Three patterns cover the shape of nearly every operational workflow. All three share the same
governance posture. The difference is how the agent reaches the work, not whether the work is
controlled.
Pattern 01
Chat Agent
Users ask in natural language. The agent switches to structured input when precision matters.
People describe what they need in the chat tools they already use. The agent reads your context,
drafts the work, waits for approval, and then executes against your systems of record. For critical
workflows, the agent renders structured input cards inside the chat window:
explicit fields, validations, and choices that keep free-text ambiguity out of high-stakes
requests.
Replaces: helpdesk overflow, repetitive request handling, slow first response, and misread intent on high-stakes requests.
Pattern 02
Desktop Agent (Computer Use)
The agent takes action on the user's system, on their behalf.
The Desktop Agent operates the applications your team already uses, including legacy applications
that never got an API. It clicks, types, fills forms, and moves data between systems on the
user's behalf. It runs on the user's own desktop or on a managed virtual desktop pool, and
it carries the same approval and audit posture as the Chat Agent. The human stays in control; the
agent handles the swivel-chair work.
Replaces: legacy app fragility, manual data entry, and hand-offs between apps.
Pattern 03
Ambient Agent
The agent orchestrates AI, people, and systems across end-to-end business processes.
Always on. The Ambient Agent watches events, schedules, and signals from your systems, then runs a
governed end-to-end flow that coordinates AI, people, and systems of record across
the full business process. Humans stay on the approval gate for anything that matters, with the
trust, governance, and control that mission-critical operations demand.
Replaces: missed SLAs, late detection, after-hours escalations, and manual hand-offs across teams and tools.
Knowledge of your business is a capability inside all three patterns. It is not a
fourth product to buy. The agent reads your context (documents, tickets, runbooks, history) wherever
the work needs it.
YAA\IF