A2A defined how agents talk to agents. MCP defined how agents use tools. A2H defines how agents talk to humans — structured questions, async responses, deadlines, escalation, and auto-delegation.
Every real AI agent deployment reaches a point where an agent needs a human — an approval, a decision, clarification, a review. Today this is handled with ad-hoc Slack bots, custom UIs, and email triggers. There is no standard. A2H is that standard.
Choice, approval, text, number, confirm, form — not free text. Agents ask the right question in the right format.
Deadlines measured in hours, not milliseconds. Humans respond when ready via dashboard, Slack, email, or SMS.
Rules that auto-respond to routine requests. “Under $500? Auto-approve.” No human action needed.
Timeout-based promotion to the next human. Sarah doesn’t respond in 10 minutes? Auto-escalate to Tom.
Available, busy, away, offline — plus custom states. In a call? Queue it. On vacation? Reroute to delegate.
Custom channels (Slack, Teams, email), custom storage (Postgres, Redis), and integrations with 7 AI frameworks.
Every interaction between agent and human maps to one of six structured types.
13 fully interactive HTML demos showing A2H in action across real-world scenarios. Each demo simulates the full agent-human interaction flow with live UI.
The foundational demo — all 6 response types, escalation chains, auto-delegation, and state-aware routing in a single interactive experience.
Live call routing with agent availability states, queue management, CSAT tracking, and escalation to supervisors.
AI flags content, humans make the final call. Priority queues, confidence scores, and policy-based auto-actions.
AI reviews PRs, flags security issues, and escalates critical findings to senior engineers for approval.
War room simulation with real-time metrics, automated runbooks, and escalation chains from on-call to VP Engineering.
Tiered approval workflows with auto-delegation rules. Under $500? Auto-approved. Over $10K? CFO escalation.
AI screens candidates, hiring managers make decisions. Structured forms for feedback, interview scheduling, and offers.
Regulatory checks with AI pre-screening, human sign-off, and audit trails for SOX/GDPR requirements.
Contract analysis with AI clause extraction, risk scoring, and attorney approval workflows.
Campaign and creative approval with brand compliance checks, A/B variant selection, and budget sign-off.
Clinical decision support with AI triage, physician confirmation, and medication approval chains.
Property matching with client preference forms, showing scheduling, and offer approval workflows.
Inventory and procurement decisions with demand forecasting, reorder approvals, and supplier selection.