Agent platforms take on more of the execution stack

What happened

Azure adds managed agent execution

On 28 October, Microsoft introduced the Copilot harness in Foundry Agent Service in public preview, combining reusable skills, connected tools and source-controlled agent definitions. Separately, long-running resilience for hosted agents became generally available. Application code must still save progress and prevent duplicate actions after a restart.

Confidence: high in announced availability and documented behaviour. Primary vendor source: Microsoft announcement, 28 October 2026.

GitHub Agentic Workflows reaches general availability

On 28 October, GitHub made Agentic Workflows generally available. Teams define reasoning-based automation in Markdown, compiled into GitHub Actions. Safeguards include read-only permissions by default, sandboxing and validation of proposed changes.

Confidence: high in announced availability; outcomes depend on the workflow and configuration. Primary vendor source: GitHub release, 28 October 2026.

Research challenges assumptions about human oversight

In research published on 27 October, 50 professional analysts using Claude Code followed deliberately assigned agent preferences on 86% of analytical choices. The experiment covered one task and artificially strengthened the agents’ preferences. It does not establish an equivalent rate in production.

Confidence: high in the reported finding; moderate in broader applicability. Primary research source: Anthropic study and limitations, 27 October 2026.

Google expands governed business context for agents

On 29 October, Google announced general availability of expanded semantic-model support in Knowledge Catalog. It connects definitions from systems including dbt, Snowflake, Databricks and SAP with lineage and quality metadata. Conflicting definitions retain their provenance.

Confidence: high in announced availability. Performance gains remain vendor-reported. Primary vendor source: Google announcement, 29 October 2026.

Why it matters

Interpretation and key rationale for enterprise delivery teams.

• AI-DLC and repository governance: GitHub’s release offers a concrete option for generating traceability artefacts, analysing failed CI and proposing documentation changes across service repositories. Review and approval rules still need to be explicit.
• Azure architecture: Managed execution could reduce orchestration work, but agents that update business systems still need checkpoints and protection against repeated writes after restarts. Keep those responsibilities clear at service boundaries.
• Governance: Human sign-off alone may provide weak evidence of independent judgment. Record alternatives considered, supporting evidence and the reviewer’s rationale.
• Reliable business outcomes: Approved definitions of operational and financial metrics deserve versioning alongside golden datasets. Better models cannot resolve conflicting business rules by themselves.