Topics
June 16, 2026
June 16, 2026
AI isn’t just changing the tools and structures of teams—it’s creating an identity crisis among workers at large. And leaders are trying to find their way to navigate the new challenges ahead.
In the recent webinar “Building the Research Team of Tomorrow," leaders from Pinterest, TD Bank, TaxWell, Cisco Networking, and more gathered to discuss how AI is not just changing their tools, but fundamentally reshaping the profiles and fundamental structures of their teams.
Below, they share some tips for how to navigate the continually changing path ahead.
Building the research team of tomorrow requires a shift from viewing research as a support function to viewing it as a core engineering and product partner. That team is already here, and its members are acting as architects of the systems they once only studied.
Dana Cho (Pinterest) emphasizes that the most impactful research output is no longer a slide deck. Instead, researchers should distill years of accumulated user insights into "system-ready principles" that can be coded directly into the product.
In this new paradigm, UXR and machine learning engineers work side-by-side in iterative cycles, rather than through traditional handoffs. Focus on how your systems thinking lands in the development process, and ensure your insights live on as institutional knowledge within the system itself.
Andy Vitale (TaxWell) argues that in a fast-moving AI environment, what someone knows today is less important than how fast they can learn. He advises hiring for "slope"—the rate at which an individual improves.
The future researcher needs the agency to not only find a problem, but to fix it. This means screening for practitioners who are comfortable pushing a pull request to fix a mislabeled field or a broken workflow discovered during testing, rather than just filing a ticket.
Mary Piontkowski (Cisco Networking) sees researchers as builders who evangelize for user-centered outcomes. They do this by embedding persona and research context into the system. This involves building synthetic agents that pull in fresh insights from multiple sources on a regular basis. Many researchers’ roles are shifting from exercising authority over decisions to influencing outcomes across a distributed network of AI-powered agents.
Daniel Avrahami (Cisco Networking) points out that internal artifacts—like research plans and transcripts—are no longer just for the researcher’s consumption. They are now valuable data that feed into a project's long-term memory. Build and maintain these artifacts so that they can be used by the system or other researchers to improve subsequent projects.
The integration of AI into workflows is about more than just speed, it’s about reclaiming time for high-level strategy and higher-stakes judgment. As the floor for execution rises, it becomes more important than ever to point the team toward the right questions.
Andy Vitale (TaxWell) notes that AI has made "no s***" research—findings that merely confirm what everyone already suspected—incredibly cheap and easy for anyone with an LLM account to produce. To be a high performer, you must focus on "oh s***", research: surfacing insights that the business genuinely did not know, and that change a fundamental decision. Performance is now measured by your judgment in choosing which questions are worth asking in the first place.
Daniel Avrahami (Cisco Networking) warns of a potential structural skill gap between those who have the luxury to play with AI tools and those who do not, such as high performers overloaded with work or parents with young children. He advises leaders to explicitly carve out time for teams to experiment and learn new tools, like Cursor, even if those tools might eventually become obsolete. The goal is to flex the skill of learning rather than mastering a specific piece of software.
Riley Heintz (TaxWell) shares an individual contributor's perspective: AI should be used to offload busy work to free up capacity for research strategy. By streamlining workflows with AI, you can facilitate research more quickly without skipping critical steps, ensuring user insights are incorporated into the process—even at high speeds.
Christian Rohrer (TD Bank) suggests using AI to fortify your ability to conduct heuristic analyses. For example, AI can act as an assistant to produce scorecards (like the PURE method) that evaluate the ease of an experience. This allows you to catch issues and maintain quality standards at the same cadence as the rapid shipping schedules typical of the AI era.
As AI democratizes the ability to conduct research, the danger of low-quality, surface-level insights increases. Keeping the focus on users requires new layers of governance and a commitment to human judgment.
Daniel Avrahami (Cisco Networking) highlights a growing concern: non-researchers, like PMs, taking raw transcripts, running them through AI, and immediately acting on the output without proper synthesis.
Make an effort to build governance into your research artifacts and repositories. This ensures that when someone not formally trained as a researcher queries a repository, the insights generated remain genuine, trustworthy, and reflective of data integrity.
Mary Piontkowski (Cisco Networking) emphasizes that while engineering teams are moving faster, researchers need to remain the guardians of "bold and brave human judgment". It’s critical to protect discovery and foundational work to listen for the things users don't even know they need yet. Work to influence from both the top-down and bottom-up to ensure a user-centered approach isn't sacrificed for speed at all costs.
Dana Cho (Pinterest) acknowledges that AI can trigger an existential crisis for researchers who feel their core value—such as moderating or synthesizing—is being automated. She advises reframing this by asking: "What could you do today that you literally could not do before?" By viewing AI as an expansion of scope rather than a limitation of identity, researchers can engage with the technology as makers who build agents and prototype synthetic workflows.
Daniel Avrahami (Cisco Networking) argues that while organizations are great at shipping, they often fail at learning and refining. Researchers should invest heavily in post-launch insights to understand how generative AI experiences are actually landing with customers. Because AI outputs are often non-deterministic, it’s important to move beyond testing fixed prototypes to researching how real workflows evolve and interact with users over time.
Scaling research in the AI era means moving from manual, one-off studies to recurring evaluation programs that measure trust, accuracy, and performance over time.
Dana Cho (Pinterest) points out that a new downstream role for researchers is building eval frameworks that determine if an AI feature is actually good enough to ship. This requires high levels of judgment to decide what quality looks like for an AI system.
Additionally, researchers can contribute to collective knowledge by sharing best practices on how to create new "judges" (AI evaluators) and tooling with their engineering partners.
Mary Piontkowski (Cisco Networking) suggests that researchers are in a unique position to own the metrics that matter for AI, such as accuracy, latency, and model drift. As agents pile on top of one another, the experience becomes something that no one explicitly designed. Researchers can use these metrics to evangelize for user-centered outcomes in complex environments.
Andy Vitale (TaxWell) uses AI to solve the problem of seasonal user access. By taking deep institutional knowledge of user behaviors, his team builds synthetic users that can run through experiences during the off-season. This allows for continuous testing and refinement, even when real participants are difficult to reach.
Daniel Avrahami (Cisco Networking) notes that his team has shifted toward recurring programs that routinely evaluate the quality of AI responses and track shifts in user trust. These programs show progress and identify changes in customer attitudes toward features that allow AI to perform actions on their behalf.
The overall sentiment is clear: embrace the tools, but lead with judgment. By shifting from report-generators to system-builders, you ensure that as research scales at unprecedented speeds, it never loses sight of the human at the center of the technology.
As TaxWell’s Riley Heintz put it, the best approach right now is to be a sponge—soak in the new information, stay adaptable, and never stop experimenting.
.png)
