queries as their data source, so they’re designed to catch regressions and instrumentation gaps earlier (before alerts hit customers) — a useful pattern if you want agents that surface reproducible evidence rather than speculative fixes.

Try/watch: Start by giving a Team Agent a narrow, measurable goal (watch this release for regressions) and require evidence links to traces and the originating query. Monitor false positives and how much human review the agent’s recommendations need before you widen its remit.

CrowdStrike adds an “AI Kill Switch” and related agent-security moves

What changed: CrowdStrike introduced an AI Kill Switch in its Falcon Guardian product on November 3, 2026 — administrators can terminate a misbehaving agent where it runs, block it from relaunching, and cut its network access; the company also published a set of same-day press items describing additional agent-observability and security integrations.

Why it matters: Security controls that let defenders stop an agent in-flight are a practical precondition for broader agent adoption: buyers and operators can demand a runtime safety mechanism before they permit agents deeper access to systems or data, and incident playbooks should now include kill-switch steps.

Try/watch: Add the Kill Switch (or equivalent) to your acceptance checklist for any agent with system access, and validate that audit trails record who triggered it and why. Watch vendor claims about “automatic remediation” and require rollback/replay tests so kill actions don’t obscure root-cause evidence.