Three critical UX challenges facing AI agent adoption.
As the tech industry rushes to build AI agents for every industry vertical, there is often an underlying assumption that because agents can handle messy, unstructured tasks, the user experience can be equally open-ended and unstructured. In our experience this is far from the truth.
On the contrary, AI agents pose unique design challenges to both the teams building them and users adopting them. These challenges manifest as conceptual challenges causing onboarding abandonment, integration challenges limiting effectiveness, and confidence challenges restricting retention and growth.
Conceptual challenges: Agents require users to build new mental models.
Agents fundamentally change the product paradigm people expect. Rather than users performing tasks with a product, they now manage entities within the product to perform tasks. This change adds new complexity that, if not addressed properly, creates barriers to user onboarding and leads to high abandonment rates during initial adoption.
The design challenge lies in making this new paradigm understandable. This requires simplifying technical language, providing clear educational content, and explicitly introducing new conceptual frameworks, to make sure users create accurate mental models of how an agent works with the product. You can achieve this through thoughtful naming, information architecture, and intentionally reinforcing concepts throughout onboarding and regular use.
Thoughtful education as the path to user confidence.
Recently, we worked with a startup creating agents for banking and financial applications. They hired us to redesign the onboarding and information architecture to address issues with sales and adoption as they expanded into regional banking markets. Prospective customers were having difficulty understanding what the agents were capable of, and new customers were struggling to train analysts to use them.
Our primary focus for the onboarding experience was to introduce the agent in deliberately simple terms: how it worked, what it could do, and how it was trained. This provided a conceptual foundation users could build upon.
We audited the app's terminology, replacing technical terms with industry-familiar language. We introduced as few new concepts as possible, and when introducing something new, we used clear, plain language that built on existing mental models and known product patterns. To clearly distinguish between where users manage agents vs where agents work independently, we created an icon representing agents to show where and when they’re working throughout the experience.
Workflow challenges: Successful integration requires deep knowledge of user processes.
Once users have a clear understanding of an agent and its capabilities, the next challenge is to make sure it can integrate into their daily workflows. It’s easy for product teams to focus on the underlying technology, while neglecting how it's applied to people’s uses in the messy, real world. If not done well, integration challenges can limit effectiveness and result in low efficiency and time savings.
If you're instructing an agent to perform a task, you must thoroughly understand that workflow to direct it effectively. On every project, we invest heavily in collaborating with product owners, subject matter experts, and users to map out the underlying processes and their permutations. This often takes the form of demos, interviews, and FigJam workflow diagrams, but the goal is establishing a shared foundation everyone can build from.
Understanding users is key for simplifying complex experiences.
When you spend time to understand the underlying workflows, solutions often present themselves.
When redesigning the loan application agent for a client, one of our key design challenges was simplifying the agent setup process. After studying how loan applications were actually reviewed, the solution became clear: Rather than training the agent in the abstract, the user would review loans, training the agent as they went.
We wouldn’t have been able to simplify the configuration process without understanding the workflows that it was used for, and that the user was already familiar with.
Confidence challenges: Visibility into agent decisions builds essential user trust.
The new dynamic of delegating work to independent agents requires an enormous amount of trust from users who remain responsible for outcomes. While agents can technically go off and work in the background with no human visibility into their operations, this approach ignores the reality of what people need.
In highly regulated industries like financial services and critical infrastructure this black box dynamic is especially unacceptable. Teams must design visibility into agents' decision chains for traceability and auditing, while building human oversight and quality checkpoints into workflows—especially for high-risk decisions.
When users don’t have confidence and trust in these systems, it restricts long-term adoption and causes users to revert to manual processes, limiting expansion across teams and use cases.
Building confidence as a stepping stone to adoption.
We've watched this dynamic play out across our clients. End users who might be wary of costly mistakes or job displacement, hesitate to trust agents with their most complex workflows, choosing instead to test them on lower-risk tasks like data entry or basic document processing.
This often causes leaders to question agent effectiveness when they don't see immediate ROI. But we've learned this cautious approach is actually a critical step—it gives teams the opportunity to build confidence with agents before applying them to more substantial workflows that deliver bottom-line savings. Building trust is the essential first step toward meaningful adoption.
New technologies don’t change design fundamentals.
The good news is that while the technology is new, the principles remain the same. Designing products that are easy to understand, seamlessly integrate into workflows, and instill confidence has always been the goal. When building agents, these fundamentals simply become more critical than ever.
However, applying these principles to agent design requires both design expertise and real-world experience with the specific challenges agents present. The difference between successful and failed agent implementations often comes down to how well teams navigate the conceptual, integration, and transparency challenges we've outlined.
Teams that invest in understanding these challenges early—and design solutions for them—create agents that users actually adopt and trust.
–––––––
Ready to tackle these challenges? Our team has first-hand experience helping teams transform their agent concepts into products users actually trust and adopt. Let's talk about your project.