Agents in Tracks

​

Agents inside dynamic operating procedures

Tracks guide people and AI through real work — non-linear, handoff-friendly, and built for how jobs actually run. Agents participate as first-class teammates: watching progress, fulfilling work, and taking tasks the same way humans do. This is orchestration without building a complex multi-agent system. One or two well-configured agents, assigned to the right Track roles, often cover the need.

​

Agents as overseers

An overseer is anyone responsible for moving a track forward — team members, contacts, or agents. Overseers visit checkpoints, mark milestones, create tasks, run Actions, and add notes. Assign an agent as overseer when you want AI to monitor, nudge, triage, or fulfill routine checkpoint work — onboarding reminders, document checks, status summaries, and similar. See Overseers and tasks for roles, permissions, and track-level vs. instance-level detail.

​

Tasks: humans and agents in the same loop

Overseers collaborate through tasks (todo requests):

1. Create — write what needs doing
2. Assign — send to another overseer (human or agent)
3. Fulfill — complete the work and mark done

A human can assign to an agent; an agent can assign back to a human. That mixed loop mirrors real teams — and depends on the agent having the right Focus and context to fulfill what it accepts.

​

Agents at milestones and automation

Tracks connect milestones to Actions for automated steps. Agents intersect that model in two ways: Use rails (Actions, approval milestones) for steps that must be guaranteed. Use agent judgment for steps that need reading, summarization, or case-by-case decisions. See Milestones and automation for wiring Actions to milestones.

​

Continuity and handoff

Tracks preserve continuity — history, progress, pending tasks, and notes stay intact when work passes between people or agents. An agent picking up an instance should see the same context a human would: which checkpoints were visited, which milestones remain, and open tasks. Configure agents with linked context (policies, SOPs) and appropriate Focus (Contact Book, registry data) so handoffs are informed, not cold starts. See Continuity and handoff.

​

Variables on track-related agent work

When an agent runs inside a Track-related Action or agent node, variables in the system prompt may be filled from track or instance data — names, dates, checkpoint labels, and similar. Define variables in the Builder; supply values where the track or Action connects to the agent.

​

Design checklist for track agents

Before assigning an agent to a track, confirm:

1. Goal — one sentence for what this agent does on this track
2. Never — behaviors that need human approval or deterministic Actions instead
3. Context — linked files and Focus for the entity type the track follows
4. Measure — how you will spot failures (Chat Review, instance notes, Track Data)
5. Blast radius — start on low-risk checkpoints before high-stakes ones

​

Where to go next

If you would like something unique to your business or have a unique issue, use XSpada Evolve or ask a XSpada rep.