Great experiences, now that AI agents are in the conversation
How AI agents shape the way people are served, and how to design, govern and measure those experiences well.
Customers are starting to meet organisations through AI agents: assistants that answer questions, compare options, book, buy and sort out problems on someone's behalf. Some of those agents belong to you. Some belong to your customers. Increasingly, they talk to each other. Agentic experience is the discipline of making every one of those moments clear, trustworthy and genuinely useful, for the people involved and for the agents acting for them.
What is agentic experience?
Customer experience used to assume a person on one side and a brand on the other. Agents change that. An AI agent can now sit on either side, or both, deciding what to show, what to ask and what to do next.
Agentic Experience Management (AXM) treats those agent-shaped journeys as something to be designed and looked after deliberately, rather than left to whatever a model happens to do on the day.
- Agent-to-human: the tone, honesty and helpfulness of agents that serve your customers and staff
- Agents as users: making your services easy for someone else's agent to understand, navigate and act on correctly
- Agent-to-agent: the hand-offs, negotiations and confirmations that happen between agents, often out of sight
- Human in the loop: knowing when an agent should pause, check or pass to a person, and making that hand-over smooth
- Trust and measurement: disclosure, consent, accountability, and signals that tell you whether the experience is actually working
Four interactions to design for
Most organisations will end up running all four at once. Each needs its own design thinking, guardrails and ways of telling whether it is going well.
Agents that serve people
Your own assistants on the website, in the app, on the phone or inside a messaging channel.
- Say plainly that it is an AI agent, and what it can and can't do
- Keep the voice consistent with your brand and your values
- Show sources and reasoning where it matters
- Make it easy to correct, undo or start again
Designing for agents as users
A customer's assistant may visit your service before the customer ever does. It needs clear, structured and accurate information to work with.
- Publish services, prices, policies and hours in machine-readable form
- Offer predictable, well-documented actions instead of fragile screens
- State limits and conditions so agents don't over-promise
- Return clear errors an agent can explain back to its person
Agents talking to agents
Bookings, quotes, claims and support increasingly pass between one organisation's agent and another's.
- Agree what each side is allowed to decide on its own
- Confirm identity and authority before acting
- Keep a readable record of what was agreed and why
- Bring the humans back in when stakes or uncertainty rise
Hand-offs to people
The moment an agent passes a customer to a person is where trust is most easily won or lost.
- Pass on full context so nobody has to repeat themselves
- Escalate on frustration, vulnerability or high-impact decisions
- Give staff the agent's reasoning, not just its transcript
- Feed what staff learn back into the agent's design
Example: describing a service so an agent can use it well
Designing for agents as users starts with being explicit. This sketch shows the kind of information a visiting agent needs: what the service does, what it can act on, what the rules are, and when a person should take over. Protocols such as MCP and A2A define how agents connect; the experience question is what you say and promise once they do.
// Illustrative only — not a formal standard { "service": "Appointment booking", "provider": "Example Clinic, Ōtautahi Christchurch", "audience": ["people", "ai_agents"], "actions": { "check_availability": { "requires_consent": false }, "book": { "requires_consent": true, "confirm_with_person": true }, "cancel": { "requires_consent": true, "notice_hours": 24 } }, "policies": { "cancellation": "https://example.co.nz/cancellations", "privacy": "https://example.co.nz/privacy" }, "disclosure": "Agents must identify themselves and who they act for", "handoff": { "when": ["urgent symptoms", "complaint", "person asks for a human"], "to": "reception team, weekdays 8am–5pm" } }
A fictional example. The field names are for illustration, not a published specification.
A practical approach to AXM
Agentic experience is not a one-off build. It is an ongoing loop of mapping, designing, governing and measuring, much like any well-run customer experience programme.
Find where agents already touch your customers
List the assistants you run, the third-party agents that reach your services, and the journeys where an agent could reasonably act for someone. Note where decisions are made and by whom.
Design the conversation and the contract
Shape how your agents speak, what they may do on their own, and what information and actions you expose to outside agents. Write down the promises, limits and hand-off points.
Build in trust, consent and accountability
Be clear about who is responsible when an agent acts, how personal information is handled under the Privacy Act 2020, and how people can question or reverse an agent's decision.
Watch outcomes, not just activity
Track whether people and agents actually got what they needed, review samples of real interactions, and keep improving. Agents change as models and prompts change, so the loop never really ends.
Measuring agentic experience
Traditional CX scores still matter, but they miss a lot of what agents do. These are useful questions to build measures around.
Task completion
Did the person, or their agent, get the outcome they came for without a workaround?
Accuracy and grounding
Were answers and actions correct, consistent with policy, and traceable to a source?
Hand-off quality
When an agent passed to a person, was it at the right moment, with the right context?
Corrections and reversals
How often do people undo, override or complain about something an agent did?
Agent-readability
Can outside agents find, understand and correctly use your services and policies?
Trust signals
Do people know they're dealing with an agent, feel in control, and choose to come back?
Who it's for
Anyone in Aotearoa New Zealand responsible for how customers, citizens or members are served, now that some of that service runs through AI agents.
Customer experience teams
Leaders extending CX practice to journeys that agents now shape
Product and service designers
Designers working out how to design for people and their agents
Contact centres
Teams blending AI agents with people on the phone and in chat
Public and community services
Agencies, councils and not-for-profits where trust and fairness come first
Retail and service businesses
Brands whose customers are starting to shop and book through assistants
Digital, data and risk leaders
People accountable for privacy, safety and governance of AI in practice
Part of the AXM network
Agentic Experience sits alongside the other sites in the AXM network, each looking at a different side of how AI agents are changing the way New Zealand organisations work and serve people. Start at the hub to see the whole picture.
Visit axm.co.nz