AI got fast. The learning didn't come with it.

AI can now write a research report, sort your data, and spit out twenty mockups before your standup ends. That speed is real, and it's not going away. But the leaders who actually run design and research teams are landing on the same warning: the fast output is the cheap part. The thinking that used to come with it is the part you can lose without noticing. Let me catch you up.

The report shows up, the learning doesn't

When AI runs your interviews, sorts your data, and writes the summary, your team gets a clean report. What they don't get is the wrestle that made the old reports stick. Maria Rosala at NN/g calls this the self-generation effect: people remember what they worked out themselves far better than what they read. Hand off the interviewing and synthesis, and "what's left is consumption."

There's a trap inside that. A tidy report creates what Rosala calls an "illusion of learning." Your team reads it, feels informed, and moves on, the same way a student who rereads notes feels ready for a test and then bombs it. They retain about as much.

So the risk isn't a bad report. It's a good report that convinces everyone they learned something they didn't.

Stories move people. Summaries don't.

Rosala tells one about a man who couldn't read as a kid, who watched his mother lose herself in a book and wished he could have that too. She heard it live in a session, told her husband, and they donated to the literacy charity that helped him. That's what research does when a human is in the room. The findings changed the people who witnessed them.

A synthetic transcript can't do that. Synthetic users tend to be "overly coherent, emotionally flat," and the whole point of research, in Rosala's framing, is "the transmission of human experience into design decisions." A senior researcher she quotes put it plainly: "We were all there. We all saw it. We all heard it. We all lived it." That shared moment is what pushed the design forward. A PDF doesn't.

The screen was the hiding place

Here's the harder part, and it's about your designers, not your researchers. Jessa Parette, Head of Design at Byte by Yum!, argues design spent a decade optimizing for the exact thing AI just made cheap: fast, consistent screen production. Velocity was the metric. "The designer who could turn around ten high-fidelity mockups by Thursday was the one who got promoted."

AI automates the lowest-value work first, and for design that was the UI. Parette is blunt about what got exposed: "If the mockup was the thinking and not the artifact of the thinking," then AI didn't take a tool, it took the thing people were hiding behind. She's watched talented designers freeze, and the fear is specific. Not "my tools are changing." It's "I don't know what I am without the tools."

What stays scarce is judgment. What to build, what to kill, what breaks at scale before it visibly breaks. That can't be prompted.

Fast, cheap, good, pick two, and say which

Everyone wants speed. The useful move is getting specific about which speed. Jen Dunnam, now VP of Futures at Hinge, points at the old tradeoff: fast, cheap, good, you get two. Leaders want all three, so the real work is "having hard, honest conversations about that tradeoff." At Shopify, speed meant rebuilding a core product. At Patreon, it meant shipping a lot of MVPs. Different jobs, different speed.

Dunnam also reframed her own role. Some peers think their job is to inject urgency. Her good designers already bring plenty, so her job is to steady things: "break it down, strategize an approach, play with the right tools, and sharpen our principles."

And she hires for one trait above the rest now: critical thinking. In interviews she probes for it hard. Where did they dissent? How did they push back? What are they still upset about that didn't make the product?

Cut the busywork, protect the sense-making

None of this is an argument against AI. Rosala offers a clean test for any task: does doing it teach the team anything? If no, hand it off. Recruiting, scheduling, transcribing, cleaning raw data, formatting a report once the thinking is done. That's support work. Let AI have it.

Protect the parts where the learning lives: moderating sessions, watching them live, debriefing together, deciding what the data means. As Figma's leaders put it after Config, "As more gets built automatically, human judgment becomes the thing that matters most." Airbnb's Teo Connor expects a stretch of slop first, but "the great work will always rise above it."

Collaboration is the other thing to defend. AI tools pull people into solo workflows. The leaders at Config are doubling down the other way, on crits, feedback in the open, debating outcomes as a group.

The deep cut

Watch what your performance reviews reward. Parette's regret is precise: she sat in reviews that were entirely about delivery. How many products touched, how fast designs moved to handoff, how consistently the design system got used. "Rarely did anyone ask: did we design the right thing? Did we see what was coming before it shipped?" The team had language for production and no language for prevention.

That's the concrete fix. If your review criteria still count screens shipped and speed to handoff, you're paying people for the work AI now does for free, and you're training your best people to compete on the dimension that's getting cheaper. Rewrite what you measure. Reward the dissent, the caught risk, the thing killed early. Otherwise the muscle Parette says already atrophied keeps atrophying, on your watch.

Three questions for your team

1. Look at your last research handoff to AI. Did anyone on the team actually watch a session live, or did everyone just read the summary? If it's all summary, what do you change next round?
2. Pull up your review criteria. How many of them reward production speed versus judgment, dissent, and prevention? Which line gets rewritten before the next cycle?
3. For your current top priority: fast, cheap, or good, which two are you actually choosing, and does your team know? If you can't name the two, you owe them the honest conversation this week.