Weekly signal

This week (Aug 31–Sep 8, 2026) the agentic stack for business automation continued its shift from experimentation to operational controls: cloud platforms tightened billing and FinOps controls for always‑on agents, enterprise tooling added environment- and agent-level governance, and data/AI platforms shipped scheduled and runtime features that make agents practical for recurring operational work.

What changed

1. Copilot Studio agents built with the GitHub Copilot harness move out of preview billing — developer and trial environments begin consumption-based Copilot Credits on September 1, 2026. Admins must now treat previously “free” dev/trial agents as billable runtime workloads.

2. Power Platform added a tenant-level cost control rule ("Draw from tenant credit pool") that reached GA in late August 2026 — admins can block environments from drawing shared tenant credits after their allocation is exhausted. This gives a straightforward enforcement knob for Copilot/Copilot Studio FinOps.

3. Databricks released platform updates (late Aug / Sept release notes) that make AI-driven automation more production-ready for scheduled and long-running tasks: Genie Code scheduled tasks are now GA and AI Runtime/Genie Code features support scheduled runs and job-style automation. That lowers friction for turning agent output into recurring operational jobs.

4. Microsoft Agent Framework published agent-and-workflow “channels” (Python) to make agents reachable from multiple surfaces (Responses, Telegram, MCP, A2A), improving integration patterns for operational automation across apps and chat surfaces. This helps teams move agents from prototypes into the flow of work.

What to do with it

1. Inventory and owner assignment: identify all Copilot‑harness agents (use the Harness column in Power Platform inventory), assign a business owner, and document expected monthly activity before Sept 1.

2. Apply FinOps controls: allocate prepaid Copilot Credits to environments, set per‑agent monthly limits and stop‑on‑limit behavior, and link high‑usage environments to pay‑as‑you‑go plans with Azure budgets and alerts. Use the Power Platform admin APIs to apply rules at scale.

3. Schedule and test automation: when migrating prototypes to production, prefer scheduled Genie Code jobs or task-style runtimes (Databricks) for reliable, auditable automation rather than ad‑hoc agent previews. Instrument runbooks and SLOs for agent jobs.

4. Integrate with channels and guardrails: expose agents only to required channels, add approval gates for write actions, and log decision traces. Use the Agent Framework channel packages to standardize how agents are invoked and monitored.

Sources: Microsoft Power Platform — Manage costs for GitHub Copilot harness; Microsoft MC / Power Platform cost-control announcement; Databricks product release notes; Microsoft Agent Framework devblog.