Evan Boyle works on GitHub Copilot agents at Microsoft and founded GenSX, an open-source framework and cloud platform for building agent applications. He specializes in making long-running software operations observable, resilient, and accessible to application developers.
Boyle previously worked at Microsoft, Convoy, and Amazon before becoming an early employee and director of engineering at Pulumi. There, he introduced the Pulumi Automation API, which lets applications provision and manage cloud infrastructure directly through familiar programming languages, and contributed to Pulumi Deployments, extending infrastructure automation into remotely executed, API-driven operations.
He subsequently founded GenSX, creating an open-source TypeScript framework organized around React-inspired agent components: reusable, independently testable functions assembled into observable workflows. In 2025, GenSX Cloud added a full Node.js runtime, executions lasting up to 60 minutes, workflow tracing, and storage that agents can provision while running. Boyle later joined Microsoft with the GenSX team to work on GitHub Copilot agents.
- Build infrastructure for agent time. Model latency, provider outages, rate limits, and background data ingestion routinely outlast conventional HTTP requests. Boyle treats durable execution and resumability as product requirements: users should retain progress when connections drop, pages refresh, or workflows encounter errors.
- Keep agent components transparent. GenSX components preserve direct access to model APIs while adding component-level retries, error boundaries, caching, and traces. Developers can inspect prompts, token usage, and intermediate results without burying agent behavior beneath rigid orchestration abstractions.
- Separate requests from execution. His agent-infrastructure architecture uses decoupled API and compute layers, with Redis Streams retaining status updates and output. Worker heartbeats support failure detection and restarts, clients can reconnect without losing context, and compute scales independently of request handling.
- Constrain ambitious agent experiments. Boyle also built self-modifying code agents capable of changing their own source and launching new instances, while emphasizing sandboxing, controlled command execution, and protection against runaway loops and model costs.
