Case Study: OpenAI achieves scalable multi-agent orchestration with Firecrawl

OpenAI boosts pull requests by 500% with Firecrawl

OpenAI faced a challenge scaling their agentic coding workflows, as their engineers found they could only comfortably manage three to five concurrent agent sessions before context switching became a bottleneck. This human ceiling limited their ability to scale parallel work effectively. To overcome this, they implemented multi-agent orchestration using their own product, OpenAI Codex, and integrated Firecrawl's web scraping service via an MCP server.

The solution involved using Codex to spawn a team of specialized, parallel agents for code review, with one agent utilizing Firecrawl to verify third-party APIs against live web documentation. This orchestration allowed one human to manage the output of multiple agents working simultaneously. OpenAI reported a concrete 500% increase in landed pull requests on some teams, significantly raising their productivity ceiling by shifting the engineer's role from active coding to review and direction.

OpenAI