Scenario mapping becomes an increasingly valuable way to understand the experiences they create.
Experience mapping has always helped professionals understand how people interact with products and services across different situations. It brings together intent, expectations, decisions, friction and outcomes, helping us see the experience beyond individual screens or isolated interactions.
As technology evolves, so does the nature of that experience.
Traditional applications have largely been shaped around predefined screens, navigation paths and structured user inputs. Headless architectures, including the evolution of headless CRM, are already changing this relationship now by separating backend capabilities from the interfaces through which people access them.
AI agents extend this shift further..
They can interpret requests, retrieve information, call APIs, remember previous interactions and coordinate multiple steps before presenting an outcome. What emerges is an experience that is no longer defined solely by the interface. Much of it is shaped by what happens behind it: what the system understands, what it remembers, what it can access and how it responds when circumstances change.
This introduces another dimension to experience design. Understanding what a person sees and does is still important, but understanding how the system behaves across different situations is becoming just as relevant.
And this is where scenario mapping takes on a greater role.
From mapping journeys to anticipating behaviour
Traditional journey mapping gives us a view of how people move through an experience. Scenario mapping brings another perspective by exploring how that experience unfolds under different conditions.
Imagine a customer might begin with one intention and change direction halfway through. They may provide incomplete information, contradict something shared earlier, encounter an unexpected error or ask for an action that requires additional permissions.
In a conventional application, such situations might translate into different screens, validation messages or alternative paths or even escalation call. With agentic systems, they can influence how system interprets the request, what information it retrieves, which actions it takes and how it responds next.
For example, take a customer contacting their bank about an unfamiliar transaction. Initially, they may simply want an explanation. Once they understand the transaction, they might decide to freeze their card. That's where interaction has moved from information seeking to a consequential action.
The experience now depends on whether the agent recognises that change in intent, retains relevant context and responds appropriately yet ensuring it doesn't talk too much. A technically successful interaction can still feel disconnected if the system continues following original direction. As a result, frustration could creep in at that moment!
Scenario mapping becomes valuable because it brings these less predictable situations into conversation. It moves the focus beyond expected path and towards the range of behaviours that shape actual experience.
Human intent remains central
As agents become more capable, much of the attention naturally shifts towards their technical possibilities. Yet experience continues to revolve around a person trying to achieve something. Intent, context, expectations and control remain central, even when the interaction is conversational and the underlying execution is increasingly autonomous.
Now, another example : consider someone checking personal loan eligibility. The request might sound transactional, but the underlying intention could simply be to explore available options. They may not be ready to apply, share additional information or make a financial commitment.
An agent that immediately initiates an application flow has misunderstood the experience, even if it has technically responded to topic.
This distinction becomes more significant as conversational interactions replace some of predictability associated with predefined navigation. People can move between questions, explore alternatives, change their minds and introduce new concerns as they learn more.
Intent is not necessarily established at beginning and carried unchanged throughout an interaction. It evolves with the conversation. The experience depends on whether system can recognise that evolution without losing context that came before it.
KARC: The capabilities shaping agent behaviour
An agent's behaviour is shaped by more than its ability to interpret a prompt. It depends on information it can access, actions it can perform, context it retains and way it coordinates different parts of a workflow.
These four connected capability areas can be understood through KARC: Know, Act, Remember and Coordinate.
Know — Information and retrieval
Agents increasingly draw on knowledge bases, documents, APIs and connected enterprise systems to interpret requests and formulate responses. The quality of an interaction is closely tied to relevance, reliability and currency of the information being retrieved.
An agent working with an outdated policy or conflicting information may still produce a confident-sounding response. From customer's perspective, the interaction may appear seamless while the underlying information is incorrect.
This creates an interesting tension between access to information and confidence in that information. Having more sources available does not automatically mean the system understands which source is appropriate for a particular situation, or when the available evidence is insufficient.
Act — Capabilities and permissions
The transition from providing information to executing actions changes the nature of the experience. Through tools, APIs and connected services, an agent may update records, submit requests, initiate transactions or coordinate activities across multiple systems.
Yet these actions carry different levels of consequence. Looking up an account balance is not equivalent to transferring money or changing account details.
The experience is influenced by authority granted to the agent, the safeguards surrounding consequential actions and what happens when an operation fails midway.
As agents gain access to more capabilities, the relationship between autonomy, permissions and user control becomes increasingly significant. The question is no longer simply whether an agent can perform an action, but whether it should perform that action in the circumstances presented.
Remember — Context and continuity
Memory influences whether an interaction feels continuous or disconnected.
An agent may retain working context during a conversation, general knowledge about a subject, relevant details from previous interactions or established procedures for completing a task. These are different forms of memory, often described as working, semantic, episodic and procedural memory.
Their value lies in helping the agent understand more than latest message. However, continuity also introduces its own complexities. Outdated information, missing context or an incorrectly remembered preference can influence subsequent decisions.
An experience that remembers everything indiscriminately is not necessarily a better experience.
The distinction lies in whether the right context is available when needed, and whether it can be corrected, updated or appropriately disregarded.
Memory therefore becomes more than a technical capability. It has a direct influence on how people experience continuity, relevance and trust.
Coordinate — Orchestration and handoffs
A seemingly simple request can involve multiple systems, actions and decisions. Agents may plan a sequence of steps, call different tools, validate intermediate results and coordinate handoffs to other agents, services or people. Each transition introduces dependencies. One action may rely on information retrieved earlier, while a failure in one part of the workflow can affect everything that follows.
Partial completion, conflicting instructions and unclear ownership become important parts of the experience. A workflow may appear to be progressing successfully while an unresolved dependency sits somewhere behind the scenes.
Coordination is therefore more than executing a sequence of tasks. It influences whether the overall outcome remains coherent when multiple systems and decisions are involved.
Beyond expected path: When scenarios change
The implications of these capabilities become clearer when we look at how agent behaviour varies across different situations. Some interactions follow the expected path. The agent understands the intent, retrieves relevant information and completes the appropriate action.
Others expose the limits of the system. Information may be missing, an API may fail, an assumption may be incorrect or a connected service may become unavailable. These situations reveal whether the agent can recognise what has gone wrong, explain the limitation and help the person recover.
Then there are edge cases: conflicting information, unusual permissions, unexpected dependencies or conditions that fall outside the standard workflow. These often expose assumptions that remain invisible during ordinary interactions.
Perhaps most interesting are changes in intent.
A person may begin by exploring an option, move towards a decision and then reconsider. They may introduce a related concern or change direction entirely. The agent needs to recognise that shift rather than continue pursuing the original objective.
Conversational interactions rarely follow perfectly linear paths. People discover what they need as they interact with the system, and sometimes the original request turns out not to be the real problem. An agent that continues pursuing an earlier instruction after the person's intention has changed can create a frustrating or even risky experience.
These scenarios reveal why agent behaviour cannot be understood through the happy path alone. What happens when information is incomplete, expectations change or the system encounters something it cannot resolve is equally important to the experience.
Experience is bigger than task completion
An agent completing a task does not automatically mean the experience was successful.
A request might be submitted correctly, a transaction might be processed or information might be retrieved. Yet the person may still be uncertain about what happened, why a particular decision was made or what comes next. Task success is only one part of the outcome.
Conversational freedom matters too. Can people ask questions, explore alternatives, clarify information and change direction naturally, without feeling constrained by the system's interpretation of their original request?
Trust matters because the experience needs to feel reliable, transparent and most important - appropriately controlled.
Recovery matters because errors and unexpected situations are unavoidable. The ability to explain limitations, acknowledge failures and help someone continue can be just as important as completing the original task. These dimensions expose the limitations of measuring agent performance purely through task completion. The quality of the interaction, the confidence it creates and way it handles unexpected situations all contribute to the outcome.
Critical thinking and trust become part of experience
As agents take on more execution, another dimension comes into focus: how much should people rely on an agent's output, and where does human judgement remain important?
Delegating execution can be useful when people retain enough context to understand and evaluate the outcome. A different pattern emerges when outputs gradually become accepted without meaningful examination.
There is a spectrum between cognitive offloading, where execution is delegated while ownership is retained; deferral, where agent recommendations increasingly shape decisions; and cognitive surrender, where outputs become accepted without a felt need to question them.
These patterns introduce a less visible dimension of agentic experiences: whether people retain enough understanding to challenge, correct or take ownership of the result.
An agent may complete a tool call successfully or return a green checkmark, but that does not necessarily establish that the original intent was understood or the intended experience was achieved.
The evidence matters, as does understanding what it actually verifies.
This is where critical thinking and verification intersect with experience design. They are not simply checks performed after an agent has completed its work. They influence how much autonomy is appropriate, how transparent the interaction needs to be and where people should remain involved.
I'll explore this further in the next article, Critical Thinking When Agents Do the Work, looking at what happens to judgement and ownership as more of the workflow moves into the hands of agents.
Bringing the pieces together
Scenario mapping connects human intent with capabilities and behaviour of increasingly autonomous systems. It brings together what people want to achieve, what agents know, what they can do, what they remember and how they coordinate actions. It also makes visible - expected flows, failures, edge cases and changes in intent that influence the quality of an experience.
This does not replace journey mapping or other established experience practices. It extends their relevance into environments where system behaviour is less predictable and where decisions can happen across multiple steps, tools and interactions.
As agentic experiences become more capable, understanding the experience requires looking beyond the interface and into the behaviour of the system itself.
The value of mapping isn't just about documenting what should happen. It’s about making visible what could happen—like when a simple checkout flow unexpectedly turns into frustrating loop, or when user effortlessly finds exactly what they need on the first try, or when agent should stop going much further comprising integrity and privacy and knows where to have human take front seat. Understanding these happy and negative paths is where scenario mapping truly finds its purpose with AI agents.
Sometimes, confidence doesn't mean having an answer for everything. It means recognising when the experience needs to take a different path.