Agentic Commerce: Own the Customer or Become Invisible
Agentic shopping will hit a fifth of the channel in five years, not twenty-five — and the winner won't be whoever has the best AI, but whoever owns the customer when the agent buys.
Executive summary
The most important thing to understand about agentic commerce is not that it is coming, but how little time there is to get ready for it. Electronic commerce took roughly twenty-five years to grow from a rounding error into about a sixth of American retail. The credible forecasts for agentic commerce — purchases that an AI agent researches, decides on, or completes on a person's behalf — put it at a comparable share of online spending inside five. Bain projects US agentic commerce at $300–500 billion by 2030, fifteen to twenty-five percent of e-commerce; McKinsey models up to $1 trillion of orchestrated US retail revenue by then, and $3–5 trillion globally. Take the low end of those ranges and the pace still runs several times faster than the web did. The comfortable assumption that adoption curves give incumbents a decade to adjust does not survive contact with these numbers.
The second conclusion matters more than the first, because it decides who wins. The scarce resource in this shift is not model quality — the frontier labs will all have capable shopping agents — it is control of the customer at the moment of purchase. The early evidence already points one way. Shoppers say they trust a retailer's own agent about three times more than a third-party agent to actually complete a transaction. Walmart's in-house agent lifts the baskets of the people who use it by around a third. OpenAI, having launched an in-chat checkout, spent the following six months quietly retreating from it toward merchant-run storefronts. And in August 2026 a US appeals court told Amazon it could not simply wall its site off from an agent a customer had sent. The pattern underneath these facts is that the platforms cannot disintermediate everyone by force, and the retailers who keep the relationship — their data, their loyalty, their own agent — hold the leverage.
The third conclusion is the practical one. Because the direction of travel is now legible even though the winning brands are not, a retailer can act without betting on a protocol or a platform. The moves that pay off in every plausible future are the same: make your catalogue and inventory readable, in real time, to machines; expose clean checkout and availability so an agent can transact without scraping; deploy or plan a first-party agent so you are a destination and not merely a supplier; and keep the customer data that makes any of it defensible. These are no-regret moves — cheap relative to the downside, useful whoever wins. The bets you should defer are the platform-specific ones. The horizon here is genuinely short and the specifics are genuinely unknowable; what follows is a set of source-grounded projections, not forecasts dressed up as certainties. But the shape of the thing is clear enough to move on now.
From answering to buying
For three years the visible face of the AI revolution was a chat box that answered questions. Agentic commerce is what happens when that same system stops answering and starts acting — and the first place it touches real money, at scale, is shopping. That makes retail an unusually good vantage point on the wider transition. Everything hard about giving an autonomous system authority over consequential decisions shows up here in miniature and with a price tag: how a machine proves it is allowed to spend your money, who is accountable when it buys the wrong thing, whether the businesses on the other side cooperate or resist, and what a person is comfortable handing over. The dynamics that will govern agents booking travel, managing money, or running a supply chain are being litigated first, cheaply, over trainers and grocery baskets.
It is worth being precise about what "agent" means here, because the loose usage hides the whole story. At one end sits an AI that influences a purchase — a recommendation, a comparison, a summarised review — while a human still clicks buy. In the middle sits an assistant that assembles the cart and waits for a nod. At the far end sits software that, given a standing instruction, completes the transaction itself. All three get called agentic commerce, and the distance between them is the difference between a marketing feature and a structural change in who does the shopping. Keeping the three separate is the only way to read the numbers without being misled by them.
Everyone means something different by agentic commerce
Start with the forecasts, because their disagreement is the first clue to what is actually happening. eMarketer models the market at around $144 billion by 2029. Morgan Stanley puts agent-driven purchases at ten to twenty percent of US e-commerce by 2030 — roughly $190–385 billion. Bain lands at $300–500 billion. McKinsey models up to $1 trillion of orchestrated US retail revenue, and $3–5 trillion worldwide. That is more than a thirty-fold spread between the low and high estimates for the same phenomenon in the same market on roughly the same date.
The spread is not analysts disagreeing about whether this is real. It is analysts counting different things. The low numbers count only the narrow case — a purchase an agent completed autonomously. The high numbers count the broad case — every sale an AI touched along the way. You can watch the same gap open in the holiday data. Salesforce estimated that AI and agents influenced around $262 billion of global online sales over the 2025 holidays, roughly a fifth of the total. Bain, counting more strictly, put agent-completed US Black Friday sales at about $3 billion. Neither is wrong. They are measuring opposite ends of the same spectrum, and the spectrum is the point: the influence layer is already enormous, and the autonomous layer is small but compounding fast. When someone tells you agentic commerce is either overhyped or already dominant, they have usually picked the end of the spectrum that suits their argument.
The influence layer is not waiting for the autonomous one to catch up. Adobe measured traffic referrals from generative-AI tools to retail sites rising by around seven hundred and seventy percent year on year last November, and found that visitors arriving from AI channels converted roughly nine times more often than those coming from social media, with revenue per visit up more than two hundred percent. Whatever you call it, shoppers are already starting their journeys inside an AI and arriving at retailers pre-qualified. Gartner, looking further out, expects about a fifth of digital-commerce transactions to run through AI platforms by 2030 and, more provocatively, expects AI agents to outnumber human sellers roughly ten to one by 2028. The reason the definition matters so much is that it changes what you are supposed to do about it. If agentic commerce means the narrow, autonomous case, it is a small experiment you can watch from a distance. If it means the broad case — AI shaping which products a shopper ever sees — then it is already the top of your funnel, and being unreadable to it is already costing you customers you will never know you lost.
It arrives in five years, not twenty-five
The useful analogy is the one every retailer already lived through. E-commerce felt slow for a long time. It was under one percent of US retail at the turn of the century, still only around four percent by 2010, and it took the pandemic to shove it past sixteen percent. A quarter-century to reach the point where ignoring it became fatal. Plenty of executives who watched that curve concluded that channel shifts move at a pace you can manage — that you get years of warning. That lesson is about to be expensively wrong.
Run the arithmetic the forecasts imply. E-commerce needed about twenty-five years to become a sixth of retail. Agentic commerce is projected to become fifteen to twenty-five percent of e-commerce — a comparable structural share of the online channel — inside five. That is not twice the speed of the web's adoption; it is closer to four or five times. Even the more cautious view, that agentic will merely double the pace of e-commerce, still collapses a decade of warning into a couple of years. The mechanism behind the compression is straightforward: e-commerce had to build its demand-side infrastructure from nothing — broadband, payment trust, logistics, the habit of buying things you cannot touch. Agentic commerce inherits all of it. The rails are laid, the trust in online payment exists, the fulfilment networks run. What is being added is a thin layer of autonomy on top of a mature stack, and thin layers propagate fast.
There is a real caveat, and it cuts against the hype in a specific way. Adoption will not be uniform across what people buy. Spec-driven purchases — batteries, printer ink, the same detergent as last month — are nearly frictionless to hand to an agent, and they will move first. Considered purchases — clothes you want to see, a holiday, anything with taste or risk in it — will lag until trust catches up, and most shoppers today still say they are not comfortable letting an AI run an end-to-end transaction. So adoption really splits into two curves: a fast one for the boring, repeatable basket, and a slow one for the emotional cart. But the boring basket is a very large fraction of retail, and it is exactly the fraction an incumbent cannot afford to lose quietly.
Three postures, one prize
Underneath the noise of a dozen product launches, every serious player has taken one of three positions. The disagreements between them are not really technical. They are disagreements about who should end up owning the customer.
The front-door strategy — OpenAI, Google, Perplexity — assumes the shopping starts inside an assistant, and whoever owns that assistant owns discovery and, eventually, the checkout. The walled-garden strategy is Amazon's: build the most polished in-house experience, refuse the shared protocols, and use an agent to pull the rest of the internet's inventory into your own flow rather than let anyone pull yours out. The third strategy — arm the retailer — is the one that gets underrated, and it is where Walmart, Shopify's merchants and the card networks have planted their flags: give the retailer its own agent and an open cart it controls, so it meets the customer wherever they are without surrendering the relationship. The economics inside the front-door camp are already diverging in a way worth watching, because it tells you how contested this is. OpenAI's commerce protocol charges merchants a fee of around four percent on the transactions it routes; Perplexity, whose Comet agent became the most active agentic checkout of 2026, charges merchants nothing at all, treating commerce as a way to make its assistant sticky rather than as a revenue line. When one entrant is willing to run the pipe for free to win the relationship, the others cannot count on tolls. That is the sound of a land grab, not a settled market.
Amazon's posture deserves its own reading, because it is the most coherent and the most exposed. Rufus — now folded into what Amazon calls Alexa for Shopping — has grown its monthly users well over a hundred percent year on year, can buy items automatically when a price target is hit, and increasingly carries sponsored placements. Its "Buy for Me" feature reaches outside Amazon to purchase items from other retailers and pull them into Amazon's own flow, which is the walled garden expanding rather than opening. Amazon's chief executive has been blunt about why he thinks third-party agents will struggle: they cannot, he argues, get the pricing or the product information right, and they lack the decades of personalisation data Amazon holds. He is not wrong about the advantages. The open question is whether being right about your own moat matters when a court has just said you cannot use it to keep out agents your customers have chosen to send. These are not friendly variations on a theme. They are three different answers to the only question that matters, and the market is going to settle somewhere in the tension between them.
The fight is the front door, and trust favours the retailer
Here is the finding that should reorder a retailer's priorities. When we covered the plumbing of all this in our Agent Internet episode, number 19, earlier this year, the single most consequential data point was about trust: as we reported then, Bain found that shoppers trust a retailer's own agent roughly three times more than a third-party agent to actually complete a purchase. That number is the whole ballgame, because it means the platforms' dream of disintermediating everyone runs straight into the customer's own preference. People do not want a neutral super-assistant spending their money at an anonymous supplier. They want the brand they already trust to do it.
Walmart is the live proof. Its agent, Sparky, is now used by around half of the retailer's app users, and those users spend about thirty-five percent more, with baskets roughly a third larger, than those who do not. Sparky-attributed sales grew about a hundred and fifty percent quarter on quarter and the units it moved rose fourfold year on year. Crucially, Walmart did not build Sparky to sit inside someone else's assistant and hope for scraps. It kept the basket, the customer data and the loyalty relationship inside Walmart, and then — in March 2026 — pushed Sparky outward into ChatGPT and Gemini so it could meet customers on those surfaces too, on Walmart's terms. As one of its executives put it, shopping can now start anywhere, whether that is Walmart or a question inside a chatbot; the discipline is to make sure it still ends inside Walmart's relationship with the customer.
The most telling move came from the other direction. OpenAI launched Instant Checkout inside ChatGPT in late 2025, the purest expression of the front-door strategy: buy without ever leaving the assistant. By March 2026 it had stepped back from it, redirecting toward merchant-specific apps inside ChatGPT — Target, Instacart, DoorDash, Walmart — where the retailer keeps control of inventory, checkout, tax and data. The stated reason was complexity, but the deeper reason is the trust gap made concrete: merchants would not hand over the transaction, and the numbers showed customers preferred it that way. The company with the most popular assistant on earth discovered it could not simply become everyone's checkout. That retreat is the clearest signal in the whole landscape of where the equilibrium is settling.
None of this is unique to a retailer the size of Walmart. Sephora runs the same play at a different scale — using AI inside the big platforms for reach while tying every interaction back to its own app and loyalty programme, so the relationship never leaves home. And the holiday data suggests the advantage is not just defensive: retailers that deployed their own agents over the 2025 peak grew around fifty-nine percent faster than those that did not. The lesson generalises into a simple rule. Meeting the customer inside someone else's assistant is fine, even necessary, as long as the basket, the data and the loyalty stay yours. The mistake is not showing up on the platforms; it is showing up on them as an anonymous supplier with nothing of your own to come back to.
The plumbing is converging, and the law just tilted
If the front-end brands are fighting, the plumbing beneath them is quietly agreeing. Ask which protocol wins and you are asking the wrong question, because they are turning out to be layers of one stack rather than rivals. A full agentic purchase reads a merchant's catalogue through the Model Context Protocol, builds and confirms the cart through a commerce protocol, and proves the shopper actually authorised the spend through a payments protocol. Those are complementary jobs, not competing ones.
The convergence is visible in who has signed on. Google's Universal Commerce Protocol, announced at the retail industry's big January 2026 gathering, was co-developed with Shopify, Etsy, Wayfair, Target and Walmart, and endorsed by more than twenty partners including Visa, Mastercard, American Express, Adyen, Stripe, Best Buy and Macy's. Google's Agent Payments Protocol, built on open web credential standards, launched with sixty-plus partners. Even OpenAI's Stripe-built commerce protocol and the card networks' own schemes plug into the same payment-mandate model. Rivals who are fighting bitterly over the customer are simultaneously agreeing on the pipes — for the same reason the tech giants gave away the underlying agent standards in the first place: the alternative, a fragmented mess of incompatible checkouts, is worse for everyone. Fragmentation risk has not vanished — some firms are still launching proprietary protocols — but the gravity is unmistakably toward a shared, open stack with a tokenised card layer underneath.
Then there is the law, which just moved in a way that matters more than any product launch. Amazon, defending its walled garden, sued Perplexity to stop its Comet agent from shopping on Amazon on customers' behalf. It had grounds to be annoyed: Amazon says it warned Perplexity at least five times from late 2024, put a technical block in place in the summer of 2025, and watched Perplexity ship an update to route around it within a day. In March 2026 it won a preliminary injunction on the argument that the agent's access violated the Computer Fraud and Abuse Act, the federal anti-hacking statute. In August 2026 the Ninth Circuit reversed it. The court's reasoning is the part to remember: when a user instructs an agent to act for them on a website, it is the user who is accessing the site, not the software vendor. That is a foundational ruling for this whole future. It means a platform cannot simply declare that agents a customer has authorised are trespassers. The walled garden can still be the best experience; it cannot be a legal fortress. The direction of travel — open plumbing, plus a legal presumption in favour of the agents customers choose to send — now points away from any single company owning the whole flow.
What to do now: become legible to agents
All of this resolves into a surprisingly concrete to-do list, and the reason it is concrete is that a specific set of moves pays off no matter which platform or protocol wins. The organising idea is legibility: an agent can only recommend, compare and buy what it can read and act on. A product that is invisible to machines is becoming invisible to a growing slice of demand.
The sequence is well understood by the people already doing it: structured data first, then API exposure, then pricing and fulfilment automation. In practice the first steps are almost embarrassingly cheap. Confirm your site is not accidentally blocking the crawlers that AI shopping tools use to read it. Audit your product data so every item exposes an accurate price, stock level and specification a machine can parse. Both are an afternoon's work and touch every product you sell. From there the ladder climbs to real-time feeds, a checkout an agent can complete without scraping, and eventually a first-party agent of your own. None of this requires knowing whether ChatGPT or Gemini or Amazon wins. It only requires being readable.
The economics of climbing that ladder are more forgiving than they look, because the payoff is front-loaded. The first rung — a clean, machine-readable feed and an unblocked crawler — costs almost nothing and immediately determines whether an agent can put you in a shopper's consideration set at all. That alone is the difference between being a candidate and being invisible. The higher rungs, a transactable checkout and a first-party agent, cost more, but they are also where the compounding value sits: an agent that can complete a purchase captures the boring, repeatable basket that would otherwise default to whoever is easiest for a machine to buy from, and an own-brand agent turns a one-off sale into the deepened, data-rich relationship the Walmart numbers describe. The right way to sequence it is to treat the cheap rungs as insurance you buy today and the expensive ones as investments you stage as the transaction volumes justify them — not as a single all-at-once re-platforming that waits until the picture is clear, by which point the cheap insurance you skipped has already cost you a year of demand.
The urgency is that most retailers are not there yet. Research in early 2026 found around forty percent of e-commerce businesses still in the middle of standardising their product pages for agentic AI, and roughly a third had not started at all. That gap is the opportunity and the threat in one number. In a channel projected to compress twenty-five years of adoption into five, being a year late to legibility is not a minor lag — it is a year of an accelerating share of demand routed to competitors an agent could actually read. And there is a compounding kicker in the data: retailers that ran their own agents over the 2025 holidays grew markedly faster than those that did not. Legibility gets you into the consideration set; your own agent deepens the basket once you are there.
What not to do, and when to act
The mistakes here are mostly over-reactions, and they share a root: treating an unsettled market as if it had a winner. Do not re-platform your business around whichever assistant is ahead this quarter — the retreat of OpenAI's in-chat checkout inside six months is a warning about how fast "the obvious winner" can change its mind. Do not commit exclusively to one protocol when the protocols are converging into a shared stack; you would be paying to back a horse in a race that is turning into a relay. Do not, above all, trade the customer relationship for reach. The platforms will offer you distribution in exchange for your data and your direct line to the shopper, and the trust gap tells you why that is a bad trade: the relationship is the one asset that is genuinely scarce, and once handed over it is very hard to get back.
So when do you act, and why? You act on the no-regret list now — this quarter — precisely because those moves are cheap and their value does not depend on resolving any of the open questions. You defer the platform-specific bets until the front-door war has a clearer shape, which the convergence of the plumbing and the drift of the law suggest is a matter of a couple of years, not ten. The forcing function is not a date on a calendar; it is the simple fact that invisibility to agents is becoming invisibility to demand, and that share is growing every quarter. The retailers who will look prescient in 2030 are not the ones who guessed the winning assistant. They are the ones who made themselves legible early, kept their customers close, and let the market sort out the rest around a business that agents could already see, read and buy from.
Bottom line
The honest epistemic position is that the direction is knowable and the details are not, and the report has tried to stay on the right side of that line. Over the next two years, expect agent-assisted buying to become ordinary in the boring, repeatable basket, while the emotional cart lags and most autonomous checkout stays supervised. Over five, expect agentic commerce to reach the fifteen-to-twenty-five percent of e-commerce the forecasts describe, with the control war between platforms and retailers largely — though never entirely — settled into a blended market where nobody owns the whole flow. Over ten, expect agents to be a primary channel of demand, and the gap between the retailers who became legible early and the ones who waited to have widened into the difference between owning a customer and being a commodity supplier to someone else's assistant. These are projections grounded in real forecasts, real adoption data and a real legal trend, not certainties; the knowability horizon in a shift moving this fast is short, and it would be dishonest to pretend otherwise. But the move a retailer should make does not depend on getting the details right. Become readable to machines, keep your customers close, and act now on the things that are cheap and certain while deferring the things that are expensive and unknowable. That is the rare strategy that is right in every version of the future that the evidence still leaves open.
Sources
- 2030 Forecast: How Agentic AI Will Reshape US Retail (Bain & Company) — the $300–500B / 15–25%-of-e-commerce forecast and the consumer-comfort gap.
- Agentic AI Commerce: The Next Retail Revolution (Bain & Company) — the 3× trust gap for retailer-owned agents, and disintermediation vs re-intermediation.
- McKinsey forecasts up to $5 trillion in agentic commerce sales by 2030 (Digital Commerce 360) — the $1T US / $3–5T global model.
- US agentic commerce revenue forecast to reach $1 trillion by 2030 (Retail Dive) — McKinsey's moderate case, agents at ~18% of US B2C spend.
- OpenAI shifts checkout plans in its agentic commerce strategy (Digital Commerce 360) — the retreat from in-chat Instant Checkout toward merchant apps.
- Stripe powers Instant Checkout in ChatGPT and releases the Agentic Commerce Protocol — ACP, the OpenAI/Stripe checkout standard.
- Under the Hood: Universal Commerce Protocol (UCP) (Google Developers Blog) — the open cart protocol and its coalition.
- Announcing the Agent Payments Protocol (AP2) (Google Cloud) — the payment-mandate model and 60+ launch partners.
- The Supply Side: Walmart's Sparky shopping agent drives more sales (Talk Business & Politics) — Sparky's basket, sales and units figures, and the ChatGPT integration.
- Amazon says Rufus gives it an edge in the agentic commerce race (PYMNTS) — Rufus growth, auto-buy, and Amazon's stance on third-party agents.
- Judge blocks Perplexity's AI bot from shopping on Amazon (GeekWire) — the March 2026 injunction and the CFAA argument.
- Perplexity overturns Amazon's injunction on its AI shopping bot (Engadget) — the August 2026 Ninth Circuit reversal and its "the user accessed the site" reasoning.
- AI and agents present a $263B holiday opportunity for retailers (Salesforce) — the ~$262B AI-influenced holiday sales figure.
- AI-driven traffic surges across industries, retail sees biggest gains (Adobe) — the AI-referral traffic and conversion data.
- Agentic-Ready Product Data: How to Get It (Shopify) — the structured-data-first readiness sequence and the readiness gap.
Transcript
Alex: In five years — not twenty-five — a fifth of everything you buy online could be chosen by a piece of software you never actually asked to shop for you.
Sam: And the company that wins that isn't the one with the smartest AI. It's whoever's still holding your hand when it hits "buy."
Alex: Welcome back to Dan's AI Intel.
Sam: This is the show that takes the one AI story each week that actually matters, and pushes past the noise to what's really going on — and what it means, for the economy, for the politics of it, for the race between the handful of labs actually building this stuff.
Alex: I'm Alex.
Sam: I'm Sam. And this week — a federal appeals court just told Amazon it couldn't block an AI shopping assistant from shopping on its site on a customer's own behalf.
Alex: Which sounds like a small legal skirmish. It is not. It's the tell for a much bigger fight — over who actually gets to own you as a customer once a piece of software is doing the shopping for you.
Sam: We're going to walk through who's building this — the chatbots racing to become your front door, Amazon building the walls higher, retailers arming themselves right back — the trust numbers that genuinely surprised us, and where the money and the law are both pointing right now.
Alex: And there's a timeline in here we checked twice, because neither of us believed it the first time.
Sam: If you're new here, hit follow wherever you're listening — it's free, and it means you don't miss the next one.
Alex: So for about three years, the AI revolution has mostly looked like a chat box that answers questions. What we're talking about today is what happens the moment that same box stops answering and starts acting — and the first place that's happening, at real scale, with real money attached, is shopping.
Sam: Which honestly makes retail a weirdly good place to watch the whole AI story play out in miniature.
Alex: It really is. Because everything that's hard about letting an autonomous system loose on a genuinely consequential decision shows up right here, cheaply, before it shows up anywhere bigger. How does a machine prove it's allowed to spend your money? Who's actually on the hook when it buys the wrong thing? Do the businesses on the other end cooperate with it, or fight it? All of that's playing out right now, over sneakers and grocery baskets, before it plays out over a mortgage or a flight booking.
Sam: Cheap rehearsal for an expensive future.
Alex: Exactly.
Sam: But here's my first problem with the term "agentic commerce" — everyone uses it constantly, and I don't actually think they all mean the same thing by it.
Alex: They really don't, and it's worth being precise, because the loose usage hides the whole story. Picture three completely different products that all get called this. One — an AI that just influences a purchase: it recommends a pair of running shoes, compares two blenders, summarises the reviews — and a human still clicks buy. Two — an assistant that goes further: it actually assembles your grocery order for the week and waits for you to glance at it and nod. Three — the far end: on a standing instruction you gave it once, software just reorders your detergent the moment it notices you're running low, no nod required, ever.
Sam: And all three of those get called "agentic commerce."
Alex: All three. And the distance between "it helped me decide" and "it bought it without me" isn't a technical nuance — it's the entire distance between a marketing feature and a genuine structural shift in who does the shopping. Keep those three tiers separate in your head, and pretty much everything else in this story gets a lot clearer.
Sam: It also raises the question that's been nagging at me since I started reading into this — the detergent example. If the thing genuinely reorders on its own, no nod from me, and it gets the size wrong, or reorders twice by mistake, whose problem is that? Mine, because I set it up? The store's, because it fulfilled a bad order? Or whoever built the agent in the first place?
Alex: That's exactly the right question to be sitting with, and it's not a hypothetical one — it's already showing up in court, which is where this episode is eventually headed. Hold onto it.
Sam: Okay, so let's start with the number that actually made me want to do this episode. I went looking for "how big is agentic commerce going to be," and the estimates I found were not remotely in the same universe. eMarketer says about a hundred and forty-four billion dollars by 2029 — the narrowest figure out there. Morgan Stanley says ten to twenty percent of US e-commerce by 2030, call it one-ninety to three-eighty-five billion. Bain says three hundred to five hundred billion. And McKinsey — the broadest of the lot — says up to a trillion dollars of US retail revenue running through this by 2030, with agents making up around eighteen percent of all US consumer spending, and three to five trillion dollars globally once you count everything.
Alex: That's more than a thirty-times spread between the lowest estimate and the highest one. For the same phenomenon. In the same market. On roughly the same date.
Sam: Which is the kind of spread that would normally tell you one side is just wrong.
Alex: Except nobody here is actually wrong — they're counting completely different things. The low numbers only count the narrow case: a purchase an agent completed entirely on its own, tier three from a minute ago. The high numbers count the broad case: every single sale AI touched anywhere along the way, tiers one through three, all lumped together.
Sam: So it's less "who's right" and more —
Alex: More like asking how much of dinner a robot cooked tonight. If you only count the arm that flipped the burger and put it on the plate, that's a small number. If you count the smart fridge that noticed the mince was about to go off and suggested burgers in the first place, that's basically the whole meal, start to finish.
Sam: Okay, that actually clicks — same kitchen, wildly different answer, depending entirely on where you decide to draw the line around "the robot did it."
Alex: And you can watch that exact gap open up in real, already-happened data, not just forecasts. Over the 2025 holidays, Salesforce estimated AI and agents influenced around two hundred and sixty-two billion dollars of global online sales — roughly a fifth of the total. Bain, counting far more strictly — agent-completed sales only, no influence counted — put US Black Friday at about three billion.
Sam: Two hundred and sixty-two billion, versus three billion. Same holiday season.
Alex: Same exact holiday season. Neither number is wrong — they're measuring opposite ends of the same spectrum. The influence layer is already enormous. The fully autonomous layer is still small. But it's compounding fast.
Sam: And the influence layer isn't exactly sitting still while the autonomous layer tries to catch up, either.
Alex: It's really not. Adobe measured traffic from generative-AI tools into retail sites rising about seven hundred and seventy percent year on year last November — and the visitors who arrived that way converted roughly nine times more often than people coming in from social media, with revenue per visit up over two hundred percent.
Sam: Nine times the conversion rate. That's not a rounding error — that's a completely different kind of shopper showing up at your door.
Alex: Pre-qualified, basically, before they even land on the site. And Gartner's looking further out than any of that — they expect about a fifth of digital-commerce transactions to run through AI platforms by 2030, and, the genuinely provocative one, AI agents to outnumber actual human sellers roughly ten to one by 2028.
Sam: So why does this definition fight actually matter, beyond just being annoying to a podcast host trying to find one clean number?
Alex: Because it changes what you're supposed to do about it. If "agentic commerce" just means the narrow, fully-autonomous case, it's a small experiment you can comfortably watch from a distance for a few years. If it means the broad case — AI quietly shaping which products a shopper even gets shown — it's already sitting at the top of your sales funnel, today. And being unreadable to it is already costing you customers you'll never even know you lost.
Sam: So if I'm keeping score — narrowest to broadest, it's eMarketer, then Morgan Stanley, then Bain, then McKinsey sitting way out at the top with that near-trillion-dollar US figure.
Alex: That's the order. And none of the four firms are disagreeing about direction — every single one of them thinks this gets big. They're just drawing the burger-and-fridge line in a different place.
Sam: Right, which is a reasonable place to leave the "how big" question for now. What I actually want to know next is the "how fast" question — because "big" and "big by when" are very different problems for a business to plan around. Right, so this next part is the number I said we checked twice. Everyone kind of assumes channel shifts move at a pace you can plan around — like, e-commerce didn't kill traditional retail overnight, it took forever to matter.
Alex: It took about twenty-five years, actually, being precise about it. E-commerce was under one percent of US retail at the turn of the millennium. Still only around four percent by 2010. It genuinely took a pandemic to shove it past sixteen percent.
Sam: A quarter of a century to go from a rounding error to "ignoring this channel is now fatal for your business."
Alex: Right — and a lot of very smart retail executives watched that slow, generous curve, and reasonably concluded that channel shifts always hand you a decade or more of warning to adjust. That lesson is about to be expensively, expensively wrong.
Sam: Because agentic commerce is supposed to hit that same kind of structural share — fifteen to twenty-five percent, remember, from Bain and McKinsey — in about five years, not twenty-five.
Alex: Run the actual arithmetic on that. E-commerce needed roughly twenty-five years to become a sixth of total retail. Agentic commerce is projected to become fifteen to twenty-five percent of e-commerce — a comparable structural share of the online channel — inside five years. That's not double the speed of the web's own adoption curve. That's closer to four or five times as fast.
Sam: Which is the bit I genuinely didn't believe the first time — four or five times, not two.
Alex: And even the more conservative reading of the same numbers — that agentic merely doubles the pace of e-commerce, nothing more dramatic than that — still crushes a decade of warning down into a couple of years, tops.
Sam: So what's actually different this time around? Why so much faster?
Alex: Think of it like moving into a fully plumbed house, versus building one from nothing. E-commerce had to lay every single pipe itself, from scratch — broadband, the basic trust that your card details were safe typed into a website, an entire logistics and delivery network, the plain habit of buying something you can't touch first. Agentic commerce inherits every bit of that, already built, already working. It's not laying new pipes at all. It's adding a smart thermostat to a house that's already fully wired.
Sam: A thin layer of autonomy, dropped on top of an already-mature stack.
Alex: Exactly — and thin layers propagate fast, because there's so little friction left underneath them. There is a real caveat here, though, and it actually cuts against the hype in an interesting way. Adoption is not going to be even across everything you buy. The stuff you already buy on autopilot — the same razor blades, the dog food, the printer ink, whatever detergent you bought last time — hands to an agent almost frictionlessly, and it moves first. Considered purchases — clothes you genuinely want to see and feel, a holiday, a sofa, anything with real taste or real risk baked into it — lag well behind, because most people still say, today, they're not comfortable letting an AI run that kind of transaction start to finish on its own.
Sam: So it's really two separate curves, not one. A fast one for the boring, repeatable basket. A slow one for the emotional cart.
Alex: Right — but the boring basket is still a genuinely enormous slice of total retail spending. And it's exactly the slice an incumbent can't afford to quietly lose while it's busy worrying about the emotional cart.
Sam: Put the two curves side by side for a second, though — because I think that's actually the useful way to hold this, not the single blended "fifteen to twenty-five percent" headline number.
Alex: Fair. On the fast curve — the repeatable basket — you should genuinely expect agentic behaviour to feel completely normal within a couple of years, not five. On the slow curve — the considered, emotional purchase — five years probably still isn't enough for it to feel normal at all; that one moves on trust, and trust is earned slowly, purchase by purchase, not legislated into existence. The blended number in the forecasts is really an average of a curve that's already arrived and a curve that hasn't really started.
Sam: Which means a retailer selling mostly considered goods — furniture, say, or fashion — has more breathing room than the headline number implies.
Alex: Some. Not infinite. Because even a slow curve is still a curve, and the retailers who get legible early will still be first in line the moment that trust does arrive.
Sam: Okay, so if this is moving that fast, who's actually positioned to win it? Because it feels like every company on earth has launched some kind of shopping-AI thing this year.
Alex: Underneath all that noise, though, everyone serious has really only picked one of three strategies. And the disagreement between them isn't technical at all, at its core — it's a disagreement about who ends up owning the customer relationship when the dust settles.
Sam: Lay all three out for me.
Alex: Strategy one — own the front door. That's OpenAI, with ChatGPT and its own Agentic Commerce Protocol; Google, with Gemini and its Search AI Mode — pushing shopping straight into the search results page itself, not just a separate app — plus its own protocol coalition; and Perplexity, with its Comet agent, which has actually been the single most active agentic checkout of 2026 so far, by volume. The bet, in each case, is the same: the shopping starts inside my assistant, so I own the discovery, and eventually I own the checkout too.
Sam: Strategy two?
Alex: Wall the garden. That's Amazon, pretty much alone in this camp — Rufus, now folded into what they're calling Alexa for Shopping. No industry protocols at all, keep everything in-house, and a feature called "Buy for Me" that actually reaches outward, past Amazon's own walls, to buy an item from a different retailer entirely and pull the whole transaction back into Amazon's own flow.
Sam: So the garden isn't just walled — it's reaching over the fence to grab stuff off someone else's shelf.
Alex: Which tells you something about how confident they genuinely are in it. Strategy three — and this is the one I think gets seriously underrated — arm the retailer. Give the retailer its own agent and an open cart it fully controls, so it can meet the customer wherever they happen to be, without ever handing over the relationship. That's Walmart with Sparky, it's the coalition forming around Shopify, Target, Etsy and Wayfair building a shared open protocol together, and it's Visa and Mastercard laying the neutral payment rails underneath all three camps at once.
Sam: And here's the bit that actually tells you how contested this still is — the economics inside camp one, the front-door camp, are already splitting apart from each other. OpenAI's commerce protocol charges merchants a fee, something like four percent, on every transaction it routes to them. Perplexity's Comet — the one actually moving the most volume right now — charges merchants nothing at all.
Alex: Zero. Because Perplexity isn't trying to build a toll road, it's trying to make its assistant sticky enough that you never leave it. And when one player inside your own camp is willing to run the pipe for free just to win the relationship, nobody else in that camp can count on their tolls holding up for very long.
Sam: That's not a stable, settled market. That's a land grab, dressed up in protocol names.
Alex: It's a land grab. And Amazon's posture deserves a closer look on its own, because it's the most internally coherent of the three — and, I'd argue, the most exposed. Rufus's monthly users have grown well over a hundred percent year on year. It can already auto-buy an item the instant it hits a price target you've personally set, and it's increasingly carrying sponsored placements inside those recommendations too. And Amazon's own CEO has been blunt about why he thinks outside agents will simply struggle against all of that — he argues a third-party agent can't get the pricing or the product data right, and it doesn't have Amazon's literal decades of personalization history to draw on.
Sam: Which — honestly, he's probably not wrong about the advantage itself.
Alex: He's not wrong about the moat. The open question is whether being right about your own moat still matters, once a court's just told you that you can't use it to keep out agents your own customers deliberately chose to send you. Which, funnily enough, ties straight back to that "who's accountable when the agent buys the wrong thing" question from the top of the episode — and it's exactly where we're headed next.
Sam: Before we get there, though — quick gut check, because I want this on the record. If you'd asked me cold which of those three postures wins, I'd have guessed "own the front door," honestly. Biggest brands, biggest reach, biggest headlines.
Alex: That's the intuitive pick. It's also, based on what we found digging into the actual numbers next, probably wrong — or at least, nowhere near the whole story.
Sam: Okay, so talk me out of my own gut instinct, then. Why wouldn't the biggest, best-funded assistant just win this by default, the same way a search engine or a social feed eventually did?
Alex: Because owning attention and owning a transaction turn out to be two genuinely different jobs, requiring two genuinely different kinds of trust. Getting someone to ask you a question is cheap. Getting them to hand you their card details, or a standing instruction to spend on their behalf, is not — and that gap between the two is exactly where the next act lives.
Sam: A cliffhanger. Fine, I'll take it.
Alex: So here's the single most consequential number in this entire story, and it comes from Bain. Shoppers say they trust a retailer's own agent roughly three times more than a third-party agent, when it comes to actually going through with a purchase.
Sam: Three times. That's basically the whole ballgame, isn't it — because it means the platforms' dream of disintermediating everyone runs straight into what customers themselves actually want.
Alex: People don't want some neutral super-assistant spending their money at an anonymous supplier they've never personally dealt with. They want the brand they already trust to be the one doing it for them. Back in June, we did a whole episode on the plumbing underneath all of this — episode nineteen, "The Agent Internet" — and that exact trust number was the headline finding there too, about two months back now. It's the single thread running underneath this entire fight.
Sam: And Walmart's basically the live, working proof of it in action. Its own agent, Sparky, is now used by about half of the people on Walmart's app. And the people who actually use it spend around thirty-five percent more, with baskets roughly a third bigger, than the people who don't touch it.
Alex: Sparky-attributed sales grew about a hundred and fifty percent, quarter on quarter. Units sold through it are up four times year on year. And crucially — Walmart didn't build Sparky to sit quietly inside somebody else's assistant, hoping for scraps of attention. It kept the basket, the customer data, the whole loyalty relationship, inside Walmart. Then, in March 2026, it pushed Sparky outward — into ChatGPT, into Gemini — but strictly on Walmart's own terms.
Sam: Shopping can start anywhere it wants to. It just has to end up back inside Walmart's relationship with you.
Alex: That's almost word for word how one of their own executives put it.
Sam: Okay, so here's the part that actually surprised me most, though, out of everything — OpenAI. Because they launched exactly what you'd expect the front-door strategy to look like in its purest form — Instant Checkout, right inside ChatGPT, buy the thing without ever once leaving the app.
Alex: Late 2025. And by March 2026 — barely six months later — they'd quietly stepped back from it. Redirected instead toward merchant-specific apps living inside ChatGPT — Target, Instacart, DoorDash, Walmart — where the retailer itself keeps control of the inventory, the checkout, the tax, its own data, the whole thing.
Sam: They walked the flagship feature back that fast.
Alex: The stated reason was complexity. The real reason underneath it is the trust gap, made completely concrete — merchants simply wouldn't hand over the actual moment of transaction, and the early numbers showed customers themselves preferred it that way too. Think about it like handing someone your actual cash register versus letting them stand next to it and point at things. Merchants were fine with the second. Almost nobody was fine with the first. The company with the single most popular consumer AI assistant on the planet discovered it genuinely could not just become everyone's checkout by decree.
Sam: That retreat might honestly be the clearest signal in this whole landscape.
Alex: It's not unique to a company Walmart's size, either — Sephora runs almost the identical play at a completely different scale, using the big AI platforms for reach and discovery while tying every single interaction straight back to its own app and its own loyalty programme. And the holiday data backs the whole pattern up independently — retailers that deployed their own agents over the 2025 peak season grew about fifty-nine percent faster than the ones that sat it out.
Sam: So the rule underneath all of this is actually pretty simple, once you say it out loud. Show up inside someone else's assistant, sure — even go looking for that, chase the reach — as long as the basket, the data, and the loyalty all stay yours at the end of it.
Alex: The mistake was never showing up on the platforms. The mistake is showing up on them as an anonymous supplier, with nothing of your own left for the customer to actually come back to.
Sam: Okay, so if the front-end brands are genuinely at war with each other over the relationship — what's actually happening underneath all of that, with the real technical plumbing?
Alex: This is the part I find almost funny, honestly — the front end is a genuine knife fight, and one level down, everyone's quietly agreeing with each other. Ask "which protocol wins" and you're asking the wrong question entirely, because they're turning out to be layers of one single stack, not rivals competing for the same job.
Sam: Break the layers down for me properly. This is the bit we actually mapped in real detail back in episode nineteen — MCP, the agent-to-agent handshake, the payments layer underneath it all.
Alex: Right, so think of an actual agentic purchase like sending a parcel through the post, start to finish. First, the agent has to read the shop's shelf — what's in stock, what it costs, the actual specs — and that's the Model Context Protocol, MCP, doing the reading. Then it has to actually pack the cart, which is a commerce protocol's job — OpenAI's version is called ACP, Google's is called UCP. Then it needs real proof that you, specifically, authorised this spend — that's Google's Agent Payments Protocol, AP2, essentially the signature on the parcel that says "yes, I sent this." And underneath every single layer of that, the actual delivery truck — the card rails themselves, Visa, Mastercard, or newer crypto-native rails like x402.
Sam: Four completely separate jobs, and they nest neatly into each other instead of competing for the same slot.
Alex: And you can see the convergence directly in who's actually signed up for it. Google's Universal Commerce Protocol was announced at the retail industry's big January 2026 gathering, co-built alongside Shopify, Etsy, Wayfair, Target and Walmart, and endorsed by more than twenty other partners on top of that — Visa, Mastercard, American Express, Stripe, Best Buy, Macy's. AP2 launched with sixty-plus partners already signed on, day one. Even OpenAI's own Stripe-built protocol, and the card networks' own separate schemes, all plug into that same underlying payment-authorisation model in the end.
Sam: Companies that are fighting each other bitterly over who owns the customer, quietly agreeing on the pipes underneath them.
Alex: For exactly the same reason the labs gave away the agent standards in the first place, if you think about it — the alternative, a totally fragmented mess of incompatible checkouts that nobody can reliably build against, is worse for literally everyone in the fight, including the fighters themselves.
Sam: Right, and then there's the law — which I genuinely think just moved in a way that matters more than any single product launch we've covered this year.
Alex: This is the case behind that "federal court told Amazon" line we actually opened the episode on. Amazon sued Perplexity, trying to stop its Comet agent from shopping on Amazon's own site, on a customer's explicit behalf. And to be fair to Amazon here — they genuinely had grounds to be annoyed. They say they warned Perplexity at least five separate times starting late 2024, put an actual technical block in place by summer 2025, and then watched Perplexity ship an update that routed around that block within a single day of it going up.
Sam: That's fast. That's genuinely fast.
Alex: In March 2026, Amazon won a preliminary injunction, arguing Comet's access violated the Computer Fraud and Abuse Act — that's the federal anti-hacking statute, originally written for a completely different era. Then, in August 2026, the Ninth Circuit reversed that injunction entirely.
Sam: And the reasoning behind the reversal is the part actually worth holding onto, longer than any of the numbers.
Alex: The court said, plainly: when you instruct an agent to act for you on a website, it is you accessing that site — not the software vendor who built the agent. Picture it like sending your usual dry-cleaner your ticket by courier, instead of walking in yourself. The shop isn't being trespassed on by some stranger off the street. You're still the one being served. The courier is simply your hands, for that one errand, on that one day.
Sam: So a company can no longer just unilaterally declare that an agent its own customer personally chose and authorised is automatically a trespasser the moment it shows up.
Alex: That's a genuinely foundational call for everything downstream of this. Let it breathe for a second, actually, because it's the densest thing we'll say all episode.
Sam: Take your time.
Alex: The walled garden can absolutely still be the best individual shopping experience out there — nothing about this ruling says otherwise. It just can't be a legal fortress anymore, on top of being a good product. Open plumbing underneath, plus a legal presumption now genuinely favouring the agents customers themselves choose to send out on their behalf — that combination points away from any single company controlling the entire flow, no matter how big, or how well-personalised, that company's garden happens to be.
Sam: And it actually answers that question you raised right at the top, doesn't it — who's accountable when the agent gets it wrong.
Alex: It does, and cleanly. If the legal position is that you are the one accessing the site the moment you send your agent to do it, then you're also the one who's still transacting, in the eyes of the law — the agent is your hands, not an independent party. The accountability doesn't float off to whoever built the software. It stays with whoever authorised it. That's tidy in principle. It's going to get genuinely messy in practice the first time an agent buys the wrong size, the wrong quantity, or the wrong thing entirely — but the legal starting point, at least, just got settled.
Sam: Which, tying it back to Amazon's CEO from earlier — he's still probably right that his agent gets the pricing and the personalisation right more often than a stranger's. He just can't lean on the law anymore to keep the stranger's agent off his lawn entirely.
Alex: Exactly that. The moat still helps him compete. It just stopped being a wall he's legally allowed to build across the front gate.
Sam: Okay, so — practically, after all of that. If I actually ran a retail business right now, what would you tell me to go do first thing Monday morning?
Alex: This is the part that resolves into a surprisingly short, concrete list — and the reason it's short is that a specific set of moves pays off no matter which platform, or which protocol, eventually ends up winning this fight. The organising idea underneath the whole list is legibility. An agent can only recommend, compare and buy what it can actually read and act on cleanly. A product that's invisible to machines is quietly becoming invisible to a growing slice of real demand, whether you notice it happening or not.
Sam: So what's actually on the "just go do it" list?
Alex: Five things, roughly in order of cost. First — real-time, structured product and inventory feeds: machine-readable price, stock level, specs, the works, kept current, not a stale export somebody updates once a month. Second — making sure you're not accidentally blocking the AI crawlers that read your site in the first place, and that your product schema — price, SKU, stock level — is genuinely clean; that one, specifically, is roughly an afternoon's work, nothing more, and it's the single cheapest thing on this whole list. Third — a clean checkout and availability interface, so an agent can actually transact without having to scrape your site like it's 2005 and guess at what's actually in stock. Fourth — deploying, or at the very least seriously planning, your own first-party agent, so you become a destination in your own right, and not just a supplier somebody else's agent happens to visit on its way past. And fifth — keeping hold of your customer data and your loyalty relationship, full stop — the one genuine asset none of the platforms can actually take away from you, no matter who ends up winning this fight.
Sam: Can you actually make that concrete for me, though? What does "invisible to an agent" look like in practice, versus "legible"?
Alex: Take two hypothetical hardware stores selling the exact same drill. Store one's website has the price buried in a product description paragraph, no structured stock field, and its robots.txt file quietly blocks the crawlers these tools use to read the web. An agent comparison-shopping for that drill literally cannot see it — not "ranks it lower," cannot see it at all, the same way a search engine can't index a page it's blocked from crawling. Store two has a clean feed: price, stock count, spec sheet, all machine-readable, crawler unblocked. The agent reads both. Only one of them is even in the running.
Sam: So it's not really a ranking problem yet. It's a "does this business exist to the agent at all" problem.
Alex: Right now, for a lot of retailers, that's exactly what it is. And that's precisely why it's the first rung, not the fifth — you can have the best price in the country and still lose the sale to a worse one that the agent could actually see.
Sam: And here's the part I like — you said "roughly in order of cost," and the cheapest items on that list are also, not coincidentally, the ones that matter most right away.
Alex: That's the whole trick of it, honestly. Think of it like buying insurance versus building a house extension. The cheap first rungs — the clean feed, the unblocked crawler — those are basically the insurance premium: small, and it protects you across a whole range of different possible futures, whichever one actually shows up. The expensive rungs — the fully transactable checkout, your own first-party agent — those are the house extension. You don't build the extension the day you get married. You stage it once the actual numbers, the real transaction volumes, tell you it's genuinely justified.
Sam: Cheap rungs now, as insurance. Expensive rungs staged later, as the volume proves itself out.
Alex: Exactly — never one big, all-at-once rebuild that waits patiently for the picture to become perfectly clear, because by the time it's finally clear, you've already lost a year of demand you could have cheaply captured for almost nothing. And the urgency here is real, not theoretical — research from early 2026 found around forty percent of e-commerce businesses were still, right now, mid-way through standardising their product pages for agentic AI, and roughly a third of them hadn't even started.
Sam: So most of the industry is basically sitting on the cheap, obvious insurance policy, and just... hasn't bought it yet.
Alex: Which, in a channel that's actively compressing twenty-five years of adoption into five, means being a year late isn't a minor lag you can shrug off. It's a full year of accelerating demand getting quietly routed straight to whichever competitor an agent could actually read properly.
Sam: Alright, flip side of the same coin — what's the actual mistake here? What should a retailer specifically not do?
Alex: The mistakes here mostly share one single root — treating a genuinely unsettled market as though it already had a declared, permanent winner. Don't re-platform your entire business around whichever assistant happens to be ahead this particular quarter; OpenAI walking back Instant Checkout inside six months is a live, recent warning about how fast "the obvious winner" can quietly change its own mind. Don't commit exclusively to one single protocol either, when we just watched them converge into one shared stack in real time — you'd be paying good money to back one horse in a race that's actively turning into a relay.
Sam: And don't trade the customer relationship away for reach. That's the one that actually scares me most, hearing all this.
Alex: That's the big one, and you're right to flag it. The platforms will happily offer you distribution in exchange for your data and your direct line to the shopper — and the trust numbers from earlier tell you exactly why that's a bad trade to make. The relationship itself is the genuinely scarce asset in this entire story. Once you've handed it over to someone else, it is very hard to ever get back.
Sam: So when do you actually move, then, concretely? What's the real trigger here, if it's not a date on a calendar?
Alex: You act on the cheap, no-regret list right now — this quarter — precisely because it's cheap, and its value doesn't depend on resolving a single one of the open questions we've spent this whole episode on. You defer the platform-specific bets until the front-door war has a genuinely clearer shape to it, and the convergence we just walked through, plus the direction the law is now visibly drifting, both suggest that's a matter of a couple of years, not ten. The forcing function isn't a date circled on anyone's calendar. It's the plain, compounding fact that invisibility to agents is quietly becoming invisibility to demand itself — and that slice keeps growing, every single quarter you wait to act on it.
Sam: So, wrapping all of this up — where does it actually land? Give me the honest version here, not the confident-sounding one.
Alex: The honest version is: the direction here is genuinely knowable, the exact details aren't, and it would be dishonest of us to pretend otherwise, this far out. Over the next two years, expect agent-assisted buying to just become completely ordinary for the boring, repeatable basket — the batteries, the ink, the usual weekly order — while the emotional cart lags well behind it, and most fully autonomous checkout stays supervised, a human still nodding along at the final step. Over five years, expect agentic commerce to actually reach that fifteen-to-twenty-five percent of e-commerce the forecasts describe, with the control war between the platforms and the retailers mostly — though genuinely never completely — settled into a blended market where nobody owns the entire flow outright. And over ten years, expect agents to be a primary channel of demand, full stop, no caveat — and the gap between the retailers who made themselves legible early and the ones who simply waited it out will have widened into the actual, lived difference between owning a customer and being reduced to a commodity supplier for somebody else's assistant.
Sam: Which is honestly a genuinely good thing to end on, when you put it that way — because the move you should actually make today doesn't depend on getting any of those longer-range guesses right at all.
Alex: That's exactly it, and it's the whole point of doing this episode. Become readable to machines. Keep your customers close. Act now on what's cheap and close to certain, and defer what's expensive and still genuinely unknowable. It's the rare strategy in any of this that turns out to be right in pretty much every single version of the future the evidence still leaves open to us.
Sam: What actually gets me, sitting back from all the numbers, is the shape of it — a court, a discount retailer's chatbot, and a handful of protocol acronyms nobody outside this industry has heard of, quietly deciding something as basic as who you trust to spend your money for you. That's not a small question. It just arrived wearing a genuinely boring disguise.
Alex: It always does, though, doesn't it — the actually big shifts rarely announce themselves. E-commerce didn't feel like the future while it was happening either. It felt like slightly cheaper batteries and a delivery van showing up more often. This is going to feel exactly that boring, right up until the quarter it very obviously isn't anymore.
Sam: And that's it for today — thank you so much for listening.
Alex: I hope you came away seeing a bit more clearly where this particular piece of the AI revolution is actually heading. It's a genuinely fast-moving picture with a brutally short knowability horizon on it, and honestly, that's exactly what makes it worth following closely, week to week, rather than checking back in a year and hoping you didn't miss anything important.
Sam: And one honest note on how this show actually gets made — it's AI-generated. A custom stack of AI tools does the research, the analysis, the fact-checking, and the illustrating, built mostly to help understand all of this from first principles, and published for anyone who wants to follow along with it.
Alex: AI-assisted, fact-checked, worth a second look — that's the actual standard we're holding ourselves to, every single episode, this one included.
Sam: Before we go, one genuinely useful thing you can do right now — follow the show. Whatever app you're listening on this second, there's a follow or a plus button somewhere on the screen. It's one tap, it's completely free, and it does two things: you get every new episode the moment it actually lands, and for a small independent show like this one, a follow is honestly the single biggest lever there is for helping it grow and reach other people out there trying to make sense of all this too.
Alex: So if this was worth your time today — go ahead and hit follow.
Sam: And one quick thing before we actually wrap up here — if there's something in this episode you'd genuinely push back on, or a thread you want us to pull harder on next time, tell us. Podcast at connectiveshift dot com. We read every single message that comes in, and it genuinely shapes what we go dig into next.
Alex: See you next time.