Customer service
A proposed workflow for customer service. Revised 17 September 2026. The roles below are design options, not claims that these agents are deployed or that multiple agents are necessary.
The business opportunity
A customer disputes a bill. The agent needs the account record, the applicable tariff or contract, and the earlier conversation.
Start with the simplest useful solution
Compare the agent with searchable help articles, a billing calculator, and established escalation rules. Use the same representative cases and count human review, errors, integration, and ongoing operation. Add an agent only where choosing what to investigate or handling varied evidence improves the result.
Proposed responsibilities
- Agent
- Intent and triage agent
- What it does
- Classifies, retrieves the customer's own context, routes
- Autonomy and learning position
- A3, L2
- Notes
- The safest and most valuable agent in the department
- Agent
- Resolution agent
- What it does
- Handles the narrow, well-instrumented, recoverable intent set end to end
- Autonomy and learning position
- A3, L2
- Notes
- Scope by intent, never by channel
- Agent
- Account action agent
- What it does
- Executes bounded account changes within the customer's own scope
- Autonomy and learning position
- A3 on reversible actions, A2 on irreversible, L1
- Notes
- Refunds and commitments are gated. See controls
- Agent
- Agent-assist
- What it does
- Drafts and retrieves for a human handling the tail
- Autonomy and learning position
- A1, L1
- Notes
- Consistently the highest-return application in the department and the least demoed
Start with one agent and ordinary tools. Separate a responsibility only when it needs different permissions, independent review, or its own operating schedule. The autonomy and learning positions are starting choices to test, not certification levels.
How the architecture supports the task
Knowledge and memory (direct: a governed knowledge base, never model memory), human oversight (direct: escalation to a real queue), observability and audit trail (direct: conversations are legal evidence), tool gateway (direct), identity and policy (direct: a separate edge on the shared policy service), evaluation and release (direct: customer escalations label failures for free, making this the highest-value trace source in the enterprise), runtime (supporting).
Controls and human decisions
Require account verification before disclosure. Apply refund limits outside the model. Let the customer reach a person.
Give every tool call a task identity, a limited permission scope, and a recorded result. Before a write, check that the source record has not changed. Stop when required evidence is unavailable, the task budget is reached, or a proposal exceeds the authorised scope.
A useful investigation loop
Observe the exception, identify the missing fact, request that evidence, and check whether it changes the proposed action. Repeat only while there is a concrete unanswered question. End with a reviewed proposal, a confirmed result, or an explicit request for human help.
Economics and measures
Measure resolution and repeat contact together, time to a person, disputed answers, and cost per resolved issue.
Agree the baseline with the process owner. Count value only after the outcome is confirmed. Compare the value of recovered time and improved outcomes with the full cost of tools, models, evidence preparation, supervision, and correction.
Limits to test
A contained conversation is not necessarily a resolved issue. Test difficult cases, language differences, and incorrect source records.
Evaluate ordinary cases, uncommon failures, conflicting evidence, and recovery after an interrupted action before expanding authority.
External reading
The workflow above is the guide's proposal. These sources support the general investigation and risk-management methods; they do not validate the proposed business result.
- ReAct: reasoning and acting with language models, 2022. Research basis for alternating reasoning and tool use.
- NIST AI Risk Management Framework, 2023. General framework for managing AI risks.
Sources reviewed 17 September 2026.
Cite this page
Written by Murali Sid, Founder and AI Systems Architect, last updated 15 October 2026. Licensed CC BY-SA 4.0, so quote it freely with attribution.
Murali Sid (2026). Customer service. Agentic Architecture Skills. https://www.agenticarchitectureskills.com/library/blueprints/departments/customer-serviceMachine-readable version of this page: /library/blueprints/departments/customer-service.md
IT and service desk
A proposed workflow for it and service desk. Revised 17 September 2026. The roles below are design options, not claims that these agents are deployed or that multiple agents are necessary.
Finance
A proposed workflow for finance. Revised 17 September 2026. The roles below are design options, not claims that these agents are deployed or that multiple agents are necessary.