| ADK |
Developer toolkit for building and deploying agents — CLI, SDKs, local dev environment |
Jenna Winkler, Matous Havlena, Tomas Weiss + team |
Active |
| Agent Lifecycle Management |
Managing agent lifecycle on the platform via the operator |
Paolo Dettori, Jarek Cwiklik |
Active |
| Agent Runtime |
What happens when an agent executes — guardrails, evals, context routing |
Mariusz Sabath, Paolo Dettori |
Active |
| Automation |
CI/CD, testing automation, and developer workflow tooling |
Jim Cadden, Gloire Rubambiza |
Active |
| Benchmarking / Testing |
Agent benchmarking and test infrastructure |
Kelly Abuelsaad |
Active |
| Event-Based Agents |
Event-driven agent patterns and async multi-agent communication |
Aleksander Slominski |
Early stage |
| Fault Tolerance & Context Management |
Handling agent failures and managing context across sessions |
Ed Snible, Hai Huang |
Early stage |
| Humr |
AI harness runtime for Claude Code, Codex, and Gemini CLI |
Matous Havlena, Tomas Weiss |
New |
| MCP Gateway |
Unified front door for MCP servers — tool routing, auth, token exchange |
Hai Huang, Ed Snible |
Active |
| Memory |
Persistent agent memory — long-term vision, no active plan yet |
— |
Early stage |
| Observability & Token Management |
Tracing, token cost attribution, and auto-instrumentation |
Evaline Ju, Ilya Kolchinsky |
Active |
| Resiliency & Consistency |
Ensuring environment consistency and preventing corruption from agent actions |
Sahil Suneja |
Early stage |
| Sandboxing |
Safe execution environments for code dev, AI-generated code, and MCP tools |
Ladislav Smola |
Active |
| Workload Identity and Security |
Zero-trust auth and identity management for agents and users |
Mariusz Sabath, Maia Iyer, Alan Cha, Morgan Foster, Akram Ben Aissi |
Active |