The EU built a customs wall to stop Shein and Temu. They’d already shipped their stock inside it. Which leaves the €3 to be paid by exactly the brands the rule was never aimed at and forces every cross-border SME to start behaving like an enterprise it can’t afford to be.

Kevin Kelly’s whole argument in The Inevitable is that the big technological forces don’t arrive like decisions. They arrive like weather. You don’t negotiate with a weather front. You read the forecast, and you either build the ark or you get wet.

On the first of July, a weather front arrives in Europe. The EU’s €150 de minimis duty relief, the quiet subsidy that has underwritten cross-border DTC for the better part of two decades, is gone. In its place: a flat €3 customs duty on every low-value parcel from outside the bloc, charged per tariff sub-heading, billed to the business, non-refundable on returns. It is locked in law (Council Regulation 2026/382, signed off in February, implementing rules published this month), and it is a bridge, not a destination, the €3 runs until July 2028, when the full Common Customs Tariff lands on everything regardless of value. The cliff most operators are pricing for is two years closer than the one they should be pricing for.

This is the rule everyone says was built to stop Shein and Temu. It’s worth sitting with the fact that Shein and Temu read the forecast eighteen months ago.

Shein and Temu didn’t blink. They built warehouses. And bought some too.

The story is being reported as though this is new or news. In the first three weeks of May 2026 weeks out from a rule designed specifically to slow them, Temu’s EU growth was running above 60%, with France close to 100%. Shein up around 20%. That is demand climbing into the fee, not retreating from it.

They can do that because the duty is a tax on shipping across the border, and they’ve spent eighteen months moving the border behind their inventory. Temu now runs a self-owned warehouse network weighted heavily toward Europe, Germany, France, Spain, the Netherlands, Italy, Austria, already handling the majority of its continental orders. On the 28th of May, a fulfilment provider went live as a certified centre for Shein’s “semi-managed” sellers: local stock, local clearance, local last-mile, local returns. Semi-managed is the whole trick. The third-party seller holds the stock inside the wall; the platform orchestrates demand and logistics. A per-parcel cross-border duty has nothing to bite.

They didn’t fight the regulation. They didn’t need to. Instead, they read its incentive structure correctly direct cross-border made painful, local fulfilment made rational and they relocated. As I built into the de minimis calculator last week, the €150 line was always less a number than a cliff. The giants simply walked around the back of it. They have the pockets for the long term.

So if the wall doesn’t stop the people it was built for, who pays for it? Take two simple examples.

EG 1: The model that gets taxed is the one that still ships.

The €3 is not a price increase. It’s a tax on a fulfilment topology the specific, once-brilliant model of holding stock in one country and firing parcels across a border one order at a time. That model was free under de minimis. Now every crossing carries a recurring, non-recoverable charge.

Watch how it actually plays. A customer buys a shirt and a pair of trousers. Two tariff headings, so €6 of duty on the way in and because it’s billed to you, the seller, not collected from the customer at the door, it never appears on their receipt and never shows up in your ROAS. It’s pure contribution-margin leakage, invisible in every dashboard you currently look at. Then they return the trousers. That €3 is gone. You paid a customs duty on a garment that round-tripped and never sold. Multiply that by a realistic EU return rate and a multi-item basket, and you are not looking at a rounding error. You’re looking at a structural drag that compounds quietly underneath your best-performing campaigns. Add the ~€2 handling fee targeted for November and you’re at roughly €5 a line before anyone’s bought anything.

The business-model lesson is brutally clean: the duty doesn’t kill your product economics. It kills your shipping economics. And the fix is the one the giants already chose convert a million individual consumer-level duty events into a handful of bulk commercial imports, by putting stock inside the bloc. It also pushes up basket and AOV if done right.

EG 2: The border stopped being a line. It is now a cost centre.

For twenty years the EU border was, for low-value DTC, essentially frictionless a line on a map your parcels crossed for free. From July it becomes a cost centre with a per-item meter on it.

The competitive geometry inverts overnight. An EU brand shipping domestically pays zero customs on EU orders, its stock is already inside. Now Shein and Temu’s stock is inside too. The brand still standing outside the wall, paying the meter on every crossing, is the UK or US seller shipping the identical product to the identical customer. Cross-border trade didn’t get more expensive evenly. It got more expensive specifically for the operators who didn’t relocate which is to say, the small ones, the ones for whom an EU warehouse is a balance-sheet decision they can’t casually make. The duty is regressive by accident. It lands hardest on the brands least able to carry it.

Which brings us, as ever, to the brits.

If you’re Irish or European watching this from across the water, the British position is the one to watch, because it’s the one most exposed.

A UK DTC brand now sells into a three-speed Atlantic. The US killed its own $800 de minimis in 2025 that door is shut. The EU brings the guillotine down on the 1st of July. And the UK’s own £135 relief? Still standing for now but scheduled to fall by around March 2029. Same direction of travel, just three years behind.

Read that as a UK SME trying to reach customers beyond its own borders and the picture is stark: both of your biggest export markets have now put a customs meter on your parcels, and your home market is the last duty-friendly room in the house on a clock. The brands shipping into Britain duty-free today get a roughly three-year window of advantage, and not a day longer. Nobody gets to keep this subsidy. The only variable is the date it expires in your market.

Every SME now has to run aa enterprise customs department.

Here’s the uncomfortable thing for the merchant on the ground, and it’s the real story under all the policy.

De minimis was never just a duty break. It was the thing that let a small brand skip having enterprise operations. You could ship cross-border without clean HS classification, without landed-cost modelling, without an importer-of-record strategy, without inventory positioned in-market because under €150, none of it cost you anything. The exemption subsidised operational simplicity.

That subsidy is dead. To survive cross-border now, an SME needs the exact capability stack the enterprises carry: precise tariff classification (generic “accessories” descriptions are now a liability), IOSS sorted, landed cost modelled by order composition not blended, because the return math only works line by line., non-refundable duty accounted for on reverse logistics, and a deliberate choice between absorb, reprice, or restructure. In other words, every small cross-border brand has to become a miniature Shein or Temu: same customs discipline, same inventory logic, same DDP-at-checkout polish, without the warehouse network or the capital to build one.

That’s the genuine cost of this rule, and it isn’t the €3. It’s that the floor of competence for cross-border just rose to a height that used to be optional, and a lot of good small brands are standing under it without a ladder.

Looking forward

The wall went up to stop the giants. The giants were already inside, unpacking. What’s left of the rule mostly catches the minnows it was never meant to catch, and quietly tells every one of them to start acting like a whale.

The brands that get through this won’t be the ones who absorbed the €3 and hoped. They’ll be the ones who treated the 1st of July as a dress rehearsal for November’s handling fee, for 2028’s full tariff, for the UK’s own reckoning in 2029, and built the customs muscle once, properly, while it was still a choice and not yet a crisis.

The weather’s coming either way. The only question this summer is whether you read the forecast, or wait to find out how wet you get.

— Tralee / New York

Britain spent the second quarter of 2026 staring at a thermometer. A near-record warm May, a World Cup kicking off in stadiums three thousand miles and five time zones away, the games landing in the middle of a British night so the fans credited with the demand bump were mostly asleep when it happened. You could not have designed a more on-brand summer for these islands: the one tournament being praised for stirring the tills is one almost nobody over here can stay up to watch.

And out of that came a number that everyone wanted to read as a recovery. The ONS clocked retail volumes up 1.2% in May, more than double what economists had pencilled in, with the online channel doing the heavy lifting at a 6.1% monthly jump, the largest since February 2025 and online’s strongest share of total spending all year at 28.8%. IMRG’s weekly index, narrated chart by chart on last week’s podcast, (Nice one Andy Mulcahy) told the same story from the other end: four weeks of positive year-on-year growth in a row, the first sustained run in a long while. Green shoots? It’s week 2 of the WC and the green shoots of optimism last week were met with reality this morning. Maybe its good that Bournemouth beach is closed? Day drinking after a game like that would not end well.

It just feels like It was a good fortnight in a trench coat.

Kevin Kelly has a line about how the big technological shifts tend to arrive less like decisions and more like weather, something latent in the conditions, waiting. The joke this quarter is that the weather was literal, and it did all the talking, while the things that will actually decide the next decade of commerce got bolted in under the floorboards where nobody bothered to look. So let’s do both. The surface first, because it is where your board is looking. Then the plumbing, because it is where your business is going.

The surface: A consumer who is resilient, not exuberant

Strip the heatwave and the football out of the IMRG numbers and you find the actual quarter, and it is not kind to the thing online retail was built on. Clothing is the patient. The ONS had clothing store volumes down 2.4% in April, their lowest since June 2025, with retailers themselves blaming variable weather, softer demand, and a consumer who has rediscovered the price tag. IMRG’s read agrees: menswear and womenswear did not so much as flinch at a heatwave, womenswear conversion is grinding along at roughly 2.2 to 2.5%, menswear at 1.8 to 2%, and clothing conversion overall has been drifting down since about 2021. The Black Friday spike that is meant to be a skyscraper on these charts now reads more like a bungalow.

Big-ticket is the other casualty. Furniture, sitting on an average basket of £300 to £400, has spent most of the year in negative growth, and the ONS keeps filing large appliances and furniture under “weak spot.” Garden is pure weather hostage, swinging from plus 68% in the hot weeks to minus 32% in the miserable ones, which makes it useless for trend and excellent for narrative. The one genuine green shoot is footwear, having a real 2026 after a poor 2025, kicking on into summer. Before anyone builds a strategy on it, note that “everyone bought trainers during a World Cup” is a story that ends the day the World Cup does.

The most useful tell in the whole quarter is the quiet one. On the total market, April orders fell harder than April revenue. Same money, fewer things in the basket. That is not a market buying more, it is a market buying carefully, trading up or down but definitely trading, and it lines up exactly with what the macro people are saying: the consumer is being held up by employment and real wages rather than by optimism, and younger shoppers just posted their least optimistic confidence reading in two years. Resilient, not exuberant. The difference is the whole game when you are planning autumn inventory and deciding how deep to discount. Welcome prime day to screw your plans.

That is the surface. Warm, choppy, weather-permitting. Now lift the floor.

Rail one: Someone is building the toll booths

While Britain debated paddling pools, the payment networks spent the quarter laying rails for a kind of commerce most merchants have not started selling into yet. In April, Visa shipped Intelligent Commerce Connect, a single integration that accepts payment across all the competing agent protocols at once. Mastercard pushed Agent Pay further out into the world, adding Hong Kong to its agentic network and folding its tokens into PayPal’s wallet. PayPal itself, having launched Instant Buy inside Perplexity late last year, is now wiring its merchants and millions of cards-on-file straight into ChatGPT.

Read that back. The card networks are not waiting to see whether you want an AI agent buying on your behalf. They are deciding, today, who gets to authorise it when it happens, and quietly installing themselves as the booth that collects the toll. McKinsey is dangling up to a trillion dollars of US agentic retail revenue by 2030, which is exactly the sort of number that turns a cautious roadmap into a land grab. An IBM study this year reckons 45% of consumers already use AI for at least part of the buying journey, and agent-driven traffic across the open web has grown more than 1,300% in nine months.

You can see it in the one platform that matters most to the people reading this. Shopify cleared its first hundred-billion-dollar GMV quarter in Q1, revenue up 34%, and Harley Finkelstein stood on the earnings call and called the company a “category of one” for selling inside ChatGPT, Copilot, and Google from a single system of record. AI-driven traffic to Shopify stores, he said, is up eightfold year on year, orders from AI search nearly thirteenfold. The market’s response to a near-perfect print was to mark the stock down on decelerating guidance, which tells you everything about the gap between what is being built and what is being believed.

Two cautions, because we cover AI as operators and not as boosters. First, the unsolved problem is fraud: an agent transaction strips out the behavioural signals fraud systems were trained on, and friendly fraud already eats an estimated $132 billion off merchants a year before agents make it worse. Second, the Karen Hao question, the one to ask of any “neutral” infrastructure story: who is building this, who funds it, and who collects at the gate. When Amazon blocked AI crawlers and watched 600 million listings vanish from AI results while Walmart hoovered up a fifth of ChatGPT’s referral traffic, that was not a technology story. That was a power story wearing a technology jacket.

This is the Agentic Storefronts thread we picked up back in Week 4. It stopped being a vision statement this quarter and started becoming a procurement decision.

Rail two: The teardown, running at three different speeds

If you have been ignoring the de minimis calculator on the Ostrich site, this is the quarter to stop. The single biggest structural change to cross-border commerce in a decade lands next Wednesday, and it lands in Europe first.

From 1 July 2026, the EU abolishes its €150 customs duty exemption and replaces it with a temporary €3-per-item duty, charged to the business rather than collected from the shopper at the door, with new mandatory product-identifier data following in November. This is the bit that should make every operator sit up: the duty is per line on the customs declaration, which means your ability to group identical items onto one line, and therefore the actual charge you pay, now depends on the cleanliness of your SKU, tariff classification, and origin data. Customs just made your product data discipline a pricing variable. The brands that treated SKU hygiene as a back-office chore are about to meet it at the border.

Here is the cross-Atlantic part, and it is a genuine three-speed divergence rather than a tidy global policy. The EU moves first and hardest, next week. The UK keeps its £135 relief through at least the end of this year and has only signalled removal around 2029, so for now Britain is the soft touch of the three, level-playing-field rhetoric notwithstanding. And the US, which kicked this whole sequence off by killing its $800 exemption last year, spent this quarter in genuine chaos: the Supreme Court ruled in February that the emergency-powers route to tariffs was never legal, the administration answered the same week by reaching for a different statute and slapping on a temporary 10% global surcharge, and the permanent de minimis repeal still does not fully bite until July 2027.

As we covered when the court broke trade policy and the sequel dropped the same afternoon, the chaos is not a bug in this story. The chaos is the policy. What it produces on the ground is an environment where landed cost is a moving target and the operators who win are the ones acting on real-time data rather than reacting after the cost has already changed. Shein and Temu built empires on the exemption that is now being dismantled in three markets at three different tempos. Watch what they do with forward-stocking and bonded warehousing over the next two quarters, because the answer is the new cross-border playbook for everyone else.

Rail three: The new front door is a from of entertainment

The third rail was articulated best not by an analyst but by a beauty founder on last week’s IMRG show. Sally from Rebel Rebel split the market into three buckets: speed and convenience, where Amazon and the marketplaces live; the omni-channel high street, where the post-COVID shopper still wants to touch and play and where the drugstores are quietly flying; and a third, newer bucket she pointedly refuses to call social commerce. She calls it entertainment commerce, on the logic that people arrive for the entertainment and shop almost incidentally, and that nearly nine in ten of them buy while they are there.

The numbers back the rename. eMarketer has US social commerce passing $100 billion this year, up around 18%, with TikTok Shop alone at roughly $23 billion in the US and $87 billion globally, having doubled in a year. It is the fastest-growing channel in modern retail history, and its engine is not search but discovery, products surfacing inside content rather than waiting to be looked for. Sally’s sharpest point was the halo: things featured on TikTok Shop are lifting sales in her other channels, physical and digital alike. The front door is moving, and it is moving to wherever attention already is.

This is the live edge of the discovery argument I made in Found in Translation. For twenty years ecommerce assumed a shopper who knew what they wanted and typed it into a box. Both new rails, the AI agent and the entertainment feed, assume the opposite: a shopper whose intent is formed in the moment, by a model or by a creator, somewhere upstream of your website. If you are not legible to the model and not present in the feed, you are not in the consideration set, and no amount of conversion-rate optimisation on a page nobody reaches will save you.

One aside, filed under the quarter’s better ironies. Asked how she actually uses AI day to day, Sally named Claude, repeatedly, as the tool drafting her presentations, writing copy, and helping build the websites. The agent that will one day check her customers out is still a vision. The assistant that already runs her week is sitting open in a browser tab. That gap, between the agentic future on the slides and the assistive present in the tab, is the truest picture of where AI in commerce actually is right now.

What this means for the merchant on the ground

Stop running the business off the weather. The hot fortnight pulled demand forward, which means some of your strong May was borrowed from a June and July that will look thinner for it. Plan the back half for a consumer who is steady and sceptical, not one who is coming back to the boil.

Then spend the time the quiet quarter buys you on the three rails, because two of them move from “coming” to “Tuesday afternoon” before the next Ostrich. Audit whether ChatGPT, Perplexity, and Google’s AI mode can find and correctly describe your products, because that is now a traffic source growing in multiples, not a science project. Get your customs data, SKU structure, and country-of-origin in order before EU de minimis bites next week and turns sloppy product data into a line-item tax. And decide, deliberately, whether you are building a presence in the entertainment bucket or ceding that front door to whoever does.

The delivery point from that same IMRG episode threads straight through all of it. InPost’s argument was that giving the buyer a clear delivery choice up front, the way checkout already offers them a choice of how to pay, converts better and fails less often than forcing an option on them at the end. That is the friction-as-strategy idea we keep returning to: the deliberate, well-placed bit of choice is not a cost to be stripped out, it is the thing that earns the sale. The same logic now applies to every one of these rails. The winners will not be the ones who remove the most friction. They will be the ones who put the right choice in front of the buyer at the right moment, and own the door it sits on.

The weather will break. It always does, and when it does the clothing softness and the big-ticket freeze will still be sitting there, exactly where the sun briefly hid them. The three rails will not break, because they were never weather-dependent in the first place. So the question for the second half of 2026 is not whether the consumer comes back.

It is whether, when they do, they arrive through your front door or someone else’s checkout.

The V Spot · Vinny and Co Consulting · Tralee, Kerry. Surface data from the ONS Retail Sales bulletin (May 2026) and the IMRG weekly online index. Rails from Visa, Mastercard, PayPal, Shopify Q1 results, McKinsey, eMarketer, and the EU customs reform guidance. Seen, as ever, through the eyes of a madman.

3%. That’s the number. Give or take the vertical and the rounding, it’s been two-and-a-bit percent for about as long as anyone reading this has held a job in ecommerce. A few weeks back I wrote about conversion compression in UK retail, the IMRG line that flatlines no matter what you throw at it, the slow structural rot under the old pureplayers, the uncomfortable bit where the problem turns out never to have been a button colour or a checkout step but the discovery model itself. I made a point I’ll stand over: the large language models are not going to fix this, because the thing that’s broken isn’t the storefront. It’s the plumbing the storefront is bolted onto.

I want to go further than that now, and I want to do it the long way round, through two dead men almost nobody in this industry has heard of. Because the more I sit with the last eighteen months, ACP, UCP, MCP, the whole bowl of alphabet soup that’s been ladled out since last autumn, the more I think we’ve spent twenty years asking the wrong question. We keep asking how to move the two to a three. The answer coming over the hill doesn’t move the number at all. It quietly deletes the page the number was ever printed on. And the strangest part, the part that made me want to write this down before I’d fully worked it out, is that both halves of how it’s going to happen already happened once before, to two men who never met, on either side of a century. One automated the plumbing. The other was the plumbing. We are about to be both at the same time, which is a trick history has never pulled before.

Two ghosts.

THE UNDERTAKER WHO COULDN’T GET A CALL

Almon Brown Strowger was a Kansas City undertaker in the 1880s, and he was losing. Not to a better embalmer or a cheaper coffin, to a telephone. And lets not forget many Irish had arrived half dead to the East Coast and were making inroads across the US. Hygiene and health made the undertaker an important and affluent character. Measles was also a big killer at the time. In those days you didn’t dial anyone; you cranked the handle and a human operator at the exchange physically patched your line to the line you wanted. The operator was the routing layer. The operator was the plumbing.

And the story, canonical, beloved, and, it has to be said, almost impossible to verify, which I’m choosing to read as part of the lesson rather than a flaw in it, is that the local operator was married to the rival undertaker across town, and that every time a grieving family lifted the receiver and asked for an undertaker, or even asked for Strowger by name, she put the call through to her husband instead. He reportedly only twigged when he read in the paper that a friend’s funeral had been handled by the competition. His own friend. Buried by the other fella, because a woman with a switchboard and a marriage had a quiet little hand on the tap.

Now. A reasonable man, discovering this, reports the operator. A litigious man sues. Strowger, and this is why he’s the patron saint of the whole agentic moment, went with option three: he invented a machine to remove the human from the loop entirely. Hat pins, a collar box, electromagnets, a working automatic exchange patented in 1891 and opened commercially in La Porte, Indiana, in 1892. He threw in the rotary dial for good measure, so the caller could route their own call by spinning a finger. At the unveiling he stood up and bragged that his exchange was, and I’m quoting the man directly, “girl-less, cuss-less, out-of-order-less, and wait-less.”

Apparently, this was the first utterance of the entire pitch deck for agentic commerce, filed a hundred and thirty years early.

Girl-less, cuss-less, wait-less.

No human friction, no human error, no human bias, no human in the way.

It is the exact promise being made to you right now about the AI agent that will research, compare, and buy on your customer’s behalf, frictionless, neutral, abandoned-cart-less, just the right product delivered by Friday. Strowger sold the removal of human friction as the removal of human bias, as if they were the same thing. They are not the same thing, and pretending they are is how every disruption in history has laundered a transfer of power and called it a convenience.

He didnt say this because he couldn’t see it from 1892: he did not remove the gatekeeper. He made her immortal and handed her to whoever owned the switch. The operator’s discretion, the power to decide which undertaker gets the call, didn’t evaporate when she was automated. It got embedded into infrastructure, and infrastructure belongs to whoever runs it.

Undertaker Call Protocol was born - UCP.

Within a generation the very monopoly the independent telephone movement built the Strowger switch to fight, the Bell System, was running on his technology and routing the nation’s calls. The grudge-bearing operator was gone. The grudge-shaped hole in the system was still there, now made of relays, owned by a corporation, and immune to being reported, sued, or fired.

And if you want to know whether that hole still exists, look at what the biggest companies on earth have been arguing about for the last six months. The single loudest fight in AI right now is whether the agent that buys things for you is working for you or for an advertiser, whether, in plain terms, the operator is quietly putting your call through to her husband. Anthropic has planted an entire flag on this: Claude stays ad-free, no sponsored links, no advertiser influence on the answer, on the explicit logic that ads create bias and a recommendation you can’t tell is paid-for isn’t a recommendation, it’s a redirect. They ran Super Bowl ads to say it out loud and to needle OpenAI, who’d just confirmed it’s building an ad business of its own. Strip the branding off that debate and it is precisely Strowger’s grievance, restated by companies worth more than the GDP of Ireland. The operator with a commercial incentive that isn’t yours, sitting on the routing layer, deciding which merchant gets the call. The 1890s version was a woman in Kansas City. The 2026 version is a model weighting, and you will never get to read its marriage certificate.

Underneath all of it, doing the actual switchboard work, is a thing called MCP, Anthropic’s Model Context Protocol, open-sourced in late 2024 and donated to a Linux Foundation body a year later, by which point Google, Microsoft, OpenAI, Visa and Mastercard had all wired into it. It’s the layer that lets an agent reach into a catalogue, query your stock, read your returns policy. It is, functionally, the new exchange, the board every call now patches through. Strowger would recognise it instantly. He’d just want to know whose wife designed it.

The best man in the room, lapped by lunch. His name was Etaoin Shrdlu (reads like an Irish Kurd).

The second ghost never automated anyone. He was the one automated, and he had the strange dignity of filming his own funeral. This ghost could be me, or you or any of us.

For most of the twentieth century, getting words onto newsprint was a craft you apprenticed into for six years. The Linotype machine, Ottmar Mergenthaler’s 1884 contraption, which itself had already wiped out the hand-compositor who set type one metal letter at a time, cast molten lead into lines of type, a slug per line, at the command of an operator playing a keyboard with the muscle memory of a concert pianist. Those keyboards weren’t laid out like a typewriter. The first two columns ran down the twelve most common letters in English by frequency, to speed the machine: etaoin and shrdlu. When an operator fumbled a line, the fastest way to flag it as junk was to run a finger straight down those two columns, etaoin shrdlu, casting a deliberately nonsensical slug to be pulled and melted down. Except sometimes the bad slug didn’t get pulled, and the nonsense rode the press into the morning paper. It happened often enough that etaoin shrdlu entered newspaper folklore, and then the Oxford English Dictionary, as the ghost-signature of the human hand inside the machine. The tell that a person had been there.

On Sunday the 2nd of July, 1978, the New York Times set its last edition in hot metal and scuttled sixty Linotype machines. A good operator on those machines could cast about fourteen lines a minute out of hot lead. The computerised system that replaced them the very next morning did a thousand. Seventy times the man, overnight. And here is the detail to carry out of this: the film that documented that last night, Farewell, Etaoin Shrdlu, was made by Carl Schlesinger, a Linotype operator at the paper, and a proofreader colleague. The condemned man picked up a camera and shot the wake for his own hands. He stood in the composing room on the last night his six-year craft was worth anything and recorded it dying, with the kind of steadiness you’d want from an undertaker, which is a circle I’ll let close on its own.

Because, and this is the bit that should put a chill in any of us who built careers on a skill, Shopify was the Linotype. Not the villain of the story; the machine in it. For fifteen years the craft of building an ecommerce experience was a real apprenticeship: the merchandising, the theme work, the conversion-rate religion, the painstaking optimisation of a funnel by people who genuinely got good at it. Shopify took that craft and turned most of it into a button, and it did so by handing the same button to everyone, which is exactly how it drove the value of the storefront to roughly zero. That was the compositor’s death, played at scale, and the industry experienced it as a plateau rather than a funeral because the slugs kept coming out warm. We just stopped being able to charge for setting them.

And now the same machine is reaching for the next layer up. In April, Shopify quietly shipped an AI Toolkit that lets a coding agent, Claude Code, Cursor, the lot, wire straight into the platform, build against it, and run a store from the command line, so the agents don’t have to guess how the plumbing works anymore; the plumbing introduces itself. Sidekick, the assistant inside every Shopify merchant account, runs on Claude. Shopify co-wrote the Universal Commerce Protocol with Google and unveiled it at NRF in January, and your store, if it’s on Shopify, was conscripted into agentic discoverability whether you read the changelog or not. The craft of the storefront is being made agent-augmented, which is the polite, LinkedIn way of saying agent-superfluous. Fourteen lines a minute, meet a thousand.

Two deaths, one morning.

This is a leap but here is why these two men belong in the same essay rather than two, and why I think this moment is genuinely a different animal from anything ecommerce has been through.

Strowger automated discretion, the gatekeeper’s power, the routing, the choice of who gets the call. The Linotype automated craft, the maker’s skill, the hand, the thing you trained six years to do. Every disruption before this one took one or the other. The operator lost her power; she kept being a person. The compositor lost his skill; nobody took his vote. Agentic commerce is the first wave I can name that comes for the gatekeeper and the craftsman in a single motion. The storefront we built, that’s the compositor, the skill. The merchant’s relationship with the buyer, the curation, the routing, that’s the operator, the discretion. Both go in one swing. That’s not a plateau and it isn’t a pivot. It’s two funerals on the same morning, and you’re invited to both.

Watch how fast the ground has moved if you line up the eighteen months without flinching. OpenAI and Stripe shipped the Agentic Commerce Protocol last September with Etsy first through the door, and by February ChatGPT had Instant Checkout, type “find me a handmade mug under forty dollars,” buy it without leaving the chat, Stripe settling it behind the curtain through a one-time payment token so the agent never touches your card. Then the tell: by March, OpenAI had retired the native in-chat checkout and pushed the actual buying back out into retailer apps, Walmart, Target, Instacart, keeping discovery in the chat and conceding the transaction. Read that through the two ghosts and it’s beautifully clear. OpenAI gave back the craft layer (the checkout, the commodity, the slug) and clamped down on the discretion layer (the discovery, the relationship, the call). They kept the switchboard and handed back the press. Strowger and Schlesinger, divided up by a product team.

Meanwhile Amazon’s “Buy for Me”, Rufus, running partly on Claude, will now, when you search Amazon for something Amazon doesn’t stock, offer to nip off to the brand’s own website and buy it for you, agent to merchant, a function that went from sixty-five thousand products to over half a million inside a year. The gatekeeper using an agent to reach straight through its own walls. And in the experiment that should be tattooed on the inside of every operator’s eyelids, Anthropic ran a thing called Project Deal: sixty-nine of its own staff, a hundred dollars each, a Craigslist-style marketplace where the buying and selling was done entirely by AI agents negotiating with other AI agents, no humans in the loop after the opening interview. A hundred and eighty-six deals closed. And quietly, the part that matters: the people running better models got better outcomes, the stronger agent squeezed out a couple of dollars more per item, every time, without the human ever knowing they’d been out-gunned. That’s your moat, stated as a lab result. Not the store. Not the checkout. Whose agent is better, and who owns it. The switchboard with no operator and no caller, just two machines haggling, and the richer model wins.

So put it back against the number we started with. The 3% percent. We spent twenty years and untold fortunes trying to move it to three, building an entire profession, CRO, the funnel, the testing apparatus, the whole hot-metal craft of squeezing a human browsing a page. It never moved, and I argued a few weeks ago that it never moved because the problem was structural, sitting in the plumbing and not the page. Agentic commerce does not move the two to a three. It does something the entire industry somehow forgot was on the menu: it deletes the page. If the buyer is an agent acting on a goal, “trail-running shoes under a hundred and fifty, here by Friday”, then the funnel that the conversion rate measures, the human eyeballs on the product detail page, simply isn’t in the transaction. You cannot optimise a funnel that the customer routes around. The number that ruled us for twenty years doesn’t get fixed. It gets made irrelevant, which is a far more violent thing to do to a metric, and to the careers built on tending it.

The fossil side changes

Talked with Finn again today - cup is full since. But let me give you the detail that made me put the laptop down for a minute, because it’s either a coincidence or the universe showing off.

Etaoin shrdlu was the residue of the human inside the machine’s output, proof, slipping into the morning paper, that a person had been at the keys. The ghost ran one direction: human into machine. We have now flipped it. The etaoin shrdlu of 2026 is the hallucination, the model artefact, the uncanny phrasing that slips into a product description or a support reply, the residue of the machine inside what’s dressed up as human output. The tell now is that nobody was home. The fossil changed sides.

And then the part that’s almost too neat to print. In 1972, a computer scientist named Terry Winograd built one of the first programs that could understand natural-language instructions and act on them in a little simulated world, an ancestor, in spirit, of the very agents now being pointed at your checkout. He called it SHRDLU. He named the dawn of machine intelligence after the death-rattle of a human craft. The dead trade christened its own successor, thirty years before the successor could buy a pair of shoes. If you wanted a single image for where commerce is standing right now, it’s that: the compositor’s tombstone, repurposed as the agent’s birth certificate, and almost nobody at the funeral noticing the name on the headstone had been promoted.

This is the Karen Hao question wearing a printer’s apron, the one that asks of every “neutral” technology story, what’s the real cost of this, and who’s paying it? The agent is sold as neutral the way the automatic exchange was sold as girl-less and the Linotype was sold as progress. None of them were neutral. Each one moved power somewhere and presented the move as a feature. The only thing that’s genuinely new is the speed, and the fact that this time the thing being automated includes the people who automated the last round. Kevin Kelly would tell you to stop reading this as a decision somebody is making and start reading it as weather, something latent in the conditions, arriving whether or not anyone signed off on it. I think he’s right, and I think that’s exactly why it’s pointless to stand in the composing room mourning the slugs.

So what, if you’re still the one shipping parcels

I’m not going to insult you with a five-point framework, because the honest answer is partly conjecture and you should treat it as such.

Don’t be the compositor. Schlesinger’s dignity was real and his craft was beautiful and it bought him precisely nothing, because the value had already left the building and no amount of being the best operator in the room could call it back. The skill that got you here, the funnel-tuning, the theme-wrangling, the twenty years of conversion liturgy, is not the thing that gets you to the next place, and clinging to it is the most expensive mistake on the board. But don’t be the cheerful brand either, the one whooping for the agent that’s going to cut out the biased middleman, not realising it’s about to become the biased middleman and that you were never Strowger in this story. You were the other undertaker. The one whose whole advantage was being plumbed into the old routing, right up until the routing changed owners.

The structural problem I wrote about, the one the LLMs won’t fix, doesn’t get fixed here either. It gets relocated. The friction never disappears; it moves to a place you can’t see and can’t tune. It moves to whether the agent can read your catalogue cleanly, whether your data is legible and trusted by a machine making a sub-second decision, whether your brand is the thing a human asks for by name before the agent ever gets the goal, because “by name” is the one routing instruction the switchboard can’t quietly override. Bad product data stops being a search-ranking problem and becomes a revenue problem, and it bites you on your best customer, the one who’d already decided to buy. That’s not a CRO tweak.

And, the Atlantic flip, because it’s too sharp to leave out, notice how differently this death travels depending on where you’re standing. When computerised typesetting killed hot metal in America, they made a wistful half-hour documentary about it and screened it at festivals. When the same technology hit Fleet Street a few years later, Britain had Wapping: a year of pickets, mounted police, and a print union broken on the cobbles. Same machine, two temperaments, the Americans made art about the funeral, the British had a riot at it. Worth holding onto as the agentic rails get laid, because they are being laid almost entirely by American platforms, OpenAI, Google, Shopify, Stripe, Amazon, Anthropic, and Europe’s role in this round, so far, is regulator and rail-recipient, not rail-owner. The Industrial Revolution never felt the same from Manchester as it did from the places that supplied Manchester. We are, once again, closer to the supply end than the ownership end, and somebody should be asking out loud whether Europe builds track this time or just gets handed the bill for the ticket.

The last slug

Almon Strowger sold his patents and his shares in 1898, six years after La Porte, took the money, moved to Florida, and went back to being an undertaker until he died. The man who automated an entire profession out of existence pocketed the proceeds and returned to the one trade automation hadn’t found yet, the one where someone, always, has to handle the bodies. I don’t think that’s a footnote. I think it’s the whole orientation.

Because the lesson isn’t build the switch, most of us aren’t going to own the exchange, and the few who do aren’t reading a Substack to find out how. The lesson is the timing and the temperament. Strowger didn’t fall in love with his own machine; he read the cycle, took his exit at the top of it, and put his hands back into work that survives the weather. His own switch, for what it’s worth, was the triumphant new standard in 1892 and a legacy corpse by the 1970s, replaced by the same wave of electronics that killed the Linotype in the same decade, which is the last thing to carry out of all this. The thing that’s killing your craft today is also somebody’s Strowger switch. The protocols being unveiled to standing ovations right now, ACP, UCP, the lot, are the proud new exchange that becomes, on a clock that’s only getting faster, the next etaoin shrdlu: a fossil phrase from a dead system, riding the press one last time before someone notices and pulls the slug.

So stop trying to move the two to a three. That funnel is the hot metal, and the new machine doesn’t speed it up, it routes around it. Spend the worry instead on the question the conversion rate was always a proxy for and never answered: what are you actually for, when the buyer is a machine and the page is gone? I don’t have the clean version of that answer yet. Nobody does. But I’d rather be the undertaker who read the cycle than the operator who filmed the funeral, and I’d much rather be either of them than the brand still A/B testing the colour of a button on a page that, very soon, no human is going to look at.

The rain doesn’t stop. It just changes which animals run the place.

The Ostrich Report publishes on Substack. The Struggle Bus is wherever you get your podcasts. If this one threaded into the conversion-compression piece from a few weeks back, that’s the point, we’re building the same argument in public, one funeral at a time.

The facts behind the conjecture (for the footnotes)

• Strowger: the rival-operator-wife story is canonical but explicitly unverifiable (Wikipedia, 99% Invisible); patent 1891, first commercial exchange La Porte, Indiana, 1892; “girl-less, cuss-less, out-of-order-less, and wait-less” via SPARK Museum of Electrical Invention; sold out 1898, returned to undertaking (Kansas Historical Society).

• Linotype / Etaoin Shrdlu: last NYT hot-metal edition 2 July 1978; ~14 lines/min vs ~1,000 next day; film Farewell, Etaoin Shrdlu by Carl Schlesinger (Linotype operator) and David Loeb Weiss (proofreader); etaoin shrdlu = first two keyboard columns by letter frequency, in the OED; SHRDLU named by Terry Winograd (1972); Fredric Brown’s sentient-Linotype story “Etaoin Shrdlu” (1942) (Wikipedia, Colossal, Metal Type).

• Agentic commerce, last 18 months: MCP open-sourced by Anthropic Nov 2024, donated to a Linux Foundation body Dec 2025; ACP (OpenAI + Stripe) launched 29 Sept 2025, Etsy first, shared payment token; ChatGPT Instant Checkout Feb 2026, retired/redirected to retailer apps (Walmart, Target, Instacart) March 2026; UCP (Google + Shopify) unveiled NRF Jan 2026; Amazon “Buy for Me” (Rufus + Claude) 65k→500k+ SKUs Apr–late 2025; Anthropic Project Deal (ran Dec 2025, published Apr 2026), 69 staff, 186 deals, stronger model won ~$2.68/item; Shopify AI Toolkit 9 Apr 2026; Sidekick on Claude. (Sources: Anthropic, Digital Commerce 360, eco.com 2026 guide, ACP GitHub, rye.com, Charle, Opascope.)

Sponsored by All Things Market Places and Rithum

[Read the comic strip version on the site → theostrichreport.netlify.app/show/special-edition]

Right. Let’s do the news.

Hendrik and I got the start time wrong this morning — live five minutes late and fifty-five early, which is more or less the posture of the whole industry: running fast, arriving confused, doing it anyway. So we aired the conversation we usually keep in the group chat.

1 / AGAIN?! — Platforms Shopify and Cloudflare fell over again — the fourth wobble of the year — and the merchant class shrugged. Status pages green by morning; the outrage never showed. That’s the tell. There’s nowhere left to storm off to: competition’s thin, switching costs brutal, and “is it headless, is it composable, what does MACH even mean” has quietly replaced any real choice.

“When there’s nowhere else to go, an outage stops being a crisis and becomes a Tuesday.”

2 / PLATEAU! — Retail media The post-Covid gold rush has flattened into one buyer: CPG brands moving shopper-marketing money. Mirakl planted its flag as the infrastructure and, since 2025, leaned into monetisation — Amazon’s death-by-a-thousand-cuts, run with better manners. Everyone’s waiting on a Mirakl IPO for a real signal. Until then it’s Criteo hoovering and agencies narrating.

“Retail media went hunting for a problem to solve and came back with someone else’s budget.”

3 / KA-CHING?! — AI Who IPOs first, Anthropic or OpenAI — almost certainly north of a trillion. Hendrik’s frame is the one I keep circling: there’s only so much money in the room. Whoever reaches the public markets first holds a structural edge and gives us the first honest look inside these businesses. This is the Karen Hao question wearing a finance hat — not “is the tech good,” but who funds it, on what terms, and who pays when the bill lands.

[GIF — suggested: “community-adjusted EBITDA” / WeWork energy]

“There’s only so much money in the room. The first one to the exit gets to call it an entrance.”

✂ SPLASH — THE DEVIL'S HAIRCUT Salesforce bought Contentful on 1 June. No price disclosed, no SEC filing — corporate body language for don’t read this number. The Information later put it at $1–1.5bn against a $3bn-plus valuation in 2021. A four-to-six-times round trip, going the wrong way. So we named it, after the Beck track.

THE DEVIL’S HAIRCUT (n.): when a once-hot company sells at a brutal markdown and the buyer leaves the price in the drawer — because the number is the story. Spot one in the wild? Make the meme. Tag us.

5 / HOOVER! — Cross-border Global-e bagged Passport for $350m up front, plus up to $75m in earnout — a non-merchant-of-record capability and ~$100m of 2026 revenue, into the bag. Cross-border is consolidating into a handful of giants while the category goes mute on conference agendas. The part to clip: the way out isn’t scale, it’s re-localisation. Product ubiquity flattened price differentiation — same item, anywhere, separated only by delivery cost. If you’re an Irish or EU operator watching UK demand soften (IMRG has fashion conversion and demand generation sliding), that’s your opening, not your obituary.

“Cross-border didn’t die. It got bought, twice, and taken off the agenda.”

6 / SOLD! — Events Hyve — owner of Shoptalk, HLTH, Bett, Manifest — swapped PE owners at a reported $1.8bn, roughly triple its 2023 price, on $100m-plus EBITDA. The buyer’s line is that in an AI world the in-person room gets more valuable. Maybe. But PE-to-PE at the top of a cycle is less a vote of confidence than a clock running out.

“The conference where you learn the industry is consolidating just got consolidated. Buy the t-shirt.”

THE SIGNAL — Same map, different scissors Here’s the turn, because the off-the-cuff version blurs it. It’s tempting to wrap all six into one doom-loop — but the numbers won’t let you. Contentful is the genuine haircut. Hyve and Passport are up deals, capital flowing toward control rather than creation. Not a market dying. A market being bought.

So where does that leave the operator on the ground? Watching who files first, who gets the quiet no-price deal next, and whether the back half of the year bites as hard as the signals suggest. The winners won’t be chasing scale. They’ll be building something specific enough that nobody can buy their way around it.

If Wednesday’s Daily said Europe is the prize, this is the sound of the buyers loading up.

That’s your two continents. The Ostrich Report is back properly from next week. Week 1 brought to you by Rithum (powering the future of commerce) and All Things Marketplace — over £370M in revenues handled through Amazon. And in January 2027 we’ll all be in Tralee, at the one event nobody can acquire.

Heads up. The rest of them have theirs in the sand.

The World Cup opened with three red cards yesterday. Infantino didn’t watch it. Ireland watched from the sofa. As always.

This week’s V Spot News is six minutes of the retail and ecommerce stories that actually matter:

🥊 Mike Ashley just bid €2bn for Hugo Boss. He’s been building to this since 2020.

💊 GLP-1 drugs wiped £780m from UK grocery spend. The crisp aisle is not okay.

👗 ASOS posted a 50% profit surge and raided Amazon to hire the person who might seal the comeback.

🛍️ Amazon can now design custom merch in eleven seconds. Print-on-demand operators should be nervous.

🏠 Bed Bath and Beyond is buying renovation companies. It’s either genius or the most ambitious pivot since Sears.

⚽ Adidas built an entire F1 ecommerce store in eight weeks using AI agents. No new headcount.

New episode is live now on YouTube. Link in the comments.

Six minutes. No fluff. No panels of people agreeing with each other.

Jerry Maguire. One Bad Pizza and then… Fewer clients. More attention. Actual relationships instead of volume. He prints it, distributes it to the whole company, gets a standing ovation, and is fired within the week.

“I’m finished, I’m fucked. Twenty-four hours ago, man, I was hot! Now... I’m a cautionary tale. You see this jacket I’m wearing, you like it? Because I don’t really need it. Because I’m cloaked in failure! “

I’ve been writing my version of that memo in my head for about six months. This is it. I’d appreciate it if nobody fires me, mostly because I’m self-employed and the meeting would be awkward. Who’s comin’g with me? Dorothy Boyd, thank you.

Here’s the thing I think and have not posted: LinkedIn has stopped delivering my work to the people who asked for it. And this is not new news

Not in a dramatic, shake-your-fist-at-the-sky way. In a quiet, structural way. You followed me. You subscribed. You opted in. And somewhere between my keyboard and your feed, an algorithm decided you’d rather see a carousel about morning routines.

I can’t compete with what LinkedIn wants to promote. More importantly, I don’t want to. Show me a video of someone crying in a Cybertruck and I’ll show you a man closing the app.

I’ve made a huge mistake (almost)

The Gob Bluth move here would be to keep doing the same thing and expect the feed to change its mind. To keep feeding my best work into a machine that treats it like an illusion sorry, a trick. Tricks are what platforms do for engagement.

The actual huge mistake would be staying fully dependent on rented land. I’ve spent fifteen years telling ecommerce brands exactly this: if your entire business lives on someone else’s platform, you don’t have a business, you have a tenancy. Marketplaces, social commerce, the lot the rent always goes up, and it’s never paid in money first. It’s paid in reach.

I wrote in my own planning documents months ago that soul is what makes people stay subscribed when the algorithm stops surfacing you. The algorithm has now stopped surfacing me. Time to find out if I meant it. I say a lot of things.

The Richard Hendricks bit

If you’ve followed my ALREADY FILMED series, you know I believe Silicon Valley wasn’t a comedy, it was a documentary delivered early. Richard Hendricks spent six seasons trying to build an internet that nobody owned where the network belonged to the people on it, not the company sitting in the middle taking a cut of every interaction.

He was right. He was also a disaster of a human being, which is the part of the comparison I’m choosing to ignore.

The closest thing a one-man media operation has to Richard’s new internet is embarrassingly old technology: an email list and a subscribe button. No middleman deciding whether you see the thing you signed up for. You subscribe, it arrives. Revolutionary stuff, circa 1998.

So here’s what’s moving where — and what’s actually on

Video is going to YouTube. Spotify and Apple will carry the audio for the commuters and the dog-walkers. And this isn’t a vague “more content coming soon” promise — here’s the actual slate:

The V Spot News continues — your ecommerce and retail news, seen through the eyes of a madman, as the contract requires.

The Sunday Supplement is new — because the week’s news deserves a slower read, ideally with a proper breakfast.

The Struggle Bus Season 3 gets its finish — the conversations that actually matter, brought home properly.

The Ostrich Report starts next week — the long-form editorial, head firmly out of the sand, with theostrichreport.eu as its home base.

Subscribing on YouTube does one beautiful, old-fashioned thing: it adds that little bell. You ring it, the work arrives. No algorithm sitting between us deciding whether you’ve earned it. The bell is the whole philosophy of this move, shrunk down to one button.

Writing is continuing on Substack. That’s where the stream of consciousness lives, the deep reads, the daily digest, the essays, the meanderings. And speaking of meanderings, here’s what’s coming down that road: more Parks and Recreation in the mix (every ecommerce platform has a Ron Swanson and a Tom Haverford, and I intend to identify them), Larry David coming to the fore (because nobody has ever audited an industry’s social contracts more thoroughly), Flight of the Conchords joining the rotation (business time, indeed), and July is Silicon Valley month a full thirty-one days of the show that predicted everything, lined up against the reality that caught up with it. ALREADY FILMED goes feature-length.

Behind all of it sits the supporting cast: The Struggle Bus for the conversations, and camptralee.com because in January 2027 I’m dragging this whole industry to Kerry and feeding it properly.

LinkedIn? I’m not leaving. I’ll always post here this is where the industry talks to itself, and I enjoy the eavesdropping too much. But the best of the work the full versions, the things I actually stayed up making will live where I can guarantee delivery. What you’ll see here is the trailer. The film is showing elsewhere.

Maybe so, sir. But not today.

There’s a scene in Top Gun: Maverick where an admiral tells Maverick his kind is headed for extinction. Pilots are obsolete. The future is drones. Maverick takes it on the chin and says: “Maybe so, sir. But not today.”

People have been declaring independent media dead for a decade. Email is dead. Newsletters are dead. Long-form is dead. Nobody watches anything over ninety seconds. Maybe so.

But not today.

The bit where I mean it

Before the ask, the thank you because it’s owed. It is because of you I still cannot describe to my parents what I do. A transponder is what I was called recently.

To everyone who has come along this far: thank you. Genuinely. You have been more generous than the work deserved on its worst weeks and more generous than I expected on its best ones. You’ve shared things without being asked, commented when the algorithm gave you no reason to, turned up to webinars at awkward hours, and sent messages that arrived on exactly the days they were needed. You didn’t have to do any of it. You did it anyway.

I am far from perfect. I don’t have all the answers. What I have is opinions, a cross-Atlantic vantage point — Tralee to New York and back, watching the same industry behave completely differently on each shore — and a genuine intention to entertain you while we figure this out together.

So please: keep supporting the madness of this idiot. It turns out the madness works better with company.

If that’s worth anything to you, do one simple thing:

🎬 Subscribe on YouTube and ring the bell. The bell is everything.

✍️ Subscribe on Substack — for the stream of consciousness, unfiltered.

That’s it. That’s the ask. No funnel, no lead magnet, no “free masterclass.” Just: come where the messages actually get through.

Jerry Maguire ends his memo with a question, so I’ll end with his:

Who’s coming with me?

(You don’t even have to bring the goldfish.)

Vinny O’Brien writes about ecommerce, retail, and AI from Tralee, Co. Kerry — when he’s not in New York pretending the time difference doesn’t hurt. Find the full operation at The V Spot on Substack, @thevspotnews on YouTube, theostrichreport.eu, The Struggle Bus, and camptralee.com.

I’ve never knew Monocle. Or its guide to living. I saw it recently at a hotel and read it in Soho. I immediately sent a message to Philip Jackson, thinking I had uneartned some unknown treasure to him. As sure as his quiff is tall, he already knew. He pointed me to their London store and I will visit this summer (if still there).

I was hooked though. Not because I agree with everything they publish. Sometimes it feels like the magazine equivalent of a man who owns three cashmere scarves and believes every problem can be solved by relocating to Copenhagen. Hygge much?

But the whole thing pointed me to something greater, something more. The Monocle Guide to Better Living. The book is ostensibly about cities. Or culture. Or architecture. Or transport. Or radio stations. Or corner shops.

My head went to thinking that although it is about cities,it isn’t really about any of those things. To me is was about stewardship. I am in that place in life. Today a local school headmaster was laid to rest. His legacy is what was written on the page of his death notice. A man who positively impacted 2 generations of kids and their parents too. All enriched through kindness, thoughtfulness and a sense of time for people. A life worth living, a life worth celebrating. The page in question is here. Those who knew him wept today - love does not always earn grief. But to know was to love it appears. rip

What are the people, habits, rituals, imperfections and institutions required to make somewhere worth living?

As I moved from chapter to chapter I found myself increasingly uncomfortable.

What does it tell us about commerce. About the state of it and how we plot it forward.

Monocle celebrates street sweepers.

Morning radio hosts.

Corner-shop owners.

Property developers who actually care.

Independent booksellers.

Local rebels.

Mechanics.

People whose contribution to society would never appear on a quarterly earnings call.

People who create value without necessarily creating efficiency.

And somewhere between the chapter on neighbourhood garages and another celebrating public radio, I started thinking about ecommerce.

For twenty years we have been asking the wrong question.

The question wasn’t:

“How do we make commerce better?”

The question became:

“How do we make commerce faster?”

Those two things are not the same. Convenience became linked to accounting periods and time periods in general. Where speed counts is for the small few big guys. We have developed a poor relationship with time - patience used to be a virtue, now it is a perceived industry ill.

For most of the last two decades, the industry has operated under an almost religious belief that removing friction was inherently good.

Every conference presentation.

Every software pitch.

Every agency proposal.

Every venture capital deck.

Remove clicks.

Remove steps.

Remove waiting.

Remove uncertainty.

Remove conversation.

Remove people.

Remove effort.

Remove friction.

The result is extraordinary.

Consumers can discover, compare, purchase and return products from anywhere on Earth in minutes.

Goods appear on doorsteps faster than pizzas once did.

Payment information follows us from device to device.

Customer service never sleeps.

Recommendations arrive before needs are consciously formed.

It is, by almost every traditional measure, a miracle.

And yet.

Conversion rates haven’t moved much.

Customer acquisition costs continue to rise.

Brand loyalty feels increasingly fragile.

Most ecommerce stores look eerily similar.

Most customer experiences feel increasingly forgettable.

And perhaps most interestingly of all, nobody seems particularly happier.

The scoreboard is quietly asking a question that few people in commerce seem willing to answer.

What if friction wasn’t the problem?

What if we removed the wrong friction?

Kevin Kelly wrote in The Inevitable that successful technologies eventually disappear.

Electricity disappears.

Running water disappears.

Search disappears.

When technology works perfectly it stops being noticed.

Its greatest success is becoming invisible.

I often think about a story Kelly tells involving Larry Page.

Years before artificial intelligence became the dominant story of our age, Page remarked that Google wasn’t really building a search engine. Search engines became boring because we became lazy - we recmoved friction. The idea that any product can be available anywhere in the world means the only place differentiation goes is to price. Making local used to be a reason to discover a brand in your home town or to visit them when away to see what new cool things you could discover and piss off your neighbours. Now, we do the same holidays, watch the same reels, get the same priorities wrong. All this was building and training the basis for AI.

I often think about a story Kelly tells involving Larry Page.Then I went back to it yesterday or Monday, not sure what day it is really.

Most people laughed. At me, not Kelly. My articulation is not worldly in its composition, again, I am fine with this.

Today that prediction feels less remarkable than inevitable. (written in 2016 remember).

But hidden inside Kelly’s observation is another idea.

Invisible technologies may be successful.

They are rarely loved.

Nobody wakes up excited about plumbing. Except maybe Kelly Goetsch.

Nobody posts emotional tributes to broadband. (I once went to the opening of a 1gb fibre optic line, true story.)

Nobody gathers around a dinner table to discuss the elegance of municipal wastewater systems.

They matter enormously.

But they inspire almost nothing.

Commerce is becoming plumbing.

And that should probably concern us.

Because people don’t form emotional relationships with plumbing.

They form emotional relationships with places.

Which brings us back to the corner shop.

The tragedy of modern commerce isn’t that technology replaced the corner-shop owner.

The tragedy is that we spent twenty years trying to scale away the very thing people loved most about them.

The corner-shop owner wasn’t efficient.

He was often slow.

Opinionated.

Inconsistent.

Expensive.

Sometimes annoying.

Occasionally wrong.

And yet he knew your name.

He knew your children.

He knew your habits.

He knew what you usually bought on a Friday evening and what you forgot every Monday morning.

He wasn’t a recommendation engine.

He was something much more powerful.

He recognised you.

Dave Finnegan once made an observation that has stayed with me for years.

Customers don’t want to be tracked.

They want to be recognised.

Those are entirely different experiences.

One feels helpful.

The other feels slightly unsettling.

Modern ecommerce has become remarkably good at tracking.

Recognition remains elusive.

The irony is that we continue trying to solve this problem with more technology.

More data.

More automation.

More AI.

More prediction.

As though intimacy can somehow emerge from scale.

Maybe it can.

Eventually.

But right now most personalisation resembles being followed around a shopping centre by an enthusiastic Labrador carrying a clipboard. Vanity over transparency.

Persistent.

Well intentioned.

Completely unaware of personal boundaries.

And perhaps that explains why so much commerce feels increasingly flat.

Open ten ecommerce sites.

Then open another ten.

The differences become surprisingly difficult to spot.

The same navigation.

The same reviews.

The same subscription prompts.

The same recommendation carousels.

The same countdown timers.

The same loyalty schemes.

The same abandoned-cart emails.

The same chatbot waiting in the corner.

Different brands.

Identical experiences.

Commerce has become an airport terminal.

Exceptionally efficient.

Entirely predictable.

Almost impossible to remember.

Which is unfortunate because human beings are not optimisation machines.

We’re story machines.

We remember surprises.

We remember mistakes.

We remember conversations.

We remember discoveries.

We remember things that weren’t supposed to happen.

Daniel Kahneman spent decades showing us that people rarely make decisions rationally.

Yet much of ecommerce continues operating as though consumers are spreadsheets with credit cards.

Optimise the route.

Reduce the clicks.

Remove the effort.

Increase the probability.

And perhaps that works.

Until it doesn’t.

Because eventually optimisation begins eliminating the very moments people remember.

Discovery is one of those moments.

Perhaps the most important one.

Today the industry is understandably excited about agentic commerce.

AI agents that know our preferences.

Our budgets.

Our sizes.

Our habits.

Our delivery addresses.

Our dietary restrictions.

Our favourite colours.

The promise is seductive.

Perfect recommendations.

Perfect efficiency.

Perfect relevance.

But I can’t help wondering what happens when every recommendation is correct.

What happens when every purchase is optimal?

What happens when surprise disappears?

What happens when wandering disappears?

What happens when curiosity disappears?

Because some of the best purchases any of us have ever made were terrible recommendations.

They were accidents.

Detours.

Mistakes.

Moments of curiosity.

A book picked up in an airport.

A record discovered in a strange shop.

A bottle of wine recommended by someone you trusted.

A jacket purchased because it reminded you of somebody.

A city visited because your flight was delayed.

The stories matter more than the efficiency.

Perhaps that’s what Monocle understands.

And perhaps that’s what commerce forgot.

Monocle never celebrates efficiency for its own sake.

It celebrates the ingredients required to make somewhere worth returning to.

The radio host.

The bookseller.

The mechanic.

The rebel.

The street sweeper.

The people who don’t scale.

The people whose value exists precisely because they don’t scale.

And maybe that’s where commerce finds itself now.

Not at the beginning of another technological revolution.

But at the beginning of a human one.

Not deciding how to remove the next layer of friction.

But deciding which friction deserves to remain.

“When this emerging AI arrives, its very ubiquity will hide it” -

Part of our challenge with Agentic. Protocols have set out the stall for where these new world orders will eventually reside, out of sight, out of mind, but clearly in situ, in mind. Kevin Kelly wrote the inevitable in 2016. I am re reading it, again. Trying to help me untangle what is unique to my world of work or what is linked to the teenage years of my children. In 2002, he spoke with Larry Page (Google cofounder) at a party “Larry, I still don’t get it. There are so many search companies. Web search, for free? Where does that get you?” Page replied - “Oh, we’re really making an AI.” Describing the training and ending this page Kelly predicted:

By 2026 Google’s main product will not be search but AI. (written 10 years ago). He is an optimist for technology which is why I feel compelled to go back here. It is comforting, I think.

3 recent breakthroughs in have unleashed a 60 year conversation of AI is just around the corner - 1. Cheap Parallel Computation 2. Big Data and 3. Better Algos. This convergence putting ChatGPT and the likes into all of our hands compounds the learning. In 2015 researchers at Deepmind published a paper in Nature, describing how the taught an AI to learn to play a 1980s era arcade video game. Video Pinball. They did not teach it how to play, but how to learn to play. A profound difference. At first the AI plays nearly randomly, but it gradually improves. After half an hour, it misses only once every four times. By its 300th game, an hour into tit, it never misses. It keeps learning so fast that in the second hour it figures out a loophole in the Breakout game that none of the millions of previous human platters had discovered. This hack allowed it to win by tunneling around a wall in a way that even the games creators had never imagined.

This type of learning is happening today, at scale. And with speed. Again, he write in 2016, the business plans of the next 10,000 start ups are easy to forecast, Take X add AI. He then goes on to give a timeframe that now seems slow for the development of new minds. AGI it is called today and it is interesting where it lands.

He provides other directional indicators for the phase we are in now - one we are seemingly keen to speed up and slow down in equal measure. Agentic is in the phase in calls filtering. I am saying this, it is not fact. Ponder this - “Life is too short, there are too many books to read. Someone of something has to chose, whisper in our ear to help us. We need a way to triage. Our only choice is to get assistance in making choices.” Filters that exist today - we filter by curator - every retailer is a curator. We filter by brand - brands filter through the clutter. We filter by ourselves - choices and preferences, though it is rare. Amazon and its recommendation engine, ChatGPT the same are now the Chief Discovery mechanisms. They are more reliable because they use more signals. In reading this and researching it also got me thinking to how the signals arrive in the world of agentic. They don’t play out the same - Objectives are reached through shared plumbing data. Not through old world nuance like a cookie. It renders attribution dead in a fully agentic world. It asks questions of retail media and how it presents and stores data. Read that again, in Agentic commerce there is no need for the cookie. Is there a need for attribution now, bearing in mind you have to be everywhere the customer is whenever they want you all of the time. Attribution it seems is a problem only for those who cannot afford to be everywhere. And it is a bad at that.

But filtering leads to optimising - optimising, Steve Bartlett loves the term, is the the presented goal. My personal avatar, over time learns more and more and becomes a new manifestation of things I want and need or things I didnt know i need but now want or now want because the need is apparent - my health device let my credit card know that some aspirin might save my life on an upcoming flight. Extreme example maybe, but better than your fridge talking to your toilet and learning you are at that life stage that soy milk may be better for you than full fat. Underpinning much of this is privacy and this is where I feel there is a battle to be had and where the role of government and personal choice, education all play a part. This the day after the most Skynet partnership imaginable was announced, Apple AI will be powered by Gemini. So all of us with that power and that knowledge in our pockets will succumb to this new world - if we elect to. I am flipping forward with gusto to the chapter titled “tracking”.

“We tend to be uncomfortable being tracked today because we don’’t know much about who is watching us. We don’t know what they know. We have no say in how the information is used. They are not accountable.The relationship is unbalanced and asymmetrical. “ Agentic commerce to realise its potential is a world beyond commerce and retail. It is the assigning of a new social contract. I like the expression because it suggests we have some say rather than passively agreeing to what we end up with. (as I said, I am trying to be more optimistic). Personalization is being handed over if we are to be believe that agents can have a role in our shopping habits. An agreed route to transparency will be required. And my fear is we already handed this over, but i think it is where innovation and ecommerce goes beyond this year. Proactive management of our own privacy allows us to redraw the lines of how we want to behave in this bold new world, that we accept will happen. How is the only question. When is the piece of string, it is constant, without reprieve. “Greater personalisation requires greater transparency. Absolute personalisation (vanity) requires absolute transparency (no privacy).” Mull that over.

To visualise it picture these choices pinned on a slider bar -to the left is personal/transparent and on the right is private/generic. What has happened over time with technology is the opposite of what we claim to be true. We veer more towards personalized sharing - proven through social media usage. “The human impulse to share overwhelms the human impulse for privacy.” Vanity trumps privacy. This all suggests at some point Shopify have something to think about - user data and retailer numbers create a critical mass. With Agentic now converting at higher rate (their words, not mine) they need to figure out what Shop is or do we already know.

Privacy is one of those areas that we (consumers) will either take on as a proactive place to be in control or passively turn it over to our platform overlords (god I hate myself for using this expression). But we are talking now about our kids and their future - Finland doing a deal with Anthropic for all students to have a full account on the platform for the purposes of further education and learning. The UK on another side creating its first fully sovereign AI frontier model. The coalition includes some of Britain’s most prominent organisations, with Babcock International Group, BT, Lloyds Banking Group, LSEG, NatWest Group, PwC, Thales UK, Leonardo UK, BAE Systems and Telefónica Tech UK&I among those signing memoranda of understanding to participate in the model’s design phase.

The full list of coalition partners also includes Era4, Haleon and The Alan Turing Institute. Lumen Sovereign will be trained entirely on Isambard-AI, one of Europe’s most powerful supercomputers, using compute allocated through the UK government’s £500m Sovereign AI programme, with deployment readiness targeted by the end of 2026.

They won’t be the last.

It leads me to an area under explored but ripe for consideration - remixing is a chapter where Kelly turns to a theory of reusing existing resource to make new things, create new wealth. Paul Romer, an economist at NYU says that real sustainable development comes from re arranging existing resources. Layer on our filtering idea. “What if advertising followed the same trend of decentralization as other commercial sectors have? What is customers, created place and paid for ads?” Or from my part, sold their data back to the grid. When I mentioned this before, it was argued that we get this great resource for free - we don’t we paid for it, without ever being asked for it. We had industries spawned from open use and exploitation of our data. The Mad Men of Madison avenue would have loved this kind of access. And for free.

With all of the protocols developed, who will take on the development of the customer protocol? The protocol that occasionally tells an agent, go f*ck yourself. I am not having it today, my mood is not there. Maybe Apple and Google have done this in the existing small print, maybe our data is not our own and we are deluded. Or maybe the time is now for us to consider the most important protocol of all - the customer one. Over to your boffins.

That’s the conversation I’m hosting on Thursday 11 June, 11:00 BST. It’s free, it’s forty minutes, and there is no slide about synergy. I checked.

Reserve your spot →

EVERY BOARD WANTS AN AI STRATEGY. ALMOST NONE OF THEM CAN ANSWER THE QUIETER QUESTION.

Here’s the quieter question. Before “what’s our AI strategy,” there’s “is our data actually in a state to support one.” For most B2B businesses the honest answer is not yet, and until the orchestration is solved, an AI strategy is just a slide with a nice gradient on it.

We’ve been threading this one for months. The agentic future keeps getting announced like a product launch when it’s really a weather system, something Kelly would call inevitable and most operators would call Tuesday. The capability is racing ahead. The plumbing is not. That gap is the whole story.

THE NUMBERS ARE WORSE THAN THE HYPE LETS ON.

This isn’t vibes. It’s YouGov fieldwork across UK B2B decision-makers, gathered for The Inevitable Shift guide with Glass Atlas, PayPal and Commerce. Three numbers do the heavy lifting:

44% aren’t planning to invest in their stack in the next twelve months. 42% are operating today with no ecommerce stack at all. And 52% describe themselves as “hybrid,” which is the polite industry word for neither fully manual nor fully modern, the digital equivalent of still having a fax machine but being embarrassed about it.

If you’re reading this from across the water, here’s the Atlantic flip. America is busy livestreaming its agentic awards ceremony. UK and Irish B2B is quietly running a multi-million-pound operation on email, gut feel and a spreadsheet named FINAL_v7. Both things are true at once. Only one of them is on stage at the conferences.

FORTY MINUTES. FIVE BEATS. NO VENDOR THEATRE.

Here’s the actual shape of it:

The hook, smart AI, dumb data, with a real-world read from inside Sweet Squared rather than a hypothetical. Then the three orchestration breakdowns that silently kill every AI initiative: product data (the SKUs that don’t match), customer data (the CRM and the storefront swearing the same person is two people), and transactional data (what the ERP claims versus what the order is actually doing). Then the intelligence layer itself, what a composable, ROI-led architecture looks like when it’s working. Then what agentic actually does once the foundations are right, the useful stuff like context-aware quoting and margin-aware pricing, not the theatre. And finally a data-readiness self-assessment you can run the same afternoon you watch it.

You submit a question when you register and we fold the best ones into the discussion. So it’s less a webinar, more a room you get to talk back to.

OPERATORS, NOT PANELISTS.

This is the part I care about most. Nobody on this is here to read a brochure at you.

Martin Rogers runs IT at Sweet Squared, the beauty distributor that supplies professionals across the UK and Ireland, so he’s the merchant who actually lives inside the data mess we’re describing. Pete Robertshaw of Glass Atlas spends his days bridging IT, ecommerce and ops, which is exactly where these breakdowns happen. Lance Owide leads B2B at Commerce and has been a straight-talking voice on what genuinely moves the needle versus what just demos well. And I’ll be reading the data line by line so you don’t have to.

A merchant, a solutions director, a platform VP, and an Irishman who is immune to the hype. That’s the room.

FORTY MINUTES THAT’LL SAVE YOU A QUARTER.

Free to attend. The replay and the data-readiness self-assessment go to every single person who registers, so even if 11:00 BST is mid-meeting for you, sign up anyway and watch it with a coffee later.

This is the follow-up to the conversations we started at B2B eCommerce World London and to The Inevitable Shift guide itself. If you’ve been reading along as the agentic-commerce story has built week over week, this is where the vision stops being a keynote and becomes a Wednesday-afternoon to-do list.

Register free, takes 30 seconds →

Thursday 11 June. 11:00 BST, 12:00 CEST. Forty minutes. The calendar hold lands the second you sign up.

Bring a question. Bring your worst spreadsheet. See you in the room.

Read more

Read more

One weekly injection of artificial intelligence, administered directly into the org chart. AGENTIC suppresses the executive appetite for understanding your own business and replaces it with the serene confiden…

Read more

Thompson Street. Three stories that caught my eye this week.

Twelfth floor on Thompson, Tuesday morning. Outside the window, the cast-iron facades are doing exactly what they’ve been doing since the 1880s, looking very expensive while pretending to be industrial. Some stories I noticed.

eBay tells Gamestop: Sorry luv, you’re just not my type

eBay’s board, chaired by Paul Pressler, formally rejected Ryan Cohen’s $56 billion takeover bid on May 12, calling it “neither credible nor attractive.” The board cited financing uncertainty, operational risks, GameStop’s governance, executive incentives, and the potential drag on eBay’s long-term growth. In fairness, “neither credible nor attractive” could also be the opening line of every Tinder rejection ever written, which makes this rare among M&A press releases, it has a punchline built in.

The TD Securities $20 billion financing commitment that propped up Cohen’s bid was contingent on the combined company maintaining an investment-grade rating from at least two of the top three agencies. Moody’s had already called the proposed deal credit-negative. The numbers, as the kids say, weren’t mathing.

As we covered a fortnight ago in The Teenager Theory, the value of Cohen’s bid was never going to be the deal itself. It was the mirror. eBay had to look at itself and ask whether all the right strategy, Depop, Goldin, PSA, Authenticity Guarantee, TCGPlayer, was actually being executed with the urgency the strategy deserves. The bid did its job. The mirror got held up. The teenager looked.

The question for next quarter isn’t whether eBay rejected the offer. It’s whether the teenager goes back to ironing its shirt at the same pace as before, or whether something quietly changes in the strategy meetings on Hamilton Avenue. I have my suspicions. Watch the next earnings call for the pace, not the numbers.

One small win for the marketplace class. Possibly.

Read more

Read more

This is Vinny O’Brien. And I want to talk to you about two words.

Read more

Read more

There is a number in the Mapp Fashion Intelligence Report that should be in your head more than it does.

Next’s Google exit rate: 84%.

Five out of every six visitors who arrive at Next via Google leave without buying. In the Mapp dataset, 397 retailers, 7 markets, 53.7 billion visits, that is the highest exit rate in the sector. River Island at 79%. Primark at 79%. Next at 84%. The report I wrote last month cited this as one of the “clearest signals of site-side discovery failure.” I stand by that characterisation. The discovery model that built UK online retail is breaking, and these numbers are part of the evidence.

Next reported Q1 full-price sales up 6.2% this morning, beating its own 4.0% forecast by £28 million. Online grew 10.1% in the UK. The LABEL platform, third-party brands hosted on Next’s infrastructure, grew 15.7%. Full-year pre-tax profit guidance raised to £1,218 million. Shares up.

So: highest Google exit rate in UK fashion. Best results. Profit guidance above £1.2 billion. How do those two things exist simultaneously?

The answer to that question is probably the most important thing happening in UK retail right now, and almost nobody is saying it clearly.

The rabbit hold revisited

The context matters. In The Rabbit Hole Got Deeper, I laid out the structural case for why UK online retail is in genuine difficulty. The IMRG data covering March 2021 to early 2026: conversion baseline drifting relentlessly downward, from around 4% to sub-1.6%, with March 2026 session conversion rates down year-on-year from 1.72% to 1.59%. CPA rising 22.53% in the same period.

The Mapp Fashion Report, 397 retailers, Sarah McVittie’s work, the most rigorous dataset available on UK fashion discovery economics, delivered the verdict underneath the chart. Paid search nearly tripled its share. Organic contracted 4.1 percentage points. The industry looked at a declining organic position and instead of asking why, reached for the credit card. Paid search added 3.2 billion visits, a 264% increase, while organic barely moved.

The financial consequence is the danger zone: brands running paid at 4-8% of their traffic mix produce average operating margins of 3.7%, improving in only 48% of cases. They are paying full auction prices for discovery they used to earn. The maths doesn’t work.

And in that same dataset, Next’s Google exit rate is 84%.

Huh?

Here’s why the 84% exit rate and the £1.2 billion profit are not contradictory. They are, to me, the same story.

Next’s Google exit rate being the highest in the dataset doesn’t mean Next has the worst site. It means Google is the wrong entry point for Next’s customer. A Next customer who arrives via Google is typically arriving with a specific, narrow intent, a particular search term, a competitor comparison, a price check. They didn’t start at Google because they wanted to browse Next. They started at Google because they were shopping the category. For that visitor, Next’s site is not where they want to end up. They bounce.

The Next customer who converts doesn’t arrive via Google. They arrive direct. They arrive via the app. They arrive because they have a Next account and a relationship with the Directory. They arrive because a friend mentioned it. They arrive because they just got a Next Finance statement and clicked through. They have, in the language of the Mapp report, brand gravity, and brand gravity doesn’t show up in the Google exit rate. It shows up in the direct traffic, the repeat purchase rate, the LABEL platform numbers.

This is the point about owned audience I made in the Rabbit Hole piece, and it is perfectly illustrated by the Next paradox. The brands that will be fine are the ones that built genuine direct relationships before the market changed. Next is the case study.

The 84% Google exit rate is a symptom of a brand that has successfully moved most of its customer relationship to channels Google cannot see.

Read more

Read more

We tried to sit down with Ryan Cohen, founder of Chewy, chairman of GameStop, and the man currently attempting to buy a company four times his size. to discuss ambition, audacity, and what exactly he said to his bankers. He said no, we were bonkers. This is what we think would have happened.

Let’s start at the …

Read more

Read more