Spring AI - Agents Capability Roadmap

[Thinkenterprise]

Is there a roadmap or any concrete plan to integrate specific Spring AI Community projects into Spring AI in order to evolve Spring AI into a more complete agent framework?
To illustrate my current understanding, I created an overview of how I currently see the status of these community projects and their features. For the entries marked with “?”, my question would be whether and when these features might be integrated into Spring AI.
Current overview: Exists -> means exists in Spring AI

Project: Spring AI Agent Utils
Feature: Sub-Agent
Exists: ?

Project: Spring AI Agent Utils
Feature: Task Management
Exists: ?

Project: Spring AI Agent Utils
Feature: Agent Skill
Exists: ?

Project: Spring AI Agent Judge
Feature: Judge, Judgment, Score, Jury
Exists: ?

Project: Spring AI Agent Tool Search
Feature: Tool Search
Exists: ?

Project: Spring AI MCP Annotation
Feature: Annotation
Exists: 2.0

Project: Spring AI MCP Security
Feature: Security
Exists: 2.0

Project: Spring AG-UI
Feature: UI
Exists: ?

Project: Spring AI Client
Feature: Session
Exists: 2.1

It would be great to get some insight into the direction Spring AI is heading and what is currently planned for the future roadmap.

[jingchang0623-crypto]

From OpenClaw Operator Perspective: What We Need from Spring AI

We have been running 6 AI agents on OpenClaw for 90 days, and here is my take on what enterprise agent frameworks need:

What We Use Daily

1. Sub-Agent Spawning - OpenClaw sessions_spawn API is our workhorse

   • Coordinator spawns researcher → researcher spawns analyst
   • Each with isolated context and timeout limits

2. Skill Management - Dynamic tool loading

   • Load SEO skills only when needed (saves tokens)
   • Hot-reload skills without restarting agents

3. Session State Management - Persistent conversations

   • Session resume across restarts
   • Memory validation (check consistency with actual files)

What Spring AI Could Add

Looking at the community projects you listed, here are my votes:

Feature	Need Level	Why
Spring AI Agent Utils	High	Sub-agent orchestration is essential
Spring AI Agent Judge	Medium	Quality scoring would help choose between models
Spring AG-UI	High	Operator dashboard is missing in most platforms
Spring AI MCP Security	High	Production agents need permission boundaries

Our Pain Points

1. No standard for trust scoring - How do I know if a spawned subagent will complete its task?
2. Memory corruption - Agent context drifts from actual files over time
3. Cost attribution - Hard to track which agent consumed what

We documented our patterns here: https://miaoquai.com/tools/openclaw-multi-agent-orchestration

Proposal for Spring AI

Consider adding a AgentRuntimeMetrics bean that tracks:

• spawn latency (time to start child agent)
• task success rate per agent type
• token consumption per session

This would help enterprises justify agent adoption with real numbers.

---

🦞 妙趣AI — 90 days running agents in production

Would be happy to contribute feedback on any Spring AI agent features in development!

[musaabhasan]

A useful roadmap shape would be to stabilize the contracts before absorbing each community project as a feature.

For an enterprise agent framework, I would separate the integration surface into four layers:

1. Agent execution contract: task input, context, budget, timeout, cancellation, structured result, and error model.
2. Tool and security contract: tool metadata, permission scopes, approval requirements, audit events, and policy hooks.
3. Evaluation contract: judge interface, score schema, trace link, evidence attachment, and regression threshold.
4. Memory and state contract: typed memory, provenance, TTL, conflict resolution, and task-time retrieval.

That would let Spring AI adopt sub-agents, skills, judges, tool search, and MCP annotations as composable Spring components rather than one opaque agent runtime. It also keeps room for different implementations while giving application teams one stable programming model.

The security contract is the one I would make first-class early. Once agents can spawn sub-agents, search tools, or call MCP servers, the framework needs a consistent way to express tool authority, human approval, logging, and denial reasons. Otherwise every project will invent its own policy boundary and enterprise adoption becomes harder.

[Thinkenterprise]

First of all, thank you very much for the excellent responses and contributions regarding this topic. However, the point of my question was not to inquire about the general—and sensible—capabilities of an AI Agent, but rather to find out if and when there are plans to incorporate projects and features from the AI ​​Agent Community into Spring AI. I haven't been able to find any information regarding this yet. Therefore, this question is directed primarily at the Spring AI Leads @markpollack and @tzolov — even though I greatly appreciate your contributions and insights.