Weekly signal

This week (Oct 26–Nov 3, 2026) the agentic automation market moved from build-and-experiment to scale-and-control. Major vendors shipped governance and trust tooling while industry-focused vendors published production agents for supply chain, healthcare and security operations. The practical story: organizations can now buy, run and govern fleets of business automation agents — but you must add runtime evals, identity binding, cost controls and detection to avoid liability, fraud and runaway spend.

What changed

1. Cross‑platform agent governance: Salesforce expanded Agent Fabric with Agent Evals (continuous testing/evaluation), API Evals (automatic API change checks), a kill‑switch and hard spending controls for agent fleets. The feature set is explicitly aimed at preventing silent failures and runaway costs as agents scale.

2. Agent identity & trust: Socure launched “Agent Intelligence” (Agent Detection GA, Agent TrustScore in beta, Agent Credential planned Q1 2027) to detect agent traffic, score agents by identity/authorization/risk, and issue verifiable credentials that tie an agent back to a human or business principal. This is positioned as a KYA (Know Your Agent) capability for commerce and risk systems.

3. Behavioral agent detection: Resemble released RAD‑S1 / Resemble Agent Detection, a behavioral model that claims high detection of agents that disguise automation (internal benchmarks show ~98.8% detection vs <15% for conventional bot tools on disguised agents). It targets login, checkout and account actions.

4. Sector readiness & marketplaces: Appian secured IL5 provisional authorization for its Defense Cloud (US Department of War) to run agentic process automation in mission contexts; Peloton Consulting published three supply‑chain agents on Oracle Marketplace; Evidently launched “Evidently Anywhere” to push patient‑context agents into EHRs. Those moves show vertical, compliance and marketplace channels maturing for agentic automation.

5. Agentic security ops: Arctic Wolf published a report showing their Aurora agentic SOC resolves the majority of investigations with AI-human collaboration — evidence agentic automation is operational in security at scale.

What to do with it

• Treat agent projects as platforms: add runtime Evals, budget/hard limits, and a kill switch before scaling. Start with Canary agents and synthetic tests tied to Agent Evals or CI.
• Add KYA and detection: deploy behavioral agent detection and an identity/trust layer (or pilot Socure/Resemble) before exposing sensitive customer flows. Label traffic first; don’t block by default.
• Use marketplaces & vertical agents to speed value capture: evaluate prebuilt agents (Oracle Marketplace, Evidently) for short time‑to‑value but insist on observability, data residency and governance.
• Security and compliance: for regulated or high‑impact workloads (defense, health, finance) require IL‑level or equivalent authorizations, audit trails, and human‑in‑the‑loop escalation.

Sources: see numbered list below.