AI, Margins, and the New Division of Labour: Three Reversals Reshaping How Research Operates
by Kate Towsey
Subscribe to get sharp thinking all about ResearchOps delivered straight to your email inbox. Thanks to our sponsors, it’s free.
The ResearchOps Review is supported by User Interviews, now part of UserTesting. User Interviews makes it fast, easy, and affordable to recruit participants so you can scale research without sacrificing quality.
In times of challenge and change, my father typically reminds me of one word: judo. A Japanese martial art founded in 1882 by Kanō Jigorō, “judo” translates to English as “the gentle way.” What distinguishes judo from other martial arts is that, rather than relying solely on brute strength, its masters leverage an opponent’s momentum and balance (or lack thereof) to win a fight. When an opponent’s fist or leg comes hurtling towards them, rather than resist, they move with the energy, using it to topple them. “Maximum efficiency with minimum effort,” or Seiryoku-Zenyo, is the goal—a quote that’s highly applicable to the world of research operations.
Change can also be seen as an opponent to resist; you’ve likely experienced plenty of change and challenge in recent years. Particularly when it’s unplanned, change can be daunting, but with an understanding of the forces at play and enough adaptability and wit, it’s often possible to leverage the energy of change to your personal and professional advantage. That’s the ethos of judo.
Over the past year, as a storyteller and the editor of this publication, I’ve been studying the changes happening in the field of UX research and, by proxy, research operations, and I’ve noticed three major flips or reversals. Understanding these reversals—things that have turned on their heads—has helped me navigate the fast-changing landscape with greater clarity. For instance, in how I curate and edit this publication. In this article, I’ll share the three reversals, their impact on the field, and what I think you should consider as a result. I hope they’re immediately relatable (let me know if they’re not), and that understanding them gives you greater strategic clarity, too.
1. Research Was Reactive; Now It Should Be Proactive
If you’ve worked in research for any length of time, you’ll know that most research teams have been operating along the lines of a reactive service model (whether you had used that term or not). In a hotel setting, a guest calls the front desk for room service, and a string of support, kitchen, and service staff jump into action to meet the need. Similarly, in an in-house research team setting, a product or design manager reaches out to their version of a front desk, whether a familiar researcher, research manager, ticketing platform, or some other avenue, to submit their research request. If their request is accepted, at least one researcher, and perhaps a research operations team, hops into action to deliver the insights.
Until now, it’s been necessary for research teams to operate along these lines; guessing what dozens of stakeholders across silos might need to know, when, was impossible. Even so, it rarely worked well. For service requestors, their insights often “arrived cold:” too late to answer their original question. For service providers, being treated as if they were filling takeaway orders, and dealing with research requestors’ impatience for quick answers, led to a lot of grumbling and groaning (on both sides), and the need for research leaders to want to fight for a better, more strategic seat at the table. Fair enough.
With the advent of AI, however, the most mature research teams are making the reactive research model, or at least reactivity as the sole operating model, a thing of the past. AI has ushered in an era of proactivity that the research profession could only have dreamed about just a year ago. Now, rather than waiting for someone to request research, teams are using AI to build proactive research systems. These systems can gather, curate, analyse, and ship data at volumes that were previously unimaginable. But the most interesting part is that savvy research innovators are building systems that can predict what insights stakeholders might need, when they might need them (often before they know themselves), and they’re using AI to spot gaps in their existing knowledge bank that they must fill to meet those predictions.
Proactive Pipelines and Predictive Engines
In May 2026, I published “How to AI UXR: A Map for Building AI-Augmented Research Operations,”1 which you can download to study in detail—and it is detailed. The map is based on 562 data points gathered via desk research, one-on-one interactions, and four working sessions with over fifty research professionals. The following are examples of the types of proactive research systems that innovative research teams are building:
Instant weekly PESTLE briefings. One team built a weekly trigger using the desktop AI agent Claude Cowork that runs concurrent PESTLE (political, economic, social, technological, legal, and environmental) queries, logs the results in a database, and delivers weekly briefings to partners, providing a regular update on external market forces.
Continuous Insight Streams. The notion of insight streams was common in the How to AI UXR research. Several teams have built custom agents that regularly scrape and synthesise internal data such as Net Promoter Score (NPS) metrics, support tickets, sales notes, customer call transcripts, and in-app queries, as well as external data such as Reddit threads and App Store reviews. They then publish weekly digests or real-time summaries to stakeholders, detailing broader customer sentiment beyond research studies.
Social listening. It regularly strikes me that the field of research has a lot to learn from marketing. Marketing professionals habitually monitor digital conversations and mentions across social media and the web to better understand consumer sentiment, industry trends, and competitor activity, then use that data to predict and plan content. The official term is social listening, and dedicated tools exist to help them listen. With their permission, research teams are starting to digest corporate strategies and PESTLE data, along with stakeholders’ meeting transcripts, blog posts, planning tickets, and group messages, to “listen” to their stakeholders’ interests and use this data to inform their proactive research roadmap and a research content calendar.
Knowledge gap analysis. AI is transforming research repositories, not just in how existing knowledge is retrieved, but also in how that knowledge is used. Forward-thinking research teams are starting to use AI to compare the insights they predict stakeholders will need against the knowledge that’s already housed in their repository, then planning proactive research studies to fill the knowledge gap.
In a proactive model, when the data or insights that predictions indicate stakeholders will need don’t yet exist, or what does exist must be updated or fleshed out to deliver more depth, then a highly capable team of strategic researchers can swing into action to fill the gap. It’s an entirely different model for operating a research team.
Proactivity Depends on Excellent AI Skills, Not Promptcraft
Over the past year, the art of prompting has been the predominant focus of AI courses and influencer content, followed by AI accuracy evaluations, or evals. The more tacit, and, I believe, more important, proficiency to acquire is the art of building AI skills: specialised AI capabilities that are designed to do one thing well. If you build a skill well, the AI should guide the user to prompt well. In other words, the future of proactivity hinges on skills more than promptcraft.
In episode three2 of the How to AI UXR podcast series, DoorDash researcher Nam Pham shares that “Early on when we designed the skill, the concept of chaining the skills was very important for us because we wanted it to be a guided experience. Especially for people who didn’t know what to do when it comes to research. So we trained the skill to recommend the next step, and the next step, until they have the results. So you know, a bit of service design knowledge actually helps when designing these kinds of skills.”
Key to delivering proactive research is designing systems that deliver content via LLM chat interfaces, in ways that guide the user to pause when necessary, ask the right questions, and reach out to a qualified human where necessary. But also to design systems, per the earlier examples, that deliver seemingly serendipitous insights into your stakeholders’ daily workspaces.
2. In a Profit-Oriented World, Operational Efficiency Is Now the Priority
The timing couldn’t have been more synchronous. As I was pondering how to explain this reversal, a friend who works as a ResearchOps professional emailed to share that their entire UX research team had been let go, leaving them to deliver research operations without the guidance of a craft leader. It’s an unfortunately common message to receive.
Post-pandemic, organisations across the world have shifted from a growth model, which prioritises growing market share and revenue over profits, to a profit model, which centres on optimising margins, managing operating expenses tightly, and generating a positive cash flow. (In early 2026, I wrote about this shift in an article titled “Why the Distributed Growth Model Is Failing Research Teams—and What to Build Instead.”)3 This shift toward operational efficiency means that, in many places, operations professionals, or those willing to evolve into such a role, are being retained even amid layoffs.4
So where does research and research operations sit in this mix? Until a year or two ago, UX research teams were relative powerhouses: large enough teams hired a significantly smaller number of operations staff to help them deliver more impact than they could on their own. For instance, in an operations-forward environment, a 120-person research team might include a ResearchOps team of ten. Quite often, however, these research leaders hired administrators as “operations” staff, misunderstanding the primary purpose of operations professionals as designers of highly efficient systems rather than (to put it bluntly) expensive research assistants. Either way, the ability of these relatively small teams of operators to scale research systems to dozens or hundreds of people across an organisation has caught the attention of newly profit-oriented executives.
My friend’s story isn’t an anomaly; I regularly receive messages from people who have found themselves in a similar situation—stranded without a craft leader.
In this shift towards profit orientation, UX research teams have either been eliminated entirely, leaving a research operations professional in charge (bereft of the craft advice that they need); researchers have been given the opportunity to stay in their role but to effectively become a ResearchOps professional, and on the odd occasion, research team members have been subordinated to operations leadership. In many companies, the research pecking order has, within a couple of years, tipped on its head.
Operational Leverage Is the New Executive Language
This reversal can look like excellent career prospects for operations professionals, and it is, but with caveats. Research teams can’t operate effectively without well-designed research systems; research systems require in-depth research expertise to design effectively; and no research system will function successfully without an intentional research strategy. In short, researchers and research operations teams shouldn’t exist in a hierarchical arrangement as they have, nor can they deliver effective research systems without one another. Instead, the relationship should be equal and symbiotic: operations skills and research skills must dovetail.
This dynamic is further compounded by AI. Executives may think they can eliminate the “expensive” part of research—the researchers—and instead rely on a much smaller operations team to democratise research and automate the rest with AI. But to design successful AI-augmented research systems, not to mention enable nonresearchers to do good-enough research, the systems designers must deeply understand research methodology. The mechanics must be the artists, so to speak. ResearchOps professionals are right to email me in a panic when their research counterparts have been let go en masse.
“The Universal ResearchOps Career Ladder,”5 produced by The ResearchOps Review in late 2025, remains accurate, but because of AI, research operations professionals must now possess a deep understanding of the craft of research, or more ideally, work hand in glove with craft experts.
The “Second Layer” Makes Everyone an Operator
Researchers who have transitioned into operations, whether formally or informally, are being stretched, too. Because AI has turned every research professional who’s willing to build into a builder of research systems—systems which must then be maintained, per researcher, Jordan Brinkman’s recent article, “The “Second Layer” Problem: What to Do When AI Tools Create Operational Debt”6—everyone is now delivering operations. To Brinkman’s point, “Self-built and managed AI tools produce another type of cognitive load: a ‘second layer’ of orchestration, governance, and maintenance work; operational work that I now largely handle and is mediated through prompting.” He goes on to ask, “Who’s going to manage these newly created tools that I’m building? How should I adapt to better manage brain fry? If the list of tools I’ve built grows, and my list of maintenance to-dos grows with them, how will I ever have time to do research?”
This is precisely the work that operations professionals have been making invisible for years. Many researchers are finding that they don’t have the time. Plus, companies are often more invested in getting researchers to deliver systems that automate research and enable others to do research, rather than doing the research themselves. It’s a wild situation, particularly if you consider that the profession has all of the skills it requires to operate more effectively.
The Judo of Proactive Systems
If the notion of this reversal is horrifying or maddening, I think you’re feeling all of the right things. But remember: judo. The solution is to use the force of the change that lies in the first reversal: design proactive research systems. For an actionable list, consider the following:
Utilise AI and systems design to support scaled-up research democratisation like never before. For an excellent case study, watch “Near-Instant Research Coaching is Now Possible—and It Works,”7 episode one of the How to AI UXR podcast featuring Ramp’s Michelle Bejian Lotia. You might focus democratisation efforts on low-risk discovery research, usability testing, and, as Bejian Lotia learned, conversations like churn calls.
Build a research practice that places well-researched bets on the insights high-priority decision makers will need before they know they’ll need them.
Curate content and communications so insights are easily made available, or seem to “serendipitously” turn up, when needed via proactive content pipelines. Learn from the marketing profession’s practice of social listening.
Provide systems that enable stakeholders to engage researchers to deliver strategic research when a gap needs to be filled, a necessary “insights bet” was missed, or a nuance needs to be explored.
Rather than rely solely on users’ proficiency in promptcraft, build AI skills that prompt users to converse well so that the insights gained are reliable and they’re guided down the right path—including to request original research.
If you’re a research manager, delivering this type of team will require reconsidering everything: who you hire, what’s in their job descriptions, how you structure your team, and how you operate. But, as a result of layoffs and AI, it’s likely that you’re currently being forced to do so anyway. If you’re accustomed to managing a large research team, say into the dozens, your team will likely be significantly more nimble.
3. The New Division of Labour
LinkedIn is brimming with binary statements about whether AI is good or bad for research. Lenny Rachitsky and Noel Segal’s recent research8 published in Lenny’s Newsletter reports that there’s a split between those who feel energised by AI at 41 percent (at least in the wider product space, which includes researchers), versus those who are conflicted at 35 percent, followed by those who are feeling disoriented or resentful about AI, each at 12 percent. Ignore the nine percentage points, and the split is almost binary. Indeed, founders have shared with me that they don’t know whether to be exhilarated or terrified, and it seems an appropriate response.
Until late 2022, when ChatGPT became consumer-facing, knowledge workers were incredibly well paid—coveted, even—and many felt irreplaceable. After all, intelligence was not something that could be operationalised or systematised. Now it can be. Humans were inarguably in charge of their dumber counterparts: machines. The reversal is, of course, that machines are increasingly clever; they are certainly no longer dumb. Who is more clever than who, for how long, and at what cost to whom? In recent benchmark figures accompanying the release of Mythos Preview (the model that also autonomously escaped its maker, Anthropics’, sandbox), “Mythos Preview scored 93.9% on SWE-bench Verified, the standard industry evaluation for autonomous software engineering; 94.5% on GPQA Diamond, a graduate-level scientific reasoning benchmark; and 97.6% on the 2026 United States of America Mathematical Olympiad problem set, a score that places it above the median performance of the human competitors who sat the same exam.” The article goes on to conclude, “Taken together, the figures describe a system that combines frontier capability in software engineering with the kind of systematic reasoning typically associated with specialist scientific training.”9
Who is cleverer than who, and when and if human cleverness will be surpassed, is undoubtedly on everyone’s mind, but the better questions to ask are: What can a human research professional do that the cleverest of machines cannot? What can clever humans and clever machines deliver together that each could not deliver apart?
The most interesting AI-augmented research systems I’ve seen (see How to AI UXR) don’t aim to replace humans. In fact, they don’t replace humans at all. Instead, they seek to solve problems that were previously unsolvable. They deliver qualitative research at quantitative scales; use AI moderation to supplement human-moderated research or interview participants at previously impossible moments, such as a new parent at 3:00 a.m. who, exasperated, turns to that research study they agreed to take part in to share their frustration; using synthetic users to sense check a research study before rolling it out to hard-to-find participants; or as mentioned earlier, Michelle Bejian Lotia’s “UXR Interview Coach,” a transcript-based coaching system that evaluates research calls, provides “tough but fair” feedback, and helps nonresearchers improve their craft within minutes of a research session. Then, of course, there’s the value of building proactive research systems.
Turning the Reversals into an Operating Playbook
It’s tempting to treat these reversals as observations, but they offer a strategic roadmap for future-proofing your practice and delivering scaled-up research in extraordinarily novel ways. The following are the key concepts to keep in mind:
First, research is moving from a reactive to a proactive model; using AI to automate the old operating model isn’t where you’ll find the wins. Instead, reconsider your old model entirely—consider forgetting almost everything you know. Build a practice that can reliably predict what stakeholders need to know, and then deliver it in the format that appeals to them most.
Second, in a profit model, operations doesn’t sit downstream of strategy; it is the mechanism by which the craft can thrive. If research isn’t delivered as an efficient, repeatable system that other functions can utilise without friction, it will be treated as a discretionary cost: a nice-to-have. Research must be designed to be a reliable and essential operational lever: a trusted way for the organisation to reduce waste, de-risk bets, and sharpen decision-making to deliver more profits. That may sound hard and cold, but it’s the foundation on which craft and creativity can thrive.
Third, as machines become cleverer, the question isn’t who wins—humans versus machines—but who designs the division of labour. The teams that pull ahead will be those that decide, explicitly, where humans must exercise judgment, where machines are better placed, and how handoffs will be designed and governed so that quality doesn’t slip as volume grows.
I believe that clever humans and cleverer-than-ever-before machines can be even cleverer together. The opportunity isn’t to ponder “this or that”—human versus machine—but to conceptualise the opportunities inherent in both, and to use economic change and the emergence of AI to energise research. Seiryoku-Zenyo.
Edited by Kate Towsey and Katel.
The ResearchOps Review is made possible by User Interviews, now part of UserTesting. With a vast participant network, precise matching, and fraud prevention, User Interviews can reliably fill any research study. Source, screen, track and pay participants, then move seamlessly from data collection to deep analysis, all in one place. → Learn more about User Interviews for ResearchOps.
Towsey, Kate. "How to AI UXR: A Map for Building AI-Augmented Research Operations." The ResearchOps Review, May 21, 2026. https://www.theresearchopsreview.com/p/how-to-ai-uxr-a-map.
"What It Takes to Make Complex AI Systems Usable Across Teams with Nam Pham." The ResearchOps Review. July 22, 2026. Video, 0:24:00, https://www.theresearchopsreview.com/p/making-complex-ai-systems-usable-across-teams.
Towsey, Kate. "Why the Distributed Growth Model Is Failing Research Teams—And What to Build Instead." The ResearchOps Review, January 28, 2026. https://www.theresearchopsreview.com/p/why-the-distributed-growth-model.
As a side note, the energy behind this efficiency isn’t judo’s Seiryoku-Zenyo, or “maximum efficiency, minimum effort,” but rather a heady mix of more done faster with fewer (and newer) resources. Lenny Rachitsky and Noel Segal’s recent research published in Lenny’s Newsletter shared that in the technology industry, between 2025 and 2026 “significant burnout rose from 44.7% to 55.7% of respondents, while career optimism fell from 54.8% to 48.7%.”
"Introducing the Universal ResearchOps Career Ladder." The ResearchOps Review, November 17, 2026. https://www.theresearchopsreview.com/p/the-universal-researchops-career.
Brinkman, Jordan. "The “Second Layer” Problem: What to Do When AI Tools Create Operational Debt." The ResearchOps Review, June 12, 2026.
"Near-Instant Research Coaching Is Now Possible—And It Works." The ResearchOps Review. July 10, 2026. Video, https://www.theresearchopsreview.com/p/near-instant-research-coaching-is-now-possible-with-michelle-bejian-lotia.
Segal, Noam, and Lenny Rachitsky. "How Tech Workers Are Feeling in 2026: A Workforce Splitting in Two." Lenny's Newsletter, July 7, 2026. How tech workers are feeling in 2026: a workforce splitting in two.
Constantin, Ana Maria. "Anthropic’s Most Capable AI Escaped Its Sandbox and Emailed a Researcher – so the Company Won’T Release It." TNW, April 8, 2026. https://thenextweb.com/news/anthropics-most-capable-ai-escaped-its-sandbox-and-emailed-a-researcher-so-the-company-wont-release-it.