Faster, Not Lighter: Living With AI Brain Fry

Welcome to Think First: Perspectives on Research with AI. Across this season we’ve watched researchers rebuild their workflows, reinvent evaluation, wrestle with synthetic users, and sort out what AI changes for careers at every stage. Here we are asking the most human question of all: what does it feel like doing this work now that AI accelerates everything? The shorthand our researchers keep reaching for is “AI brain fry.”

Two researchers took it on. Reeta Banerjee, a Principal UX Researcher at Microsoft AI, is the reframer – equal parts realist about the fatigue and evangelist for grace, community, and AI as a first-principles thought partner. MJ Kwak, an AI Experience Architect at XD Studio, is the systems diagnostician, precise about where the pressure actually comes from and candid about its toll.

They land in the same place coming from different starting points. AI brain fry is real – but it is driven less by the technology than by the culture and expectations that have grown around it. And the way through is not to work faster; it’s to be more intentional, more forgiving, and clearer about the durable human core of the craft.

The synthesis machine

MJ names the origin bluntly based on observations from both within and outside Microsoft. In an interview a year ago, she’d asked leadership to “stop turning researchers into a synthesis machine,” and she stands by the diagnosis: “the expectations have shifted faster than the technology itself matured.” Suddenly researchers were expected to “synthesize overnight or within days instead of weeks,” which is how it started to feel like “we were expected to work like a machine, not as a human.” She notes that the tools have since improved, but expectations were set before the technology could catch up.

“There are more and more demands on researchers right now in the age of AI,” Reeta says – a reflection of how fast the industry is moving. Yet she resists that premise, noting that “research exists not just as a synthesis machine, but also as an insights function and a decision-making aid,” helping leaders “see around a corner that they may not have foreseen.” Her conclusion is almost defiant: “research is more important than ever” – it’s the toolset that has expanded, not the mission that has shrunk.

The manual work that didn’t disappear

Both push back hard on the myth that AI simply halves the work. MJ estimates that manual verification, once 40 to 50 percent of a project, has fallen to “30 to 40 percent” as accuracy improved – but no lower, because the standard is higher than mere synthesis: “sense-making goes into the workload.” Her discipline is to be “very intentional about where I let AI move fast versus where I slow down.” AI takes the first pass, never the final word: “I still go back to the source materials, especially when the work involves customer signals, sensitive data, PII, or anything that could shape leadership decisions.” The result, she admits, is that the workload “wasn’t really lighter” – in some ways it “was harder on ourselves.”

“Anyone who’s actually tried to use AI realizes there’s a lot of review needed,” Reeta says – the “AI just replaces what we do” story collapses on contact, giving way to iterations, correction for hallucinations, and constant learning as the tools change. AI is genuinely “accelerating our work and giving us the ability to do more, scale at a faster pace,” but the manual layer doesn’t vanish; sometimes it just moves – into review, or into “creating agents that review that work.” She points to one example from a company-wide UX Research “Agents and Skills Bonanza” she co-organized: a teammate showcased an agentic system he had built with a self-improvement loop that “reviews its own work and improves upon it” – the kind of step-level gain that compounds.

The time you save – and who claims it

“Honestly, both happens,” MJ says of whether the time AI frees becomes thinking time. “It does free up the time, but at the same time it gets filled with more projects.” That, for her, is “one of the biggest tensions right now”: “the time saved is real, but it is not automatically protected.” Reclaiming it takes deliberate work – being “extra mindful,” embedding strategic thinking into her processes, and “setting that expectation with leadership” rather than hoping the space survives on its own.

Reeta reframes the accounting entirely: “AI is not just about saving time. It’s about expanding what’s possible.” On a fast-moving visioning initiative initially proceeding without new research, AI didn’t just speed up her usual work – it let her do work she couldn’t otherwise have done at that scale in that timeframe. She proposed and led a study with 50+ users, pairing human interviews with AI-moderated tools and accelerating from research to report in just two weeks. Her research-based user framework helped shape the team’s vision and the framing for two executive presentations – impact that wouldn’t have been feasible without AI. But whether AI makes the work better, she stresses, “depends entirely on how we apply it” – responsibly, additively, with real guardrails. She invokes “what Jess Holbrook and Judd Anton have called AI research slop”: the researcher’s job is knowing “where to bring AI in, and where to step in to make sure we’re bringing the full context,” so speed never quietly becomes sludge.

The technology, or the culture around it?

So is AI brain fry a technology problem or a cultural one? Both were clear: the technology supplies the possibility, but the culture is what can convert it into burnout.

“The technology creates the possibility of faster work, more outputs, and more cognitive switching,” MJ says, “but the culture decides whether that becomes better work or just more work.” Brain fry sets in “when we use AI to move faster without changing the expectations around volume, attention, and quality.” She locates a genuinely new kind of load in it: regular burnout comes from too much work, but AI burnout adds a layer – “you are not just doing work, you are managing the work of the machine, checking its reasoning, and protecting the quality of the human judgment around it.” And a warning: “if everyone is just summarizing but nobody is actually reading, we lose the empathy and depth that make research valuable.”

Reeta agrees this wave of change “feels more dramatic and is unfolding faster” than earlier technological shifts, and puts her finger on the same compounding load: AI burnout is “the existing load of all the work” plus “the need to learn all the new things about AI and immediately gain value.” As she puts it, it can feel like doing your regular job while “getting a master’s degree in all of these new AI tools on a weekly, daily basis.” Her team’s answer is structural – dedicated learning time they call “The Royal Academy of AI,” and a cross-company UXR AI tools and workflows learning series she co-leads – because “having those dedicated times and spaces to learn and share with each other” both accelerates the work and lets people “empathize with and help each other.”

The ducks paddling underneath

Reeta sees a strong analogy for this moment: Duck syndrome, a concept she first heard about at Stanford. Everyone around you looks like they’re “smoothly sailing on the water” – a new agent here, a slick workflow there – but “underneath the surface, people are paddling rapidly.” The antidote is grace: “remind yourself that most of the people you’re working with are also feeling like that.” She insists the overwhelm isn’t only negative, either. There’s a “positive overwhelm” too – the “wow, I cannot believe I could do that” of watching hours of work collapse into minutes – which she likens to “coming back from Disneyland” dazzled and a little dazed. She notes, “Even the most amazing experiences can take time to process and fully take in.”

MJ is candid about a quieter cost. When AI first entered her day-to-day, “it amplified my imposter syndrome” – “is my judgment good enough? Is this accurate?” – and she found herself “second-guessing and bringing myself down.” What pulled her out was mapping her own value against the machine’s: deciding, deliberately, “what are my strengths that I can amplify instead of just looking at the weaknesses AI could bring out.” She may not connect the dots as fast as AI, she says, but she brings other strengths “that could continuously be valuable as this AI matures.” Underneath the vertigo of the moment – “we don’t know if we’ll be researchers in five to ten years” – is a practice of steady self-evaluation.

Closing the season

Across five episodes, the same figure keeps surviving every disruption: the researcher whose judgment, empathy, and appetite for learning outlasts any single tool. Reeta puts the season’s quiet thesis best. Researchers’ “entire job is to learn,” she says, which makes them “especially equipped” for this moment even though it may not feel that way. “The core skills you’ve developed – the ability to learn, to empathize, to understand people and context, and to solve complex problems – have longevity far beyond the length of life of a certain technology.” The work now is to hold onto that while the ground keeps moving: to be intentional about where the machine helps, generous with the ducks paddling beside us, and clear-eyed that faster was never the point – better was.