Welcome back, picnickers.
Last week, we started talking about UX cheerleading — when the purpose of research is diverted away from learning about customers, and towards soothing executives during the “build” stage of the project.
But “build” eventually graduates into “learn.” And the learning tears their false comfort to shreds.
In the Headlines
“We’re finding a lot of AI fatigue among our users,” reads an internal memo.
Adoption of LLM tools has stalled, and sales orgs are fiddling with their numbers to pretend otherwise. Even vibe coding — once the poster child of AI’s transformative potential — has run into the limitations of the technology, with enthusiasm evaporating as teams realize that all the automatically generated code still needs to be maintained by human hand.
This backlash has caught executives with their pants down, because for months and years they were told that AI was the future and everyone was going to love it.
Microsoft AI CEO Mustafa Suleyman finds it “mindblowing” that people hate his bait-and-switch products, which fail at the tasks shown in their own ads. Suffice it to say that customers are less than enthusiastic about having that sub-part experience welded onto even more products.
Dedicated fans are calling Mozilla “astoundingly out of touch” after the abrupt announcement of Firefox’s pivot towards becoming an “AI browser.” Mozilla has since backed down; the AI features are still coming, but they will now be opt-in rather than opt-out.
“It’s a beta” is WaPo’s defense of a dismal AI-generated podcast offering where 84% of the content did not meet the company’s own journalistic standards, but which was launched anyway over objections from staff (I actually covered their AI-focused mission statement nearly a year ago; it’s not surprising that this LLM_first strategy ended up putting customers last).
"I think at this point everyone at the company is more or less OK with the way we’re using it," claims Larian’s new CEO in response to fan and staffer outrage that LLM tools would be used to create the formerly-hotly-anticipated game Divinity.
These are not just cherry-picked examples. Aside from the Microsoft one (which has been a slow motion train wreck for years) these are from the same day last week. I could go find more, but I don’t want to talk about LLMs for the rest of the year.
The important thing is not the technology. It’s the determination of these people to ignore any sign that contradicted the narrative of inevitable success.
The Timeline
This lack of executive foresight is a direct result of what Adam Thomas calls “polite failure” — saying “yes” to your manager’s face and shipping an experience you know users will hate. Because the people dealing with the consequences of those decisions are in some other room. The manager is right in front of you. It is easier not to rock the boat.
And so, product orgs have optimized themselves for polite failure. The sensing mechanisms (the processes meant to do the “measure” and “learn” parts of the SDLC) have been pared down to the bone. The minimum viable, democratized, high velocity research process looks something like this:
Find the noisiest people who agree with you (and don’t worry that they are guaranteed to be outliers).
Take what they are saying at face value (ignoring the nuance of feedback).
Immediately start building (without doing the necessary sense-making to integrate your learnings into the wider body of knowledge).
Test the solution for usability (even though usability without usefulness is good for nothing).
Collect some cool metrics (and don’t worry that they will never be capable of informing a decision)
Find a new job before the chickens come home to roost.
Good Questions
User-centered design is like a court jester (wait, don’t close the tab!)
On the surface, the court jester is entertainment. Ringing bells, telling jokes, and keeping everyone’s spirits up until The Simpsons comes out. But the jester’s real role is to say the things no one else is permitted to say. The jokes are the only way to question the wisdom of the king’s decisions without undermining his authority.
UCD is the same. The job is not to come back and tell everyone “turns out we were right all along.” It’s to identify the problems before they become problems.
Hence, our question of the week:
“If this project fails 6 months from now, what will have been the reasons?
This is where you should be starting. Not mining your data for nuggets that indicate success. But aligning on the signs of danger, and then designing a data collection apparatus suitable for more than just leadership optics.
As an aside, I’m thrilled that I was able to make today’s post fit into the Picnic’s classic section headers. What do you think? Did they help you pick what links were worth clicking? Or do you prefer the contextual ones that were more like an article?
This may be the year’s last issue of the newsletter, so it’s a perfect time to send in your feedback!
— Pavel at the Product Picnic