LastMile AI

LastMile AI built developer tools for creating, evaluating and deploying AI agents. Its mcp-agent Python framework uses Model Context Protocol to connect agents with tools and compose workflows such as routing, map-reduce and evaluation. The accompanying mcp-c tooling supports managed cloud deployment, including publishing agents as MCP servers. Its evaluation platform, AutoEval, lets developers label application data, fine-tune custom evaluators, measure response quality and build guardrails for LLM applications.

Founded in 2023, LastMile AI’s cofounders include CEO Sarmad Qadri, CPO and president Andrew Hoh, and Suyog Sonwalkar, who served as CTO in 2023–2024. Its engineering approach combines reusable agent workflows with Temporal for durable execution and OpenTelemetry for observability. For evaluation, its alBERTa family includes a 400-million-parameter entailment model that developers can fine-tune for application-specific criteria, including checking whether responses remain faithful to supplied context.

The LastMile AI team joined Meta and is now part of the Product and Applied Research organization at Meta Superintelligence Labs, building agents for executives and knowledge workers. This continues its work toward a cognitive computer: interconnected cloud agents designed to operate over days, weeks or months. The focus extends beyond individual agent interactions to long-running workflows involving other agents, people and authorized tools.