The Map Is Not the Terrain: What the AI-Native UXR Frontier Leaves Out

A response to Jess Holbrook's "Frontier UX Research Circa May 2026."

Jess Holbrook just published the clearest map of AI-native UX research I've read (Holbrook, 2026). If you lead a research team and you haven't read "Frontier UX Research Circa May 2026," go read it first, then come back.

He gets the big thing right: being AI-native is a systems problem, not a tools problem. Buying Dovetail or shipping one Claude project won't make you AI-native. You have to redesign how you work — how evidence flows from intake to insight, collection to synthesis to decision, so AI can participate at every step. I've been making the same argument to researchers and clients for a year, in nearly the same words: context engineering over prompting, systems over tools, amplification over automation. Human judgment and growth are the things you protect.

So this isn't a rebuttal. It's an extension. A “yes! and…” if you will.

Because here's what a map can't show you: a route that looks short on paper can wreck you on the road. The frontier isn't just a destination you navigate to. It's a distance you have to stay conditioned to keep riding.

Two things the map leaves out — the terrain, and whether your team is fit enough to finish. Let's take them one at a time.

THE MAP IS NOT THE TERRAIN

But first, a detour down memory lane.

In 2011, I completed the AIDS Lifecycle—a 545-mile bike ride from San Francisco to Los Angeles—to raise awareness and funds for the San Francisco AIDS Foundation. On day 5, we rode from Santa Maria to Lompoc—just 42 miles. Short in comparison to the 104-mile day. A breeze, right? Think again.

I have never cursed so much in my life. What appeared on the map to be simple turned into a hilly slog that left my body and spirit in shambles. The map gave me the direction. It said nothing about the Twisted Sisters climb and the compounding exhaustion from riding 4 days non-stop.

This is all to say, maps like Jess’s are mostly descriptive—and I say "mostly" because he does point a direction, with some light guidance on how to get there. But like any map, it tells you where, not what it'll cost to get there. His article takes the 35,000-foot view. It clearly shows the vision of the frontier and gives some turn-by-turn directions, but it stops short of what it actually takes to move a real team — with real people, real anxiety, and real Tuesday-afternoon problems — from where they are to where he's pointing. (And that’s likely by design. After all, that’d be a novel—not an article.)

And teams don't start from the same place. Different sizes, maturity, and pressures. Different access, aptitude, and attitudes toward AI. All of which change the route.

Watch how the hardest moves get compressed into a clause. "Formalize a small research engineering function — could be fractional." "Invest in AI literacy training." Each of those is a fraction of a sentence carrying a mountain of change management.

I don't disagree with either of these recommendations (I 100% agree with both). The line I nodded hardest at was his case for a research engineering function—it's a flag I've planted for a while: every team needs a dedicated research-AI engineer, ideally a Staff- or Principal-level researcher who's become AI-fluent, not an engineer learning research on the fly.

Fractional is a fallacy.

"Could be fractional" compresses a hiring fight, a reskilling plan, and a headcount defense into three words. Same with the "specify the what, let AI plan the how" model he borrows from spec-driven development. It works — if you have a researcher fluent enough in their craft to tango with the AI and stay in the driver's seat. That's a high bar, and it's not where most teams are standing. Anyone who’s tried doing AI enablement on the side of their full-time day job knows that’s a fast road to burnout (more on this in a bit).

This is part of why 95% of generative AI pilots never reach production (MIT, cited in Holbrook 2026). Not because teams lack the map or an idea of the destination, but because the actual journey from the map to a working team is made up of a hundred unglamorous operational decisions with little direction or training, and they definitely don’t want to spare a resource or pay for it.

AI Literacy & Fluency is complex.

"Invest in literacy training" quietly assumes you already know where your team's literacy starts. But every team I’ve worked with spreads the gamut, with too few at the top of AI fluency (if at all).

Here's one from the field. This May I ran a week-long AI hackathon with the research and design team at a major financial institution — 30 people, a custom agent I'd built for the engagement. The plan assumed a certain starting line. Day one, a sharp, experienced staff researcher asked, "What's an .md file?" And that was the one person brave enough to say it out loud. Plenty more were in the same boat, quietly wondering what a "skill" even is and how it maps to their work, let alone their day-to-day.

The reality of AI literacy is buried deep in the assumptions in Jess’s article. The literacy gap is bigger than most enablement programs, research leaders, and execs assume. True literacy doesn’t live in a self-reported survey, but instead in the practical knowledge of why an AI hallucinates and more so how to avoid it (as best as possible). The stumbles still came on the pebbles — the small, unglamorous obstacles that litter the path.

That's the gap between "here's the frontier" and "here's how your team gets there." It isn't a knowledge gap. It's an operational one. And it's where the actual work lives.

SCALING & CLIMBING DEPEND ON your Conditioning

Knowing the terrain is only half of it. The other half is whether your team can keep riding — next quarter, and the one after. This is the part I'll push hardest on, partly because it's personal.

Veterans of the AIDS Lifecycle would tell you: “Your ride depends on the quality of your training.” I started training 8 months in advance, and my focus during that time was twofold: I had to develop stamina to endure 7 days riding on a bike across all manner of California terrain and the skills to do so effectively and safely.

Stamina COMES THROUGH sustainability.

There was a girl on my team who literally did almost no training. She pushed through on sheer determination and riding the sweep van at the end of the day that picked up straggling riders. My sustainable approach toward regular training allowed me to ride all 545 miles and cross the finish line with pride. She crossed the same finish line as me (mainly because no rider was left behind) but with a little less pride and a lot more injuries.

Jess puts AI fatigue and burnout in the final section — "Risks, traps, and dead ends." 14% of AI-intensive workers report "AI brain fry" (BCG/HBR, via Holbrook 2026). Teams lose 40% of their AI efficiency gains to correcting and fact-checking AI output (Workday, via Holbrook 2026). NN/g declared 2026 "the year of AI fatigue."

Jess names something important here, almost in passing: the source of this fatigue is organizational, not individual. I'd go one step further. If burnout is structural, you can't bolt awareness of it onto the end as a risk.

That's the approach most teams take, and it wastes time, money, tokens, and people. Harnessing the power of AI requires intentionally designing for it—with it—from the start. Thinking programmatically and systematically over the tools and technology. Retrofitting AI onto old systems, at the speed the field is moving, doesn't just produce poor results. It burns out your most AI-forward researchers (speaking from experience).

The road to AI-native isn't a sprint. It's a marathon. The long-term winners aren't the teams that sprint to ship an agent this quarter to check a box in a strategy doc. They're the teams still doing rigorous, AI-augmented work two quarters from now without grinding down the seniors who hold the craft. Going AI-native is as much a sustainability problem as it is a systems and infrastructure one. It takes resilience.

(RE)SKILLING FOR THE ROAD AHEAD

Prior to the AIDS Lifecycle, the last time I had been on a bike was nearly 15 years prior when I was 15 years old. Riding a huffy to the city pool is different than a road bike at 40 mph down a winding mountain. I had to learn all new skills: how to clip in and out of the bike, brake properly, signal and communicate with bikers on the road, let alone how to tune my bike or—God forbid—change a flat tire on the side of CA 101. I had to develop all of these skills, the mechanics, and muscle memory in order to actually ride effectively and safely. The same is true with AI.

Jess argues that the frontier of UXR work depends on a new function: the "Pattern Skeptic" — the researcher whose job is to challenge AI output rather than produce it. The problem is that this depends on two fatal assumptions:

  1. This only works if the researcher’s judgment stays sharp. A skeptic who hasn’t internalized the data first-hand can't actually be skeptical. AI will lie to you with conviction, and you need to have watched the sessions yourself to call bullsh!t.

  2. The continued development of a talent pipeline to build the cognitive skills and craft needed to become the pattern skeptics who hold AI accountable in the future.


Staying Sharp

Discernment is a muscle, and it comes from reps. Derek Thompson put this better than I ever could:

“Use AI for the jobs in your life.
Don’t use AI for the gyms in your life.”
— Derek Thompson

“A simple way to figure out whether to use AI at work or in life is to think about the difference between a gym and a job. At a gym, the point isn’t for the weight to be lifted, but for you to lift the weight. At a mere job, however, ‘the point is for the weight to be lifted.'“

That isn't just a tidy metaphor — it's the finding of a 2026 Microsoft and Carnegie Mellon study of 319 knowledge workers (Lee et al., 2026). The more people trusted the AI, the less critical thinking they reported, and their effort shifted from doing the task to "stewarding" the AI's output. The authors land on Lisanne Bainbridge's old "Ironies of Automation": mechanize routine work, and you strip away the everyday reps that keep judgment and critical thinking sharp, leaving people's cognitive musculature atrophied until a real exception shows up. (The study leans on a 45-minute self-report survey, so I'd hold the exact size of the effect loosely — but the direction matches what I see in the field.)

So if your journey to AI-native quietly removes the reps that build judgment, you hollow out the one human capability the whole system is built around. The real threat to research teams was never automation. It's atrophy—of craft, of critical thinking, of the people doing the skeptical work, and the future of our profession.

Here's how I make that concrete. In my AI transformation workshops, I have teams map their process and hunt for time-sucks—then sort them. Some time-sucks are necessary: rewatching a participant video, taking notes by hand, sitting with the raw data until the pattern earns itself. Those are gym work—the point is the rep, not the output. Others are unnecessary and purely on the job end of the spectrum: formatting transcripts, first-pass tagging, reformatting a deck. Hand those to the AI. Sustainable capability means intentionally choosing which work stays human— not because AI can't do it, but because it shouldn’t. Doing it by hand builds the researcher muscle and keeps the team's judgment alive. A frontier team knows the difference and budgets for it.

Preserving our Profession

This hits juniors hardest, and it's where the frontier conversation — and Jess's article — goes quiet.

And it's not hypothetical; the bottom rung of the UXR career ladder and talent pipeline is already being sawed off. Stanford's "Canaries in the Coal Mine" study found a 13% relative drop in employment for workers aged 22–25 in the most AI-exposed occupations since late 2022, while experienced workers in the same jobs stayed flat or grew (Brynjolfsson, Chandar & Chen, 2025). Tellingly, that decline clusters in roles where AI automates rather than augments — in augmentative roles, employment grew across all ages. The Burning Glass Institute found the same pattern in the postings themselves: the share of job listings asking for three years' experience or less fell from 43% to 28% in software and 35% to 22% in data analysis between 2018 and 2024 (consulting dropped too, 41% to 26%). In our own field, Google Cloud reportedly cut every researcher below L6 on certain teams — the junior tier, erased in a single reorg (Reddit).

Here's the trap inside the trap. The same shift is automating the work juniors learned on. 76% of researchers now use AI for analysis and 57% for transcription (Venkat, 2026) — the first-pass tagging and coding that used to build a researcher's eye. So if AI drafts the themes a junior used to build by hand, where do they get the reps that become judgment? You can't grow a Pattern Skeptic who never had to find a pattern.

My answer is the un-sexy one: a paired apprenticeship model. Put the junior alongside a senior on AI-assisted work, make the critique visible, and treat "evaluating the AI" as a craft you coach rather than a checkbox you tick. That's the kind of operational detail that never fits in a frontier checklist — and it's exactly what decides whether you have a pipeline in three years or just a few exhausted seniors.

Put the two halves together, and you get the whole point: sustainable means your people don't burn out, and your skills and your talent pipeline don’t atrophy. It's why, in the framework I use with clients to evaluate AI-assisted workflows — HEARTS, for Human-led, Experience-focused, Amplifying, Rigorous, Trustworthy, and Secure & Sustainable — that final letter carries real weight (more on this next week). Sustainable isn't only about energy cost or compliance. It's whether the way you've engineered the work is something humans can and want to keep doing.

The so (now) what…

Jess gave the field a real gift: a clear, honest, current picture of the frontier, built transparently with AI and disclosed end-to-end (huge kudos for AI usage disclosure at the end, complete with prompts!). The map is good, and I'll be pointing people to it.

Two things that I would humbly add and underscore.

  • First, a map is not the terrain. Your journey to the frontier will vary wildly. The hard part is the operational work of shepherding real people in the real world, and that work doesn't fit in a footnote.

  • Second, scaling & climbing depend on your conditioning. Sustainability and skill development (especially for Jr researchers) can’t be the last bullet on a risk list. It's a design constraint and principle to be proactively built from the beginning. Build for the team you'll still have in a year.

“Chasing efficiency commoditizes research. Chasing critical thinking compounds it. ”
— Me (Kaleb Loosbrock)

I finished those 545 miles in 2011—and the easy-looking 42-mile day was the one that nearly broke me. Completing the Ride is one of the biggest accomplishments of my life. Not because the route was long. But because I slogged through it all, climbed hills that nearly broke me, and crossed the finish line.

The frontier Jess maps is real, and it's worth riding toward. Just don't mistake the map for the road, and don't start the ride ill-prepared and conditioned to quit rather than finish.

What's the one part of your research craft you'd refuse to hand to AI—even if it could do it—because you need the reps?

Sources

AI Disclosure

In the spirit of practicing what I preach, here's how this one came together.

This was a human-led, AI-assisted piece. I used Claude Opus 4.8, in Anthropic's Cowork app, as my main research and drafting partner: it pulled together my own Readwise highlights and margin notes on Jess's article and the Lee et al. study, helped me gather and format sources, and worked with me through several rounds of structural editing. I then used Google Gemini to independently double-check and verify every source and citation, and Grammarly to tighten the grammar.

The argument is stories, the workshop moments, the frameworks, the point of view, and the final word on every line. I reviewed each source myself, edited everything into my own voice, and made the calls on what stayed and what got cut.

Using AI as a collaborator let me turn my own reading and notes into a tighter argument, faster than I could alone (what easily would have taken a week, took a day) — without handing over the judgment ;-). Which, fittingly, is the whole point of this piece.