A few weeks ago, I was deep in a rabbit hole. I'd been analysing the top design leadership subreddits, trying to understand what's actually keeping leaders up at night.
One post stopped me cold. 920 upvotes. The title: "How do I evaluate a portfolio that is 90% AI generated?"
The hiring manager wrote:
"I'm hiring for a Senior UI role. Every portfolio looks the same. Perfectly polished Bento grids, abstract 3D shapes, pristine copy. But when I ask 'Why did you choose this layout?', the answer is vague. I suspect 90% of the work is Relume/Midjourney output. How are you filtering for actual design thinking versus curation skills?"
The top responses were fascinating. One leader said whiteboard challenges are back. Another asks candidates to show "the ugly phase" - the messy sketches, the failed prompts, the bad ideas. A third stopped looking at portfolios entirely. They now pay candidates for a one-hour critique of an existing app.
I'd heard versions of this concern more than a dozen times in the past year. But seeing it laid out like that, with 920 people agreeing - something clicked.
We have a problem. And it's not the one most people think.
The Paradox Nobody's Talking About
Here's what's strange.
Google, Anthropic, McKinsey, Amazon, Goldman Sachs, Deloitte. These companies are betting billions on AI. They're building the tools that promise to make knowledge work faster.
And yet.
When it comes to their own hiring, they're going the opposite direction. Google brought back whiteboard interviews. Deloitte doubled their case study time. Goldman Sachs added more in-person rounds.
Why would companies that sell AI efficiency demand slower, more human evaluation of their own people?
Because they've figured something out that the rest of us are only starting to see.
What the Research Shows
In late 2024, MIT researchers ran an experiment. They gave knowledge workers AI assistants and tracked what happened to their thinking over time.
The finding was uncomfortable.
People using AI showed 20% less activity in the brain regions tied to critical thinking and decision-making. The researchers called it "cognitive offloading." I'd call it something simpler: our thinking muscles get weaker when we stop using them.
This isn't about whether AI is good or bad. It's about what happens when we hand over certain kinds of work without realising what we're giving up.
I noticed this in my own work in mid 2024 before the study came out. When I first started using AI tools, I went all-in. AI-first was the goal. At first I felt like I had super-powers and productive it skyrockets. But after a few months, I caught myself doing something worrying. I was accepting suggestions I hadn't fully thought through. Not because I was lazy. Because my brain had quietly started to trust the machine more than my own judgement.
The research gave me a name for it: automation bias. The more we use AI, the more we tend to trust it - even when we shouldn't.
The impact for me though, it was so much worse than slipping on reading an article, email or proposal.
A sharp very noticeable decline in memory and my creativity - the thing i pride myself on - went from a sharp pencil to a dull one. I was even sent for brain scans by my doctor.
When i finally realised that neuroplasticity, it goes both ways - the old if you dont use it, you lose it... It's actually grounded in science.
It took me months to re-build what I had lost... and it had only been a few months since I lost it.
The Convergence Problem
Here's where it gets interesting for hiring.
When everyone uses the same AI tools, trained on the same data, optimised for the same outcomes - the outputs start to look the same.
That Reddit post captured it perfectly. Every portfolio: Bento grids. Abstract 3D shapes. Pristine copy. Technically excellent. Completely interchangeable.
But the problem goes deeper than portfolios.
If everyone's cover letter sounds similar, and everyone's interview prep comes from the same AI coaching tools, and everyone's case study uses the same frameworks - how do you find the person who actually thinks differently?
HR leaders are starting to ask this question out loud. One told me recently: "AI can help someone perform at 90% of average. But we're not hiring for average. We're hiring for the extra 30% that only humans bring."
That extra 30% is the part that's getting harder to see.
What "Extra 30%" Actually Means
It's worth being specific here.
When I talk to hiring managers, they describe it in concrete terms:
- The ability to challenge a brief, not just execute it
- Knowing when to break a pattern rather than follow it
- Reading the room in ways that don't show up in data
- Making ethical calls in grey areas where there's no clear answer
- Building trust with difficult stakeholders
These aren't nice-to-haves. They're the things that separate competent work from work that actually moves things forward.
And here's the uncomfortable truth: these skills don't get better when you use AI more. They get better when you practice them directly.
A Lesson from the Justice System
I want to share an example that's stayed with me.
In the United States, courts use an algorithm called COMPAS to help predict whether someone is likely to reoffend. The idea was to make sentencing fairer by removing human bias.
The system was designed with human oversight. Judges were told to use the AI score as one factor among many.
What actually happened? Judges started relying on the scores more and more. When the AI said someone was high-risk, they went along with it - even when their own experience suggested otherwise.
A legal scholar called it "the illusion of objectivity." The AI looked so precise, so scientific, that it became hard to question.
The result was bias in a different form. The algorithm had been trained on historical data that reflected existing inequalities. So it learned to rate Black defendants as higher risk, and white defendants as lower risk - even when they weren't.
Human oversight was there. But the humans had stopped really thinking.
What the Smart Companies Understand
The organisations going slower on hiring aren't anti-AI. Many of them are building the AI.
What they understand is this: some decisions need more human thinking, not less. Hiring is one of them.
When Google brings back whiteboard interviews, they're not being nostalgic. They're creating a situation where candidates have to think in real time, without AI assistance, in front of people who are watching the process - not just the output.
When that hiring manager asks for "the ugly phase," they're looking for evidence of struggle. Struggle is a sign that someone actually wrestled with the problem rather than just generating a polished answer.
Gartner predicts that by 2027, 40% of organisations will require AI-free assessments for roles involving important decisions. That's not paranoia. That's pattern recognition.
The Shift That's Coming
I think we're going to look back at this moment as a turning point.
For the past two years, the question has been: "How do we use AI to go faster?"
The next question will be: "Where do we need to go slower on purpose?"
Not because speed is bad. But because some things only develop through friction. Critical thinking. Judgement. Ethics. The ability to sit with ambiguity.
I call it - strategic friction. Knowing when to build it into - to your workflows and to your customer's experiences. That is the challenge now - reversing 10 years of design convention - trying to optimise, trying to remove friction.
These are the capabilities that separate someone who can execute from someone who can lead.
That is why I teach designer that the PAUSE IS WHERE THE VALUE IS.
If it werent - you're job could be automated away.
What This Means for You
If you're hiring: consider where you might be screening out the signal you actually need. Perfect AI-assisted outputs might look impressive, but they don't tell you how someone thinks under pressure.
If you're a candidate: the best investment you can make isn't mastering the latest AI tool. It's building the skills that can't be automated. Show your messy thinking. Explain your trade-offs. Demonstrate that you can disagree with a brief when it matters.
If you're leading a team: start asking where you need deliberate friction. Not everywhere. But in the places where judgement matters most.
The smartest companies are already doing this. They're building moments into their processes where humans have to think - really think - without AI assistance.
Not because they don't trust AI.
Because they understand what makes humans valuable.
I'd love to hear what you're seeing in your own hiring or job search.
Are portfolios getting more homogeneous?
Are interviews changing?
What's working and what isn't?
Let me know - riley@ai-flywheel.com