A multi-agent AI system, from concept to product
Architecting an autonomous multi-agent system on Google's Agent Development Kit, Gemini and custom MCP servers, and taking it from concept to internal pilots and external products.
Concept to internal pilots and external products, with end-to-end ownership of the system.
We worked with Product and UX teams at Google on a new AI-powered product: an intelligent chat experience driven by agents that reason, plan and act on their own.
The product itself is confidential, so this case study focuses on the engineering.
Build an intelligent chat experience powered by AI agents that could reason and plan autonomously, and make it robust enough to move beyond demos into internal pilots and, ultimately, external products.
What we did
We architected a multi-agent system using Google's Agent Development Kit (ADK), Gemini and custom MCP servers, designing the prompts, agent orchestration and tool-calling that let the agents reason, plan and act.
We worked closely with Product and UX teams to define value propositions, deliver stakeholder updates and iterate quickly on feedback, starting as part of a team and later owning the system end to end.
We built the Python and TypeScript APIs behind the system and deployed them to Google Cloud Platform.
Results
- Taken from concept to internal pilots and external products
- Autonomous reasoning, planning and tool use in production
- System owned end to end
Work
More work
Book a call
Have a workflow you'd like to automate?
Tell us where your team loses the most time. If AI isn't the right fit, we'll say so.