Runtime plane / Tier 2 / Phase 1

Experience & Interaction

How does a human participate?

The surfaces where people meet the system — and where they steer it: channels, streaming, sessions, clarification, provenance display and the approval moments that make autonomy safe.

01 / WHAT THIS BOUNDARY OWNS
  • Channels: web, mobile, chat platforms, voice and public API
  • Streaming and partial-result rendering so long tasks stay legible
  • Session and conversation state at the edge
  • Clarification prompts when the task is under-specified
  • Approval and review surfaces for human-in-the-loop checkpoints
  • Provenance display: what the answer was based on and how confident it is
  • Multimodal input and output, and accessibility
02 / INTEGRATION POINTS
03 / ENTERPRISE ACCOUNTABILITY

Product Engineering / Digital Channels, with UX

  • Product lead — accountable owner
  • Frontend engineer — streaming, sessions, channel adapters
  • UX designer — approval, clarification and provenance patterns
  • Content designer — the words in the interface
  • Accessibility specialist

Architectural recommendation, not a research finding.

04 / WHAT DONE LOOKS LIKE
Pilot can ship one channel with streaming. Approval and clarification surfaces become mandatory the moment agents can write to production systems.

Signals to watch

  • Task completion and abandonment rate
  • Clarification rate and its effect on success
  • Approval latency and reviewer override rate
  • Accessibility conformance
  • User-reported trust or satisfaction
05 / FAILURE MODES
  1. No clarification path, so the agent invents assumptions instead of asking
  2. Approval fatigue: too many low-stakes confirmations and reviewers stop reading
  3. Provenance hidden, so users cannot calibrate trust
  4. Disclosure obligations missed for chatbots regardless of risk tier
06 / REFERENCE LANDSCAPE

Tool and standards snapshot · last reviewed approximately mid-2026

Open source

  • AG-UI protocol
  • CopilotKit
  • Vercel AI SDK
  • assistant-ui
  • Chainlit
  • LiveKit or Pipecat for voice

Managed

  • Contact-centre and CPaaS platforms
  • Enterprise chat platforms (Teams, Slack) as channels

Standards

  • AG-UI for streaming agent interfaces
  • Server-Sent Events and WebSocket
  • WCAG 2.2 accessibility
  • Art. 50 EU AI Act disclosure for machine interaction
ARCHITECTURE NOTE

This is the harness most often skipped in an architecture diagram and most often blamed in a post-mortem. Autonomy policy is only real if there is a surface where a human can actually intervene in time.