Design and Product in the Age of AI

Authors: Rob Gifford, Principal Designer; Dani Nordin, Design Architect; Victor Thuo, Principal Design, Jana Epstein, Product Manager @athenahealth

In 2026, the impact of generative AI on product development has moved from speculation to daily practice. Many people in the industry have now had time to experiment with AI tools, but the path forward can be mystifying; AI can be astonishingly helpful with some tasks while still struggling with others — making it easy to both overestimate and underestimate what these tools can actually do.

For UX Designers and Product Managers, that path is especially hard to chart — the patterns and best practices simply haven’t settled yet. Our process is often less prescribed in general, and our deliverables are typically more interpretive than our Engineering partners’, which makes the “right” way to use AI harder to pin down. That lack of a single path can be overwhelming, but it is also exciting.

At athenahealth, Product & UX teams have been testing AI across research, design, and strategy work for over a year. Through a series of conversations and a panel discussion at Boston’s User Experience Professionals Conference (UXPA) reflecting on this experience, clear themes emerged around how to leverage AI and the skills we think will be most valuable in the future.

Jana Epstein, Victor Thuo, Dani Nordin and Rob Gifford discussing AI at Boston’s UXPA Conference

The Best Uses of AI for Design and Product Work (Right Now)

Often AI utilization is discussed synonymously with “acceleration” and “efficiency”. While UX and Product are seeing significant efficiency gains, they’re not regularly achieving the 5–10x gains Engineers are. The most dramatic impact we’re seeing is AI’s ability to improve the quality of our thinking and unlock new methods to increase our understanding of a problem space.

Use 1: Using AI as a Thought Partner

One of the clearest ways AI sharpens our thinking is by serving as an always-ready thought partner. Victor Thuo, a Principal Designer in athenahealth’s Data Ecosystem & Platform division, uses AI as a digital team of experts to help him take on different perspectives before sharing work with stakeholders. He created a panel of role-based agents, such as Product Management and Engineering agents, that can critique ideas, identify blind spots, suggest next steps, or surface relevant project background.

The intent is not that AI replaces these stakeholders’ input. It is to provide a first reaction to strengthen ideas before they enter a broader conversation. By asking AI to react from the perspectives of different disciplines, Victor can bring more developed thinking to the team and make sure a wider set of considerations is on the table.

Jana Epstein, a Product Manager in Clinicals, described a more reflective version of the same pattern. She uses AI to organize her thoughts around difficult product decisions, management questions, or strategic tradeoffs. When wrestling with a hard problem, she’ll go for a “walk and talk” with AI, where she uses the tool’s conversational and reasoning abilities to help organize her thoughts and gain clarity.

That may sound simple, but it points to one of the more durable uses of these tools: AI is useful not because it is an oracle that can be trusted to produce the right answer every time, but because it can catalyze divergent thinking, expose assumptions, and help practitioners clarify their own point of view.

Use 2: Using AI to Understand More, Earlier

Another high-value use case for UX professionals is the ability to do data synthesis at a scale that would have been impractical before. In research and discovery, AI can help teams examine larger, messier bodies of information, which can lead to more focused primary research and ultimately improve hypotheses and product decisions. While Dani Nordin, a Product Design Architect in athenahealth’s Clinicals division, still uses tools like HeyMarvin and Airtable to accelerate analysis for standalone interview studies or surveys, she feels the bigger opportunity is synthesizing qualitative data sets that are too large, too complex, or too messy for traditional manual synthesis.

One example came from work exploring voice and conversational experiences in the electronic health record. Rather than starting with a speculative list of themes, Dani used Claude to analyze thousands of questions clinicians had posed to athenahealth’s chart-based chatbot. The analysis helped surface patterns in what clinicians were trying to accomplish using this tool. In her words, “the biggest innovation wasn’t that it took less time or that I used AI; it was the fact that I could do the analysis at all. Trying to manually code 7,000+ user prompts would have almost certainly produced a level of human error that I didn’t see with the AI tools.”

AI can also help practitioners ramp up on unfamiliar domains in a way they couldn’t before. Rob Gifford, a Principal Designer in Clinicals, uses Codex and NotebookLM to scan clinical information and workflow research when beginning discovery in a new space. Instead of relying only on a few subject-matter expert conversations or slowly working through academic papers one at a time, he can quickly build an evidence-based foundation of knowledge. That early foundation changes the quality of downstream research. With it, primary user research can focus on more tailored questions and test sharper hypotheses — uncovering what is not already well established.

The point is not simply speed. The point is that AI makes a broader and more rigorous starting point feasible. In both examples, the quality of AI synthesis depended on the quality of the research plan and context the model was provided. AI needs clear research questions, relevant context, guardrails, and explicit expectations about confidence — a vague question will usually produce a vague answer from AI.

Use 3: Using AI To Prototype Faster

Prototyping is another area where UX teams are already seeing meaningful acceleration. AI development tools make it possible to build interactive prototypes that feel closer to real software. In a few hours or less, designers can produce prototypes that feel and behave more like a live feature without manually mocking up every complex state and interaction in Figma.

The purpose, though, is not to ship AI-generated prototype code (this can add unnecessary time, when the solution is still forming). Rather, it is to improve the quality of feedback we get from users by giving them something more realistic to react to. A good comparison from the past is leveraging paper prototypes in concept testing — they’re quick and cheap, but the fact that users can visualize something helps them understand the idea. AI-coded prototypes go a step further than this: by letting users actually interact with the solution, they can better evaluate how it would fit into their work (not just grasp the concept behind it).

What Human Skills will be Most Important Moving Forward?

A common anxiety the Product and UX community expresses around AI is that it will make product and design roles less necessary. In practice, we are seeing the opposite. As engineering teams use AI to increase development velocity, the bottleneck shifts from “Can we build this?” toward “Should we build this?” and “How should it work?” Those are the questions Product and UX functions exist to answer. They require user understanding, business judgment, design sensibility, technical awareness, and deep organizational context. In healthcare, the stakes are especially high because clinical workflows are complex and deeply ingrained in how our users work. An AI-generated interface or spec may look plausible while missing the reality of how the work it’s targeting actually gets done. Product and UX are needed to keep AI-enabled delivery grounded in real workflows and meaningful outcomes.

Skill 1: Judgment & Communicating Intent

AI can produce plausible artifacts, but it cannot — and should not — decide what a team is trying to accomplish, what constraints matter most, or what success should look like. That responsibility belongs to humans.

Using AI can turn Product and UX practitioners into direction-setters and expert reviewers, so the quality of AI output really rests on how well we articulate the design intent, constraints, and quality bar — and, after output is produced, how well we inspect whether AI has created something faithful to the problem we asked it to solve. End quality ultimately depends on our ability to respond to AI output as a “first draft” that can be refined and iterated on based on expertise. While many practitioners who are newer to using AI can fall into the trap of thinking AI will do the work for them, that human-in-the-loop is a vital part of using AI to augment one’s own skills.

That makes communication skills — verbal and especially written — incredibly valuable. AI still behaves, in many ways, like a junior teammate: it can do a lot with clear direction, but it can also wander when the prompt is vague or the context is incomplete.

If intent defines what good should look like, judgment is what helps teams choose well among the multitude of options AI so easily creates. As synthesis, prototyping, and documentation get faster, teams will have more data to interpret, more concepts to explore, and more functionality to consider. The critical skill becomes evaluating these options.

That evaluation depends on context no model fully has: every user interview, roadmap discussion, design critique, stakeholder concern, implementation constraint, and hard-won lesson a designer or product manager has accumulated from past work.

In contrast to current LLMs, humans have an “infinite context window,” which gives them a special role in carrying that context across projects. Further, humans understand people, workflows, and what a good experience feels like in ways that are difficult for even good AI models to envision. Prompting and tool fluency matter, but the durable skill is judgment and discernment: knowing when an AI-generated output is helpful or off-target.

Skill 2: Cross-functional Collaboration

It is tempting to read AI’s productivity gains as a reason for each discipline to move faster on its own. If a Product Manager can generate a specification, a designer can generate a prototype, and an engineer can generate code, why slow down to coordinate?

While AI lowers the cost of producing artifacts, it also increases the possibility of producing disconnected artifacts when other factors — like quality, understanding of the problem being solved, etc. — aren’t taken into account.

Teams that lean primarily on speed as a success metric may move quickly toward a less coherent product. AI can amplify a single point of view just as easily as it can broaden one. Without other humans in the loop, it becomes easier to fall into tunnel vision.

The better approach is to use AI to collaborate more deliberately. Product, UX, and Engineering need shared context, shared evaluation criteria, and a shared vision for what the experience should become.

That might mean Product, UX, and Engineering jointly shaping the specification they pass to an AI to build. It could also mean leveraging some of engineers’ freed capacity to observe user interviews or engage in technical research to contribute to early product discovery.

The good news is that AI tools can make cross-functional collaboration smoother because they make it easier than ever to gain foundational knowledge in new domains. For example, a designer who isn’t familiar with a technical framework can quickly develop a cursory understanding of it. LLMs now make it possible to create tailored learning guides or quickly answer targeted questions to fill in specific knowledge gaps. Having a basic level of shared knowledge can help team members collaborate more seamlessly because they can understand each other’s language and mental models.

Skill 3: Defining Vision and Value

One way to make sure we’re building the right thing is to establish a clear product and experience vision that defines what fully solving a specific user-centered problem looks like. Historically, these visions have often been constrained by delivery capacity. In healthcare especially, teams may understand the broader workflow a user needs but still be required to release a relatively thin MVP because the ecosystem is complex and engineering capacity is limited.

As engineering velocity increases, we’re able to deliver larger slices of that vision or set a more ambitious one — closing the gap between what we know users need and what we can realistically ship. That said, we don’t typically want to ship a multitude of disconnected enhancements. We want to build towards a cohesive, seamless vision.

There are a variety of skills needed to create a vision: imagination, strategic thinking, as well as user empathy, a broad research toolkit, stakeholder management, and authentic communication, to name a few. These are the areas human designers and product managers will need to lean into, even as they use AI to augment their work.

While it is tempting to believe that an LLM’s output can unlock 5–10x productivity for all teams, the most salient opportunity accelerated AI output provides is not just a higher volume of changes to our product. It’s the greater value we can bring to customers through richer experiences that solve a greater portion of their needs.

Conclusion

We’ve already witnessed how useful AI can be for Product and UX work, but it’s not always in the ways industry hype suggests. The productivity gains are real and significant, but AI’s greatest impact may actually be increasing the quality of human thinking and the depth of experiences we deliver to our users. Its value is not in replacing Designers and Product Managers, but in sharpening their instincts by expanding what we can explore, evaluate and imagine.

In other words, Product and UX are not disappearing now that AI is here. If anything, our value is becoming more concentrated around where we’ve always added the most value: framing the right problem, understanding people in context, defining intent, and shaping a coherent vision.

Connect with the Authors:

Rob Gifford, Principal Designer, athenahealth Clinicals

• *linkedin.com/in/robertgifford9

Dani Nordin, Design Architect, athenahealth Clinicals linkedin.com/in/daninordin

Victor Thuo, Principal Design, athenahealth Data Ecosystem & Platform linkedin.com/in/victorthuo

Jana Epstein, Product Manager, athenahealth Clinicals linkedin.com/in/parsonsjana