Microsoft Agent Framework finished another important stability step with Declarative Workflows 1.0: orchestration now moves out of app code and into YAML-defined workflows, with Python agent-framework-declarative at 1.0.0 and the .NET workflow package stable too. Sources: Microsoft Agent Framework post ; Microsoft Agent Framework tag page
The harness release is the more practical companion to that workflow work. Microsoft says the released harness bundles planning, tool orchestration, approvals, memory, telemetry, and safer file handling into the framework; HN also picked it up today. Sources: Microsoft Agent Framework: The Harness is now released ; HN thread
GitHub shipped a useful cluster of agent-surface updates on Jul. 23-24: Copilot cloud agent for Linear is GA, GitHub MCP Server now tracks the next MCP spec, and agent automation controls in Issues are in public preview with reasoning, confidence, and optional human review. Sources: Copilot cloud agent for Linear ; GitHub MCP Server next spec ; Agent automation controls
GitHub Copilot added Gemini 3.6 Flash, and the release note is explicitly agentic: configurable reasoning effort, parallel tool use, and better task-completion and token efficiency across coding workflows. HN and Reddit both surfaced it quickly, which is a decent signal that people are actually trying it. Sources: GitHub changelog ; HN ; Reddit
AutoGen is now officially in maintenance mode, and the July 2026 GitHub discussion is a meaningful repo-level signal rather than chatter: Microsoft tells new users to start with Agent Framework, while the discussion cites 50.4k stars, 98 releases, 3,776 commits, and 559 contributors. Sources: AutoGen repo ; AutoGen update discussion
OpenAI’s latest eval/safety posts are worth watching together. Separating signal from noise in coding evaluations says roughly 30% of SWE-Bench Pro tasks are broken, while Safety and alignment in an era of long-horizon models says long-running internal agent tests surfaced new failure modes and pushed trajectory-level monitoring plus rollback controls. Sources: OpenAI coding evals ; HN ; OpenAI long-horizon safety ; HN ; Reddit