Who Does Your Agent Actually Work For?

Why index funds won, Alexa failed, and what that tells you about who wins agentic commerce.

At an industry event a few weeks ago, I had yet another conversation about agentic commerce – AI agents doing your shopping for you, autonomously, end to end. It’s the hot topic in (e-)commerce right now. And the argument I keep hearing (and I bet you’ve heard it too if you are even remotely connected to e-commerce) goes like this: it removes purchasing friction; friction removal always wins; therefore, it’s inevitable. Case closed – let’s talk valuations and who the big winners will be.

Except we ran this exact play once before. Remember when Alexa was going to become your shopping platform? Amazon put a microphone in a hundred million kitchens, wired it directly into the world’s largest store, and waited for the orders to roll in. Alas, they didn’t. Voice commerce stayed a rounding error, the division reportedly burned billions, and today Alexa mostly sets kitchen timers, switches your lights on, and tells you the weather.

So the friction argument has at least one big, expensive counterexample. Which raises the question my conversation partner and I debated for a while: has it ever actually happened? Has there been a moment in history where people willingly handed the decision itself – not the execution, the decision – to a machine or a system, at scale, because it was more convenient?

Turns out: yes. Constantly. We just don’t notice, because successful delegation is invisible by design.

Start with the biggest one. Index funds are the largest voluntary transfer of decision-making authority in modern economic life. Jack Bogle launched the first index fund at Vanguard in 1976, and Wall Street literally called it “Bogle’s Folly” – who would accept average returns on purpose? Roughly five decades later, passive funds hold more US equity assets than active ones (they crossed over in 2023, per Morningstar), and trillions of dollars sit in vehicles where a formula – not a person, not even the investor – picks every single holding. Target-date funds went further: a formula picks your stocks and adjusts your risk as you age, and you touch nothing for forty years. And auto-enrollment in 401(k)s – supercharged by the Pension Protection Act of 2006 – went furthest of all: people delegated the decision to even have a decision.

Or take navigation. Twenty years ago, knowing the best route across town was a point of pride – cab drivers built careers on it, and every family road trip featured a co-pilot wrestling a paper map. Today Google Maps decides, a few hundred million times a day, and nobody blinks. Uber stacked three more delegations on top: an algorithm picks your driver, your route, and your price, and we went from “never get into a stranger’s car” to tracking the stranger’s arrival on a map in about five years flat.

The list keeps going. Subscribe & Save and autopay quietly automated billions in recurring purchases. Spotify’s algorithm decides what you hear next; Netflix autoplays the next episode before you’ve formed an opinion about the last one. People delegate consequential decisions to systems all the time – happily, at massive scale.

Which makes the failures more interesting, not less. Because there’s a genuinely instructive one: Stitch Fix. The company IPO’d in 2017 on precisely the agentic-commerce thesis – tell us your size and budget, and an algorithm (plus a stylist) picks your clothes. It worked well enough to go public and then shed more than 90% of its value. Delegating clothing turned out to be fundamentally different from delegating index weightings, because what you wear is a statement about who you are. Few people’s identity is wrapped up in the precise set of holdings in their stock portfolio.

Lay the winners next to the losers and a pattern emerges. People don’t resist giving up control. People resist giving up self-expression, and they resist agents whose incentives they can’t read. Delegation wins when three things line up: the decision is a chore rather than a statement about who you are, the outcome is verifiable or the stakes are low, and it’s legible whom the agent actually works for. Vanguard won not just because indexing was mathematically sound but because its structure made the incentive question boring: the funds are owned by the fundholders. There’s no one to betray you. Alexa failed the same test from the other side – everyone could smell that the agent worked for the seller. You didn’t need an economics degree to feel the conflict of interest; you just needed to hesitate for half a second before saying “Alexa, buy batteries” and wonder whose batteries, at what price.

Hence, agentic commerce isn’t a single bet – it’s two very different ones rolled into one. The replenishment layer – detergent, dog food, printer ink, the maybe 60% of your cart that’s pure habit – is a chore with verifiable outcomes, and it likely will get eaten, probably faster than most people expect. The discovery layer – the jacket, the restaurant, the gift for your partner – is not friction to be removed. Choosing is part of the consuming. Strip it out and you haven’t made shopping efficient; you’ve made it not-shopping. The friction argument is half-right, and the half it misses is the expensive half: sometimes what looks like friction is the product.

If you’re building in this space, or advising someone who is (or, like most of the executives Jeffrey and I work with, trying to figure out what your customers will still want a human hand in), here’s my best bet: take your customer’s purchases and sort them into two piles – chores and statements. Be honest about which pile each item actually belongs in, not which pile your strategic plan needs it to be in. Then ask, for every chore you want to automate: can my customer verify the outcome, and can they tell – instantly, structurally, without reading a terms of service – that the agent works for them and not for the merchant? That second question is where this market will be won, because the technology is getting there quickly and the trust architecture mostly isn’t.

Which brings me back to where the conversation should have started. The interesting question about agentic commerce was never “will people give up control?” – history says they will, cheerfully, for the right chores. The question is the same one Bogle answered in 1976: who builds the agent whose incentives are so structurally boring that trusting it requires no leap of faith at all? Amazon likely won’t be it – it owns the store. So who becomes the Vanguard of agentic commerce?

@Pascal