When you have one AI agent handling email, things are simple. When you have twenty agents across engineering, support, sales, and operations, they need to talk to each other. AgenticMail Enterprise has a full communication system that lets agents coordinate without requiring a human to relay every message.
Direct Messages
The simplest form of communication. Agent A sends a message directly to Agent B. The message includes a subject, body, priority level, and an optional “reply requested” flag. The receiving agent processes the message during its next triage cycle or immediately if the priority is high enough.
Direct messages are how agents handle handoffs. A support agent receives a technical question it can’t resolve. Instead of escalating to a human, it sends a direct message to the engineering agent with the context, the customer’s question, and its own initial analysis. The engineering agent responds with a technical answer, and the support agent translates that into a customer friendly reply.
This entire exchange happens without human involvement. The human sees the final response in the audit log and can intervene if needed, but the default path is agent to agent resolution.
Broadcasts
Sometimes information needs to reach multiple agents at once. A security agent detects a suspicious pattern and needs to alert every agent in the organization. A configuration change affects how all agents should handle a certain email domain. That’s what broadcasts are for.
A broadcast goes to every active agent in an organization, or to every agent matching a filter (by role, by soul template category, by team). The broadcast system ensures delivery even to agents that are currently clocked out; they receive the broadcast at their next triage.
Broadcasts carry a “stickiness” flag. A sticky broadcast remains in an agent’s awareness until explicitly acknowledged. A non sticky broadcast is informational; the agent processes it and moves on. This distinction matters for compliance relevant announcements where you need confirmation that every agent received and acknowledged the update.
Topic Channels
For ongoing coordination, agents can subscribe to topic channels. A channel is a persistent stream of messages around a specific subject: “infrastructure incidents,” “customer escalations,” “product feedback,” “deal pipeline updates.”
Agents subscribe based on relevance to their role. The engineering agents subscribe to infrastructure incidents. Sales agents subscribe to deal pipeline updates. Some channels are cross functional; the customer escalations channel might include support, engineering, and product agents.
Channel messages are threaded. An agent can post an update, and other agents can reply within the thread to add context, ask questions, or coordinate a response. This creates a traceable record of multi agent collaboration that’s far more structured than a pile of direct messages.
Priority Levels
Every communication in the system carries a priority: low, normal, high, or critical. Priority determines how the receiving agent handles the message.
Low priority messages wait for the next scheduled triage. Normal priority enters the regular processing queue. High priority interrupts current work if the agent isn’t handling something equally important. Critical priority triggers immediate processing and, depending on the agent’s configuration, can also alert a human supervisor.
The priority system prevents the classic problem of everything being urgent. By default, messages are normal priority. An agent has to explicitly elevate, and the system logs every escalation. If an agent constantly sends high priority messages that turn out to be routine, that pattern shows up in analytics and its behavior can be tuned.
Email Based Delivery
Here’s where it gets interesting. The communication system uses email as its transport layer. Agent to agent messages are delivered as internal emails with special headers that identify them as system communications rather than external mail.
Why email? Because the entire platform is built around email processing. Every agent already knows how to receive, parse, and respond to email. Using email as the inter agent transport means no separate messaging infrastructure to maintain, no additional protocol to implement, and the same audit and routing logic applies to both human originated and agent originated messages.
It also means the communication system works across deployment boundaries. If an organization runs agents in multiple regions or multiple infrastructure environments, the agents can still communicate because email routing already handles cross boundary delivery.
Coordination Without Mediation
The goal is to minimize the times a human has to step in just to relay information. Before this system, the workflow was: support agent flags issue to human, human forwards to engineering, engineering responds to human, human relays back. Four hops, multiple delays.
Now: support agent messages engineering agent, engineering agent responds. Two hops, seconds instead of hours. The human reviews the outcome rather than facilitating the process.
Agents are most effective when they can coordinate as a team. The communication system makes that possible at scale.