Descriptions:
Liam Hampton, an engineer at Microsoft, demonstrates how to run multiple GitHub Copilot agent types simultaneously within VS Code at AI Engineer 2026, making a practical case for how parallel agent workflows can address the ROI questions businesses are raising about AI tooling investment. The talk distinguishes three agent modes and explains when each is appropriate: local agents for hands-on, high-context work like test writing; background agents for semi-autonomous tasks like frontend generation; and cloud agents for fully delegated work like documentation.
The live demo runs all three modes concurrently on the same Python application. The local agent—configured with Claude Opus 4.6 at medium reasoning—writes unit tests interactively and iterates on error handling with direct developer input. The background agent uses GitHub’s Autopilot feature and Git worktrees to build a new frontend in an isolated branch, creating a pull request and pausing for local testing. The cloud agent handles open-source project scaffolding (README, contribution guidelines) with no developer involvement.
Hampton’s framing is deliberately business-facing: agents are most valuable not when they replace developers entirely, but when they absorb adjacent tasks so engineers can stay in meaningful work. The session also briefly touches on token efficiency strategies emerging in the community, noting that developers are finding creative approaches—including unusual prompt formatting—to reduce spend. The demo is notable for showing parallel agent coordination as a practical, tested workflow rather than a proof of concept.
📺 Source: AI Engineer · Published May 21, 2026
🏷️ Format: Tutorial Demo
