Agent Runs now visualize subagent activity for eve projects

Post LinkedIn

▲Read original on Vercel News

#multi-agent #debugging #observabilityagent-runsagent runs eve

💡Debug complex multi-agent workflows with new granular visibility into subagent performance and costs.

⚡ 30-Second TL;DR

What Changed

New Subagents tab organizes activity by execution turn

Why It Matters

This visibility significantly improves debugging complex multi-agent systems, allowing developers to pinpoint exactly where sub-tasks fail or consume excessive tokens.

What To Do Next

Open your eve project in Agent Runs to inspect the new Subagents tab and audit your sub-agent token consumption.

Who should care:Developers & AI Engineers

Key Points

•New Subagents tab organizes activity by execution turn
•Shared timeline view for tracking subagent prompts and duration
•Drill-down capability for turns, tool calls, metadata, and token usage
•Direct visibility into subagent failures within the parent run

🧠 Deep Insight

AI-generated analysis for this event.

🔑 Enhanced Key Takeaways

•The 'eve' project framework is Vercel's specialized architecture for managing multi-agent orchestration, designed to handle complex, recursive task delegation.
•This update integrates with Vercel's observability suite, allowing developers to correlate subagent performance metrics directly with Vercel AI SDK telemetry data.
•The subagent visualization utilizes a hierarchical tree structure to represent parent-child relationships, enabling users to trace the lineage of tool calls across nested agent loops.
•The interface now supports real-time streaming of subagent logs, allowing developers to debug 'hallucination loops' or infinite recursion in agentic workflows as they happen.
•Vercel has introduced specific cost-attribution tagging for subagents, allowing teams to break down token consumption and latency by individual sub-task within a single parent project run.

📊 Competitor Analysis▸ Show

Feature	Vercel (eve)	LangSmith (LangChain)	AgentOps
Subagent Visualization	Native/Integrated	Graph-based tracing	Event-stream focused
Pricing	Usage-based (Vercel)	Tiered/Enterprise	Per-agent/Seat
Deployment	Vercel Infrastructure	Platform Agnostic	Platform Agnostic
Primary Strength	Frontend/Edge Integration	Ecosystem Compatibility	Observability/Security

🛠️ Technical Deep Dive

Implementation relies on the OpenTelemetry standard for tracing subagent spans within the Vercel AI SDK.
Subagent activity is captured via a middleware layer that intercepts tool-use events and injects parent-run IDs into the metadata headers.
The shared timeline view is rendered using a high-frequency WebSocket connection to ensure sub-millisecond latency for real-time log updates.
Token usage is calculated by aggregating the input/output tokens from the underlying LLM provider API responses mapped to each specific subagent turn.

🔮 Future ImplicationsAI analysis grounded in cited sources

Vercel will introduce automated subagent optimization suggestions.

By collecting granular performance data on subagent failures and token usage, Vercel is positioned to offer AI-driven recommendations for prompt refinement.

The eve framework will become the standard for Vercel-hosted agentic applications.

Deep integration between the observability interface and the framework creates a high switching cost for developers using Vercel's AI infrastructure.