Single Agent Deployment
When one bounded responsibility can own the outcome, we don't build more system than the work requires.
Trigger
Request or Event
A ticket, message, or scheduled trigger starts the work.
One Accountable Responsibility
Single Agent
One bounded operational responsibility.
Human Control
Approval / Escalation
Defined review points before or after action.
Outcome
Outcome
A completed, evaluated piece of work.
What It Is
One bounded operational AI responsibility, connected to approved knowledge and tools, with human controls and evaluation around it. It has a single owner and a single defined job.
When It Fits
This fits when
- One primary workflow with a clear owner
- Limited durable state between steps
- A defined handoff into or out of the workflow
- No material agent-to-agent orchestration required
Look elsewhere when
- The work spans multiple coordinated specialists
- Several workflows need to share the same knowledge and controls
- State must persist across many independent processes
What Can Live Inside a Single Agent
Retrieval, classifiers, model routing, evaluators, and deterministic tool calls can all support one commercial responsibility without turning it into a Multi-Agent Workflow. What stays singular is the accountable operational role — not the number of underlying components.
Examples
Illustrative patterns, not a list of completed customer deployments.
Knowledge Lookup with Ticket Creation
Answers a defined question set from approved knowledge and opens a ticket when it can't fully resolve the request.
Bounded Document or Operations Assistant
Works from an approved document or operations context to draft, summarize, or check a single class of output.
Intake with One Approved System Action
Takes a structured intake, validates it against rules, and performs a single approved action in a connected system.
How We Validate It
- An acceptance suite built from real, representative cases
- Tool and integration checks against approved systems
- Defined failure cases and refusal behavior
- Human control points reviewed and confirmed
- A documented deployment baseline before go-live
Evaluation Loop
- Task completion against the acceptance suite
- Regression checks against prior behavior
- Policy and tool-use adherence
- Escalation and refusal accuracy
What Happens After Launch
After launch, a Single Agent Deployment moves into Managed Intelligence, or into a documented customer-operated ownership handoff — the boundary is defined before go-live, not after.