GitHub made Dynamic Workflows available in public preview across Copilot CLI, the Copilot app, and SDK: code now defines repeatable stages, sequencing/parallelism, structured handoffs, subagent cross-checks, and human checkpoints while agents handle judgment-heavy steps. It is a useful middle ground between a fixed script and free-form delegation; surfaced in this pass as a concrete orchestration/reliability capability. Source: https://github.blog/changelog/2026-10-01-dynamic-workflows-in-copilot-cli-and-the-copilot-app
GitHub Copilot computer use entered public preview in CLI and the Copilot app on macOS and Windows, allowing GUI interaction across apps without APIs/MCP; users approve control by default, can set trusted apps, and organizations can disable it. This extends Microsoft’s coding agent into desktop workflows, with the usual preview caveat. Source: https://github.blog/changelog/2026-10-01-github-copilot-can-now-interact-with-desktop-apps/
Copilot code review can now be requested through REST and GraphQL APIs with per-review effort selection, and Balanced became the default for repositories/orgs still set to Default. This makes review automation easier to wire into external systems; available across Pro, Pro+, Max, Business, and Enterprise. Source: https://github.blog/changelog/2026-10-02-copilot-code-review-api-support-and-new-default-effort-level/
DeepSeek put its open-source, plugin-composable Harness into worldwide public preview as a desktop app for macOS and Windows, supporting background work, local file edits, coding, research, and everyday tasks. It drew 387 HN points and 209 comments, making it a notable harness/platform discovery beyond the model-release lane. Caveat from the HN thread: users report desktop telemetry enabled by default and debate plugin-centric maintenance; these are community reports, not an independent audit. Sources: https://www.deepseek.com/en/harness/ ; https://github.com/deepseek-ai/deepseek-harness ; https://news.ycombinator.com/item?id=49929489
Perplexity launched Computer Automations for agents that run on schedules or conditional events from Slack, Gmail, Outlook, Linear, or GitHub, carrying forward prior-run memory and tracking unresolved expectations. Admins can scope connectors/actions and users can review run history; trigger monitoring itself does not consume credits. This is a newly surfaced long-running/event-driven workflow pattern, not merely a recurring prompt. Source: https://www.perplexity.ai/en-GB/hub/blog/computer-adds-automations-for-ongoing-work