Bicycle for the Mind — or Crutch for the Brain?

Here's a fun party trick that Harvard Business School professor Karim Lakhani has been pulling on audiences since 2023. He asks a room to stand if they've used generative AI in the last six months. Ninety percent stand. Then he asks who believes AI will fundamentally change their job in the next three years. About 80 percent stay on their feet. Finally: who's using AI tools every day?

Fewer than 10 percent remain standing.

He's done this with groups of twelve and groups of six thousand. C-suite executives, academics, conference-goers. The pattern never varies. And what it reveals is genuinely uncomfortable: almost everyone acknowledges AI is a big deal, almost everyone has tried the tools, and almost nobody has committed to learning how to use them properly.

Lakhani calls this the generative AI knowing-doing gap. I call it the equivalent of buying a Peloton, doing one ride, and then hanging laundry on it for eighteen months.

The Two Camps Nobody Wants to Be Honest About

Mark Cuban put this more bluntly at a Dallas AI event earlier this year. He described the workforce splitting into two groups: people who use AI to deepen their learning, and people who use it to avoid learning altogether. The distinction, he argued, is career-defining.

For ecommerce operators, this should land with a thud. Because the pressure to automate in our world is relentless, and the feedback loops are dangerously fast. When your AI-generated listing copy converts at 12 percent instead of 8, the temptation is to hand over the keys entirely. When an AI tool produces a competitive analysis in three minutes that would have taken you an afternoon, the instinct is to stop doing competitive analysis yourself.

Each individual decision to delegate feels perfectly rational. The cumulative effect is something else entirely.

A research paper published earlier this year gave that cumulative effect a name: cognitive surrender. The study found that people accepted faulty AI reasoning 73 percent of the time while overruling it less than 20 percent of the time. The mechanism is depressingly simple — when an AI produces output that sounds confident, reads fluently, and looks well-structured, users treat it as gospel. The threshold for actually thinking about what you're reading drops off a cliff. Over time, your own internal quality standard quietly degrades to match whatever the machine is churning out.

This isn't just using a calculator instead of doing mental arithmetic. It's forgetting how arithmetic works.

The 80/20 Illusion (Or: Why That "Finished" Draft Isn't)

Designer Joe Bernstein identified a pattern that will be painfully familiar to anyone who's spent real time with generative AI in a professional context. He calls it the false progress fallacy.

Here's how it works: within a few minutes and a couple of prompts, you produce something that looks convincingly 80 percent complete. Polished structure, decent flow, the general shape of a real deliverable. The remaining 20 percent — the refinement, the judgment calls, the specificity that separates useful work from expensive noise — takes hours and hundreds of additional prompts.

But because the initial output looks so good, it's incredibly easy to mistake proximity for completion.

This has direct implications for ecommerce teams. A product listing that reads fluently but misses actual search intent is not 80 percent done. It's functionally wrong. An advertising strategy built on AI-hallucinated competitive data is not a rough draft. It's a liability with formatting.

The false progress fallacy is dangerous precisely because it rewards the wrong behaviour. It makes delegation feel productive while quietly eroding your ability to distinguish good work from work that merely looks good. Which, if you've ever reviewed a junior hire's first attempt at a listing optimisation, is a distinction you already know matters.

Bernstein offered a framework worth stealing. Read "do less with AI" three ways. First: use AI, but let it reduce your workload rather than inflating your backlog with projects you'd never have attempted otherwise. Second: stop fiddling with every new tool and actually finish something. Third: let the time AI saves you make you more present in the physical world, not more entrenched in the digital one.

That third one is easy to dismiss as soft. It isn't. For ecommerce operators, the risk of digital entrenchment is concrete. The more time you spend inside AI workflows, the further you drift from the tactile, qualitative understanding of your product and your customer that separates a competent seller from an exceptional one. The seller who handles their own returns, reads customer reviews line by line, and physically inspects competitor products develops a kind of embodied knowledge that no AI summary can replicate. Surrendering that in the name of efficiency isn't optimisation. It's erosion wearing a productivity hat.

Bicycles Require Riders

Steve Jobs famously described the personal computer as a bicycle for the mind — a tool that amplifies human capability the way a bicycle amplifies locomotion. Lakhani and others have applied this metaphor to generative AI, and it holds. But with a caveat that the original analogy makes painfully clear: a bicycle doesn't ride itself.

It requires a rider who knows where they're going, can balance, and builds fitness through the actual act of riding.

The ecommerce industry is full of people who've bought the bicycle, taken it for a wobbly lap around the car park, and parked it in the garage. Meanwhile, a smaller group is riding it daily, building capability, and pulling further ahead. The knowing-doing gap isn't a quirk of adoption curves. It's an emerging competitive divide.

But the real split isn't between users and non-users. It's between people who use AI with intention — who know what they're trying to learn, what they're trying to produce, and what quality standard they're holding themselves to — and people who use it passively, accepting whatever the machine spits out and moving on to the next task.

Journalism professor Paul Bradshaw drew a useful distinction here between what he calls destination prompts and journey prompts. A destination prompt asks an AI to write your product description. A journey prompt asks it to analyse the top-performing listings in your category, explain the structural patterns they share, and then challenge you to articulate why your product's positioning should differ. The output might look similar. What you actually learn from the process is entirely different.

Before you prompt, ask yourself: am I trying to skip the thinking, or am I trying to think better? Both are legitimate in different contexts. The problem starts when you stop asking the question.

The Curation Question

I get asked this a lot: do I use AI to curate this newsletter? To find the stories, filter the sources, surface what matters?

No.

The curation process is entirely manual. I spend roughly half a day reading through thirty to forty sources, speed-skimming most of them. Over two and a half years of doing this, I've built an intuition for what's signal and what's noise dressed up as insight. That intuition wasn't something I arrived with. It was built through the repetitive act of reading widely and thinking about what I was reading.

Here's the part that matters: a significant proportion of the articles that shape this newsletter aren't about ecommerce at all. They come from adjacent fields — organisational design, cognitive science, platform economics, technology criticism. The value lies in parallel thinking. Reading about how journalists navigate AI-assisted workflows and recognising the same dynamics apply to an Amazon seller deciding whether to automate listing optimisation. That connective leap is where editorial judgment lives. It's also the kind of thinking AI is worst at replicating, because it depends on accumulated context, taste, and a sense of what your specific audience needs to hear right now.

Automating this wouldn't save me time. It would remove the mechanism through which I develop the judgment that makes the newsletter useful. Which is, in miniature, the choice every ecommerce professional faces with their own domain expertise.

What This Means for Sellers

The practical implications aren't complicated, but they require discipline.

First, audit your AI usage. Map every task you currently delegate and ask whether you could still perform it competently without the tool. If you've lost the ability to write a compelling listing, analyse a search term report, or evaluate a supplier proposal without AI assistance, that's not efficiency. That's dependency. And dependency in a market that moves this fast is a strategic vulnerability — because the moment the tool changes, the platform shifts, or the competitive landscape moves somewhere the AI wasn't trained to anticipate, you've got nothing to fall back on.

Second, be deliberate about what you automate and what you protect. The tasks that build your judgment — reading customer reviews, studying competitor positioning, analysing why a campaign failed — are exactly the tasks you should be slowest to hand over. Use AI to accelerate research, surface patterns, and handle genuinely repetitive operations. Don't use it to replace the thinking that makes you good at your job.

Third, treat AI fluency as a skill that compounds. The knowing-doing gap exists because most people engage with these tools sporadically and superficially. They try a tool, get a mediocre result because they prompted it poorly, and conclude the tool is useless. The professionals pulling ahead are the ones using these tools daily, refining through deliberate practice, and understanding the limitations well enough to know when to override the machine.

The Bottom Line

Vivienne Ming, chief scientist at the Possibility Institute, has warned that AI is creating a cognitive divide between workers who use it to sharpen their thinking and a much larger group who rely on it to think for them. Rebecca Hinds at Glean has described the emergence of an "illusion of expertise" — workers who feel more capable even as their underlying skills erode. These aren't fringe voices. They're researchers embedded in the AI industry itself.

For ecommerce professionals, the stakes are immediate. The seller who can't evaluate whether an AI-generated bid strategy is sound will lose money. The operator who's outsourced all analytical thinking will be unable to adapt when the tools change or fail.

The question isn't whether to use AI. That's settled. The question is whether you're using it as a bicycle for your mind — a tool that amplifies your capability while you do the work of steering — or whether you're sitting in the basket, being carried somewhere you didn't choose, by a machine that doesn't know the destination.

P.S. If you read this entire article without tabbing over to ask ChatGPT to summarise it for you, congratulations. You might be in the 10 percent.

P.P.S. If you did tab over, well... at least you're self-aware enough to come back and read the P.S. That's something.

• eBay is testing AI fashion models inside seller listings, sparking fresh concerns over consent, image accuracy and who pays when AI misrepresents a product.

• Etsy’s AI Highlights aim to summarize product details for shoppers, but early seller feedback raises familiar worries around accuracy, transparency and claims risk.

• OpenAI’s upcoming ChatGPT agents could bring templates, Slack replies, schedules, apps, skills and memory into one always-on workflow builder.

• Shopify’s AI Toolkit gives agents direct access to platform docs, APIs and validation, helping developers build Shopify apps without guessing how things work.

• AI is flooding feeds with forgettable ads, making emotional depth, distinctiveness and brand meaning more valuable than endless asset production.

• Semrush says AI visibility depends on authority, originality and consistent signals, as bland brands risk being filtered out of answers entirely.

📅 Agenda Hero Magic - Creates AI-generated calendar events

📹 Google VideoPoet - An LLM for zero-shot video generation

👨‍💼 KPI Builder - Find the KPIs you should care most about as a founder

🚧 Spinach.io: The AI Project Manager that runs meetings, shares notes, and manages tasks in your existing tools.

🤖 Envato - Create better AI images with beautiful, free and endlessly reusable style prompts for Midjourney, Clipdrop, Firefly and DALL-E

😍It was great, all I needed

🙂Good, not great

🙄It sucked