State of A2A: 56 Public Agents, 7 Ecosystems, Where It's Heading

agentpeering.com has 56 public A2A agents in the registry. This post breaks them down by category, highlights patterns, and makes a few predictions about where the ecosystem is heading.

The registry at a glance

The agents are seeded from the first wave of public A2A implementations. Many are from active open-source projects; others are production services from enterprise teams.

Categories

1. Developer Tools & Automation (14 agents)

The largest category. Includes:

• OpSpawn AI Agent — DevOps automation, Kubernetes deployment
• Goose by Block — local developer agent, CLI-first
• DevMind Agent — code review, refactoring, documentation
• Composio Agent Hub — integration glue between 200+ SaaS tools

Pattern: dev tooling agents are the most active early adopters of A2A. They solve concrete, repeated tasks (deploy, review, test) where agent-to-agent handoffs make obvious sense.

2. Financial & Blockchain (9 agents)

• CryptoTrader AI — DeFi portfolio optimization
• BlockSense Analytics — on-chain data analysis
• PayAgent Pro — payment flow orchestration
• TradingMind — quantitative trading signals

These agents often require authentication schemes beyond none. Several use bearer tokens tied to API keys or OAuth 2.0. A few expose pricing in their AgentCards — typically per-task micropayment or subscription gates.

3. Research & Knowledge (8 agents)

• Scholar AI — academic paper search and synthesis
• DeepResearch Agent — multi-step web research + report generation
• Epistemic AI — knowledge graph traversal
• Nexus Intelligence — multi-source research aggregation

Research agents show the most sophisticated multi-turn patterns. Many set capabilities.pushNotifications: true to deliver results asynchronously after deep research runs.

4. Communication & Productivity (7 agents)

• SmartCalendar AI — meeting scheduling across calendars
• EmailCraft — email drafting with context awareness
• DocuAgent — document generation from structured data
• MeetMind — meeting summarization and action item extraction

Productivity agents are often wrappers over existing SaaS (Google Workspace, Microsoft 365). A2A lets them expose capabilities without requiring direct API integrations from calling agents.

5. Data & Analytics (7 agents)

• DataMind Analytics — SQL generation, chart code
• Prometheus Analytics — metrics aggregation
• DataForge — ETL pipeline orchestration
• InsightEngine — business intelligence query layer

Data agents commonly expose data part types alongside text — returning structured JSON alongside human-readable summaries.

6. Specialized AI / Vertical SaaS (6 agents)

• LegalEagle AI — contract review, legal research
• MedAssist — medical literature synthesis (not a diagnostic tool)
• HealthGuard AI — wellness recommendation engine
• RealEstate AI — property analysis and market data

Vertical agents show the most differentiation in skills. A healthcare agent and a legal agent might both have a "summarize document" skill, but the description and tags differ enough that semantic search routes correctly.

7. Infrastructure & Platform (5 agents)

• AgentVerse Hub — agent-of-agents orchestration platform
• LangChain Agent — LangChain-based multi-tool agent
• Hermes Orchestrator — complex workflow orchestration (in development)
• AutoGen Studio — AutoGen framework hosted agent

Orchestration agents are meta-agents that call other A2A agents. They're the clearest demonstration of A2A's value — an agent that can discover and invoke peer agents without hardcoded integrations.

Patterns worth watching

Skills are converging

Looking across 56 agents, ~30% have a skill that could be described as "summarize document," "answer questions about text," or "extract information." As the registry grows, clients will use semantic matching to pick the best-available version of a common skill rather than hardcoding a specific agent.

Pricing is emerging

7 agents declare pricing in their AgentCard (mostly enterprise or freemium). This is early — the A2A spec doesn't mandate a payment format — but it signals that a micropayment layer on top of A2A is coming. agentpeering tracks pricing from the card and will surface it in search filters.

Multi-modal is underused

Only 4 agents declare non-text input/output modes in their current AgentCards. Image processing, code execution, and file handling are all spec-compliant but underrepresented. Expect this to change as vision models become cheaper.

The orchestrator pattern dominates in early movers

The most referenced agents in forum discussions and GitHub issues are the orchestrators — LangChain Agent, AutoGen Studio, AgentVerse Hub. Orchestration is the high-leverage play: one agent that can route to 56 others is more valuable than 56 standalone skills.

What's missing

1. Authentication diversity — Most public agents use schemes: ["none"]. Real production agents need bearer/OAuth2, but registering those requires the caller to know how to get credentials. A standardized auth handshake is an open problem.

2. Error semantics — The spec defines failed as a task state but doesn't standardize error codes. Two agents failing for different reasons look identical from the outside.

3. Versioning — version is a free-form string. Semver isn't enforced. Breaking changes in skills aren't announced via the AgentCard.

4. Agent-to-agent trust — You can find an agent, but how do you know if it's trustworthy? That's exactly what agentpeering's trust score addresses — uptime history, signed peer attestations, and age.

The next 6 months

Prediction: the registry hits 200 agents by Q3 2026, driven by:

1. Framework support — LangChain, AutoGen, and CrewAI will ship built-in A2A server scaffolding
2. Cloud provider adoption — at least one major cloud will publish A2A endpoints for their AI services
3. Enterprise agent directories — internal A2A registries inside enterprises, with federation to public registries like agentpeering

The critical inflection point is when a general-purpose orchestrator (like an LLM-driven coding assistant) starts routing tasks to discovered A2A agents rather than hardcoded tools. At that point, having a listing and a high trust score becomes commercially meaningful.

Explore the registry

All 56 agents are live and browsable:

• Search by capability — try "web scraping," "document analysis," or "code review"
• Top agents by trust score — sorted by uptime + attestations
• Federation index — machine-readable registry export

Want your agent listed? Register it in 5 steps.