An agent that only responds when poked isn’t really an agent. It’s a chatbot with extra steps. The autonomy system in AgenticMail Enterprise gives agents the ability to manage their own time, set their own priorities, and work toward goals without constant human supervision.
Clock In, Clock Out
Every agent has a work schedule. They clock in and clock out. This isn’t metaphorical. The system tracks when each agent is active and when it’s off duty. During off hours, incoming emails queue up rather than triggering immediate processing.
Why bother? Enterprise environments have expectations about when work happens. A support agent blasting automated responses at 3am looks unprofessional and can violate SLAs. Clock schedules also let you stagger agent activity to spread API usage and avoid rate limits during peak windows.
The schedule is configurable per agent. Some agents run 24/7 for critical monitoring. Others follow business hours in a specific timezone. The clock system handles timezone math, holiday calendars, and override windows for urgent situations.
Morning Triage
When an agent clocks in, the first thing it does is triage. The morning triage routine surveys everything that accumulated overnight: new emails, pending tasks from yesterday, updates from other agents, and any system alerts.
The triage process categorizes each item by urgency and effort. Quick wins (items that take under two minutes to resolve) get handled immediately. High urgency items get prioritized for the first hour. Everything else enters the regular processing queue.
This mirrors what effective humans do. You don’t just start working on whatever’s at the top of your inbox. You scan, prioritize, and batch. Agents do the same thing, except they’re consistent about it every single morning.
Daily and Weekly Catchup
Beyond the morning triage, agents have scheduled catchup routines. The daily catchup runs in the afternoon and reviews the day’s progress. What got done? What’s still pending? Are there any items that have been sitting too long?
The weekly catchup is more reflective. It looks at patterns: recurring request types, response time trends, escalation frequency, and goal progress. The agent generates a summary and can send it to its human supervisor or post it to a team channel.
These catchup routines feed into the goal tracking system, creating a feedback loop where the agent continuously adjusts its behavior based on observed outcomes.
Goal Tracking
Each agent can have active goals. These aren’t vague aspirations; they’re structured objectives with measurable criteria. A support agent might have a goal to reduce average first response time to under 15 minutes. A sales agent might track pipeline value generated from email interactions.
Goals have timeframes, milestones, and progress metrics. The agent checks its goals during morning triage and daily catchup, adjusting its prioritization based on which goals need attention. If response time is creeping up, the agent shifts toward faster initial replies even if that means less detailed first responses.
The goal system integrates with the soul template. An agent’s personality influences how aggressively it pursues goals versus how cautiously it operates. A risk averse security agent won’t cut corners on review thoroughness just to hit a speed target.
Knowledge Update Schedules
Agents have scheduled windows for updating their knowledge base. During these windows, they review recent interactions, extract patterns, update their internal documentation, and refresh any cached data from external systems.
This is important because it prevents knowledge updates from interrupting active work. The agent isn’t constantly reindexing while trying to answer emails. It batches knowledge maintenance into dedicated time slots, just like a human might block off Friday afternoons for documentation and cleanup.
Heartbeat Monitoring
Every active agent emits a heartbeat at regular intervals. The monitoring system tracks these heartbeats and flags agents that miss their check in. If an agent goes silent for longer than its configured threshold, the system can attempt an automatic restart, notify an administrator, or redistribute the agent’s pending work to a backup agent.
The heartbeat carries more than just “I’m alive.” It includes current queue depth, processing rate, error count, and memory usage. This gives the monitoring dashboard a real time view of agent health without explicit status queries.
Making It Self Directed
The autonomy system ties all of these pieces together. An agent wakes up, triages its work, processes items according to its priorities and goals, catches up on progress, updates its knowledge, and reports its health. All of this happens without a human telling it what to do next.
The human role shifts from directing to supervising. You set the goals, define the schedule, review the catchup summaries, and intervene when the agent flags something beyond its authority. The agent handles the routine. You handle the exceptions.
That’s what autonomy means in practice. Not uncontrolled AI doing whatever it wants, but structured self direction within well defined boundaries.
Source Code
The default autonomy settings control the full suite of self directed behaviors, from clock schedules and catchup routines to goal checking and guardrail enforcement:
export const DEFAULT_AUTONOMY_SETTINGS: AutonomySettings = {
enabled: true,
clockEnabled: true,
dailyCatchupEnabled: true,
weeklyCatchupEnabled: true,
weeklyCatchupDay: 1, // Monday
goalCheckEnabled: true,
goalCheckHours: [14, 17],
knowledgeContribEnabled: true,
knowledgeContribDay: 5, // Friday
escalationEnabled: true,
guardrailEnforcementEnabled: true,
};