Richard Seroter goes hands-on with agent platforms, mapping 9 jobs to Google's Gemini stack
rseroter · x · 2026-10-16
Richard Seroter (Google Cloud) published a hands-on deep dive, "Your agents need a platform," building a customer-support + refunds agent pair on Google's Gemini platform and mapping what an agent platform must do.
Nine platform jobs-to-be-done:
- Build & enrich agents: Python ADK plus Antigravity-assisted phased development
- Use & route models: API Gateway model routing, Model Garden (Gemini 4, Claude, Gemma)
- Store & access context: short-term Sessions, long-term Memory Bank, Knowledge Catalog, Agent Search/RAG, AlloyDB
- Connect tools: Google MCP Servers, MCP Toolbox, connectors, Apigee
- Host & scale: Agent Runtime, Cloud Run, GKE, Managed Agents API
- Coordinate agents: A2A protocol, Eventarc, Workflows
- Constrain & govern: Agent Identity, Cloud IAM, Agent Gateway, Model Armor, VPC-SC
- Interact: Gemini Enterprise, Live API, A2UI
- Observe & evaluate: Cloud Trace, Agent Platform Evals, Prompt Optimizer
Core argument: existing monitoring, data, and IAM systems are reusable, but agents' distinct needs — state management, governance boundaries, multi-agent orchestration — call for a dedicated platform layer. The post includes a full Gemini platform architecture diagram.
More from coding & agent
- Microsoft paper: history-aware model routing lifts multi-turn agent task pass rate by 3.3 points — dair_ai · 2026-10-16
- ORA: open robot-use agents powered by DeepSeek V4.1-Flash compete with frontier models — JunGao33210520 · 2026-10-16
- Non-developer ships a full platform fighter in 30 days with AI agents — Teknium · 2026-10-16
- 2 humans + 5 agents: plain-English feature requests now drive their codebase autonomously — nptacek · 2026-10-16
- Berkeley paper: making research agents forget boosts automated discovery by up to 39% — omarsar0 · 2026-10-16
- Ptacek: voice-chat agents turn a plain-English feature request into reviewed code — nptacek · 2026-10-16