“Until now, building agents meant juggling fragmented tools: complex orchestration with no versioning, custom connectors, manual eval pipelines, prompt tuning, and weeks of frontend work before launch.”
— OpenAI, announcing AgentKit