Agentic AICon Recap: Agent Infrastructure and AgentOps Insights

Agentic AICon, a dedicated session on agent infrastructure and AgentOps, concluded recently with more than 180 technology practitioners in attendance. The conference centered on the challenges of scaling agents in production and offered a comprehensive walkthrough of the Agent Infra solution that spans building, deploying, governing, operating, and optimizing agents.

Agenda 1 – Accelerating Enterprise Agent Engineering from Build to Scale

Alibaba Cloud introduced an end‑to‑end Agent Infra solution that integrates four core products—AgentRun, AgentTeams (HiClaw), AgentLoop, and STAROps—to address pain points such as multi‑agent collaboration, cost control, and effectiveness evaluation, thereby bridging the gap between prototype and production.

Agenda 2 – HiClaw: Multi‑Agent Collaboration and Governance

HiClaw is an enterprise‑grade platform that adopts a Manager‑Team TL‑Workers hierarchy and Matrix protocol to enable transparent coordination. It uses a unified AI gateway for authentication and traffic control, combines asset management with full‑stack observability, and mitigates key leakage and cost‑hole issues while supporting mixed‑engine workloads from single‑person setups to digital federations.

Agenda 3 – From Demo to Production with AgentRun

AgentRun serves as a serverless‑based runtime for multi‑agent coordination, tackling long‑running tasks, massive tool invocations, and security isolation. Its sandboxed environment, session affinity, and elastic scaling transform complex AI inference into stable, low‑cost cloud services, demonstrated by customers such as DeepForce Technology and MiaoWu.

Agenda 4 – AgentLoop: Enabling an Evolutionary Data Flywheel for Agents

AgentLoop addresses three enterprise challenges—opaque effectiveness, uncontrolled changes, and difficult optimization—by providing non‑intrusive observation and a continuous evaluation‑optimization loop. It introduces an “Agent‑as‑Judge” mechanism that replaces manual or LLM‑based checks, delivering real‑time quality degradation alerts and a data‑driven evolution cycle for self‑reflective agents.

Agenda 5 – STAROps: Full‑Scope Intelligent Operations Platform

STAROps leverages a unified UModel to convert logs, metrics, and topology data into AI‑readable world models. With a Mission long‑running task framework and a secure ToolService, STAROps agents perform 24/7 autonomous inspection, root‑cause analysis, and self‑healing, advancing operations from reactive to proactive Agentic Ops.

Agenda 6 – Asynchronous Collaboration with RocketMQ

The RocketMQ‑based asynchronous Agent session gateway tackles message backlog and blocking in AI applications. By introducing a lightweight LiteTopic carrier, differentiated subscriptions, physical isolation, and precise flow control, it eliminates head‑of‑line blockage, supports millions of queues, and ensures stable scheduling for high‑concurrency multi‑agent collaborations.