How Agentforce Routes Conversations Between Agents and Humans
Agentforce

How Agentforce Routes Conversations Between Agents and Humans

By Troy Amyett•November 13, 2026•6 min read
Speed
AI-narrated

Agentforce manages the handoff through explicit routing patterns that delegate tasks or full control while carrying context forward. This approach turns transfers into planned steps rather than last-minute rescues.

How does Multi-Agent Orchestration handle routing?

Multi-Agent Orchestration gives you two main patterns for moving work between agents. In Handoff Mode the orchestrator passes full control to a subagent so the subagent can finish the conversation directly with the user. The orchestrator stays in the background and watches what happens without rebuilding the final reply. This pattern works well when you have specialized subagents that need to own an entire topic such as billing questions or HR policy checks.

Supervisor Mode keeps the orchestrator in charge of every reply to the user. It sends only the current request to a subagent for classification or light handling and then pulls the result back for synthesis. The choice between these modes depends on whether speed or oversight matters more for that workflow. Handoff Mode removes one synthesis step and therefore reduces latency on routine specialist queries.

What context actually travels during a handoff?

Context moves through conversation history, intent summaries, record IDs, and structured variables. You can also use Agent Script to define custom variables that persist across steps. Flows can update objects such as the Messaging Session record so the next participant sees the same state. Without these mechanisms the receiving agent or human starts the interaction cold and forces the user to repeat details.

Salesforce documentation shows that context can include both raw history and summarized intent so the handoff stays natural. Structured packages let you pass verified data such as account status or open case numbers rather than dumping an entire transcript. This matters because poor context transfer is the main reason customers feel they have started over.

When should you route to a human instead of another agent?

You route to a human when the query requires judgment that exceeds the agent’s defined scope or when policy demands a person. Agentforce supports this through an Escalation topic or the @utils.escalate action. Before the transfer the agent can complete pre-work such as verifying the account or creating a draft case so the human receives a prepared record.

Omni-Channel flows then place the interaction into the right queue with the context already attached. Voice channels add extra options such as Smart Routing that funnels every escalation through a single topic for intent capture before dispatch. The result is that humans receive work only after routine steps are complete.

Where Agentforce handoff falls short

Multi-agent orchestration works only inside a single org. It does not yet support handoffs to agents in another Salesforce org or to non-Agentforce systems outside the listed integrations. If your architecture spans multiple orgs or requires immediate bidirectional audio with external platforms such as Amazon Connect, you will need additional middleware until the planned fall 2026 connector arrives.

Timeouts also apply: the orchestrator has a 120-second limit and subagents have a 30-second limit. These constraints force you to keep individual actions small and to design fallback paths when a subagent does not respond in time.

How do you set up your first orchestrated handoff?

Start by reviewing the official help pages on Multi-Agent Orchestration and Agent Script. In Agentforce Builder create an orchestrator agent and connect the subagents you need. Map variables in the canvas so context passes cleanly and configure the After Response section to handle post-processing.

For human escalations add an Escalation topic and set the canEscalate flag where required. Write instructions that include the keyword “Escalation” so the system recognizes the intent. Connect an Omni-Channel flow that routes based on the captured intent summary. Test the entire path in the Agentforce Testing Center using handoff assertions before you activate the versions.

Begin with high-frequency, low-risk tasks so you can measure results quickly. Once those flows stabilize you can expand to more complex multi-intent scenarios.

What changes when you add voice or external channels?

Voice conversations use the same core patterns but add topic-level triggers and conditional escalation rules. You can require the agent to complete specific actions before it offers an escalation option. Smart Routing collapses every escalation into one topic whose only job is to capture intent and hand off cleanly.

External platforms such as Amazon Connect are on the roadmap for real-time bidirectional audio over WebSockets with context passing targeted for fall 2026. Until then you can still route messaging or chat sessions through Einstein Enhanced Bots into Agentforce agents using intent-based rules and preserved variables.

Frequently asked questions

How do I keep context from disappearing during an agent handoff?

You keep context by mapping variables in the orchestrator canvas, updating Messaging Session records through Flows, and using intent summaries that travel with the request. Agent Script also lets you define custom state that persists across transitions.

What is the difference between Handoff Mode and Supervisor Mode?

Handoff Mode lets the subagent respond directly to the user after the orchestrator delegates control, which removes an extra synthesis step. Supervisor Mode keeps the orchestrator in the reply loop and only sends individual requests to subagents for processing.

Can I escalate from any point in a multi-agent conversation?

Yes, escalation is supported from anywhere in the conversation once you configure the Escalation topic or action. The agent can finish verification or record creation before the transfer occurs.

Does Agentforce support handoffs outside a single Salesforce org?

Current multi-agent orchestration stays inside one org. Cross-org or external non-Agentforce agent handoffs require additional integration work until connectors such as the planned Amazon Connect link become available.

How do I test whether my handoff logic actually works?

Use the Agentforce Testing Center to run scenarios that include handoff assertions. Activate only versions that pass these tests so routing and context transfer behave as expected in production.

Found this useful? Pass it on.

Share it with someone on your team who's working through the same problem.

Funnelists Services

Agentforce & AI Agents

Autonomous agents that predict, automate, and execute revenue-driving actions across sales, service, and marketing.

Troy Amyett

Troy Amyett

Founder & Chief Solutions Architect

9x Salesforce certified. Agentforce Specialist and Agentblazer Legend, 2025–2026. Anthropic-certified in Claude Code and MCP.

Get Insights in your inbox

AI-powered perspectives on Salesforce and Agentforce, delivered weekly.

No spam. Unsubscribe anytime.

Your competitors aren't waiting. Why are you?

Let's talk about what AI agents could do for your business. 30 minutes. No pitch deck. Just answers.

Launch before they do