The Seduction of Efficiency

Don't let the backwards cap fool you. Harley Finkelstein delivered a stunning set of earnings results, his physical being now so incredibly polished that he appears not to be human. Tobi always appeared like this. Yin and yang, so frequently a great combo, and with Shopify now the largest Canadian share on the market, we are watching this leadership duo develop inside one of the perpetually exciting technology companies around. Whether they, like Tim Cook, eventually move operations to the US is a question for another day.

Harley posted those results less than 24 hours after Tobi released, via Twitter (I continue to refuse to call it by its new name), the new agentic set for Shopify. A world reimagined where agency is passed to the vendor. Removing the agency from agencies, while depending on agencies to separate such duties. Isn't that what an agent should do anyway? Answers please.

Efficiency über alles. Can I say this? Is the reference too much? It is becoming the mantra through which all commerce is delivered, but is it right? Let's reframe: who is it right for? I spoke at length with my trusted friend Hendrik Laubscher, arguing, poorly, my case for us consumers. We do not need efficiency. We will find a way to not be efficient; we always do. At least this is the Irish way. Once a rule is set, we find a way around it. Efficiency will crush our spirit. But to make ecommerce work in the long term, maybe there is no other way.

Put yourself in your own shoes for a minute. The average person interacts with their phone 55 times per day; for the next generation, closer to 100. In a 16-hour day, that means whipping it out every 17 minutes or so, across roughly nine apps: email, banking, photos, and social doing the heavy lifting. We want our kids with less screen time, more eyes up here. Maybe this is why Jony Ive has joined OpenAI: the vehicle for agentic success needs to be more humanistic, more habitual, because the iPhone points in the exact opposite direction to the agentic train. That train, though, has already left the station.

The long kiss goodnight

Ecommerce has long prized efficiency: faster fulfilment, instant recommendations, automated service, reduced errors, always-on operations. With agentic AI, autonomous systems that plan, decide, and act without human oversight, these efficiencies reach new heights, allegedly freeing humans for more "meaningful" work and maximising profit.

But this prompts a critical pause: who does this efficiency actually serve? The customer, the worker, or mainly the shareholders and platform owners? And crucially: do humans really need or want efficiency in every part of their experience?

One dumb example that gets me is payments. Visa and Mastercard dominate; somewhere in the background lurks JP Morgan, also trying to decouple the phone. Say I have an agent managing my ecommerce payments. That agent might know too much about me to let me complete a purchase. Agent says: Vinny, you are overstretched, we are not releasing more credit, despite its availability. Or my agent says yes, we will handle the risk, but the website's agent says no. Rules might govern this, but who do the rules serve? (Worth following the Retailgentic Substack by Scot Wingo on all of this. Exceptional stuff.)

"Not I," said the fox: who really benefits?

Agentic AI overwhelmingly benefits the enterprise: reduced labour costs, 24/7 operations, instant scaling, maximum throughput. Shareholders gain higher margins. The organisation achieves agility and resilience unmatched by human teams. Consumers get faster, error-free service and ultra-personalised recommendations, but their experiences often become more functional, less memorable, more transactional. And workers, especially in platform and gig roles, face stricter algorithmic oversight, loss of autonomy, and degraded work quality.

Efficiency is fantastic when you're a spreadsheet. But when you're a parent ordering a birthday present, or a worker seeking meaning, ruthless efficiency can feel like a cold, algorithmic wall.

In the path to profitability, though, is there a better way? What got us here will not get us into a better future, business-wise. Profits have long been craved as costs remain crazy.

Why do we crave more than efficiency?

Human experiences aren't always improved by efficiency. In shopping, in hospitality, in digital communities, the moments we remember are the ones that feel personal, surprising, and imperfectly human, not the ones optimised to the nanosecond.

Recognition over reduction; understanding over optimisation. Efficient systems reduce us to derived preferences and predicted behaviours rather than seeing us as individuals. Algorithmic systems can produce personalisation at scale, but not the rituals, meaning, or community people cherish. And ultra-efficient systems erode the texture of daily life, squeeze workers, and diminish the serendipity that leads to true loyalty. The inefficiencies, the moments of waiting, surprise, or even error, are often where recognition and belonging arise. This is what made eBay great back in the day.

Agentism: there is an app for that

A wide, broad, deliberately unassociated leap: iPhone apps. Roughly 2.9 million new apps have been added to the App Store since 2020, about 1,615 per day. And then reality arrives:

68% of all apps never reach 1,000 downloads. 77% of users abandon apps within the first three days. The median monthly revenue for an app after 12 months is less than $50.

Retention tells the same story: 25.3% of users remain at day one, 12% at day seven, under 3% after six months. Of 100 people who download an app, two or three are still using it half a year later. The market reached $150 billion in consumer spending in 2024 while 99.5% of consumer apps failed to achieve meaningful success. The volume game pays the platform; almost nobody else.

So what is my hypothesis? Let's start with: I am not terribly smart. But behaviour, en masse, tells us we are distracted, distressed, and unable to create even minimal utility from the technology we already have. How on earth do we think agentic futures are what we are geared for?

Broken data, bad decisions

Even if you want to deploy an agent, you probably shouldn't. A staggering 78% of organisations aren't ready for agentic AI. Their data is fragmented, outdated, poorly tagged, and wildly inconsistent, which means any agent trained on it is like a six-year-old making dinner from a half-burnt cookbook. This isn't a technology problem. It's an attention problem. A human behaviour problem. A "we forgot to do the basics" problem. When executives wax lyrical about autonomous personalisation, remember: under the hood, it's still chaos.

From Moore's Law to Cher's Law

If you know me by now, you know I am a sucker for Clueless. If Cher Horowitz had a digital shopping agent, you can bet it would be serving her Prada she already owns, gluten-free granola she never ordered, and asking if she wants more beige when she clearly said blush. And that, dear reader, is the crux: we're living through the dawn of agentic AI, an era where autonomous agents promise to optimise everything from our ecommerce journeys to how we negotiate with our fridges. If 2020s tech had a PR slogan, it would be "Don't worry, your agent will handle it."

But is that what we want? Or just what we've been sold? Or is it the tech equivalent of a midlife crisis?

Thomas Friedman tells a story about Intel engineers calculating what would happen if a 1971 Volkswagen Beetle improved at the rate of microchips: it would cost four cents, go 300,000 mph, and get two million miles per gallon. Apply that logic to commerce and we get Shopify apps that deploy AI agents in 90 seconds, and Walmart fielding a fleet of digital assistants named things like Sparky. But here's the rub: our infrastructure scaled. Our institutions didn't. Our behaviour didn't. Our brains definitely didn't.

The agentic revolution promises predictive personalisation and frictionless decisioning. But what if the friction was the point? What if the in-between moments, the pauses, the dithering, the comparisons, were the very things that made commerce feel human?

Thinking fast, shopping dumb

Kahneman gave us the framework: System 1 is fast, instinctive, emotional; System 2 is slow, deliberate, analytical. Now ask yourself which one your shopping agent appeals to. Agentic AI thrives on System 1. It guesses. It nudges. It reinforces your worst habits: dopamine over deliberation, comfort over curiosity. System 2, the part that says "do I really need another pair of black joggers?", rarely gets a look in.

In agentic commerce, we're not just outsourcing effort. We're outsourcing thought. That's not efficiency. That's atrophy.

Recent research into ecommerce AI agents found that most of them mirror human biases: they favour top-ranked products, respond to social proof, and struggle with novelty. In other words, your agent is as predictable and gullible as you.

We don't need better agents. We need better questions. Agentic systems predict what we'll do next based on what we did last. But the whole thrill of shopping is the thing we didn't see coming.

The mirage of utopia

Kevin Kelly warned that a world without discomfort is utopia, and also stagnant: no problems to solve, and therefore no opportunities either. Agentic AI promises frictionless experiences and always-on assistants. It promises utopia. But utopia is sterile. Predictable. Boring. The best ecommerce experiences involve the unplanned: the stumble-upon, the weird gift from the "you might also like" section that actually hits. Agents aren't wired for that. They're wired for optimisation, and optimisation, unchecked, breeds blandness.

Cher didn't need a perfect closet. She needed options that made her feel something. That made her think. That let her choose.

As if I'd trust my agent

Let's be honest: this is happening. Agents will be embedded in ecommerce, logistics, finance, everywhere. But must we accept their terms blindly? Can we not demand systems that pause, that explain, that reflect our intent rather than just our history? The real innovation isn't agentic autonomy. It's agentic empathy.

In the final scene of Clueless, Cher doesn't pick the dress by algorithmic sorting. She picks it because it feels right. Because it means something. Because, in her own weird way, she thought it through. That's not regression. That's being human.

So yes, the agents will rise. The best among them won't just automate our instincts; they'll honour our indecisions, prompt better questions, and pause when we need space. Because if the future of commerce is just faster checkout, frictionless choice, and auto-piloted desire?

Then, to quote the patron saint of Beverly Hills: "Ugh, as if."