Weekly signal

This week (Oct 5–13, 2026) accelerated the operationalization of agentic business automation: vendors and platform providers moved from building blocks and previews into security, governance, and acquisition activity you should treat as immediate program requirements.

What changed

• SailPoint (US) launched an identity-first approach to agent governance at Navigate 2026: Autonomous Agents (agent-side defenders), Autonomous Identity Security Posture Management (A-ISPM), agent audit/export capabilities, and Just‑In‑Time privilege controls to manage AI agents at machine speed. Availability begins rolling out globally from Navigate with some features scheduled for Q4 FY27.

• OpenAI updated ChatGPT Business/Enterprise release notes (Oct 7–8): workspace agents gained admin controls (per-app action permissions, activity visibility in the admin console), speech/audio output for agents, new model options, and workspace app templates — making ChatGPT workspace agents easier to connect to enterprise apps while giving admins explicit controls.

• UiPath publicized analyst recognition (Oct 9) after expanding agentic testing and orchestration capabilities—positioning their Test Cloud and agentic orchestration stack as production-ready for enterprise QA and process automation. The message: agentic testing + governance is becoming a mainstream purchase category.

• Automation Anywhere moved on two fronts (Oct 7): announced an acquisition of Boost.ai (conversational agents) and shipped Automation 360 v.42 with explicit Model Context Protocol (MCP) support so its automations and AI Agents can interoperate with MCP toolsets and external agent runtimes.

• IBM published actionable guidance and pattern articles about Model Context Protocol (MCP) architectures and how to use watsonx Orchestrate / MCP servers to safely expose business automation capabilities to AI agents—practical guidance for integrating agents with workflows, decisions and content in production.

What to do with it

1. Immediately inventory agent entry points (MCP servers, connectors, ChatGPT agents, orchestration endpoints) and map them to identity owners — treat these like production APIs.

2. Turn on admin controls for workspace agents and restrict per-app action scopes while you pilot (ChatGPT Business/Enterprise). Capture admin logs and activity.

3. Update access policies to short‑lived/JIT credentials for agents; integrate agent lifecycle events into identity tooling and SIEM. Pilot SailPoint Autonomous Agents where identity/privilege risk is highest.

4. If you run production QA or test automation, evaluate agentic testing platforms (UiPath Test Cloud) and add agent performance & evaluation gates to CI/CD.

5. Standardize on MCP or A2A-compatible connectors for new agent integrations and require vendor MCP compatibility (Automation Anywhere, IBM, Microsoft, others are converging on MCP).