How’s your AI imposter syndrome doing? From unknown, to Hype Cycle to deep reality and the overarching feeling of Anxiety, in all its forms. From kids, to job safety to now, learning and implementing. So. Many. Emotions. It’s like being a teenager all over again. Outwardly projecting confidence whilst crying into your pillow every night. Technology introduction to those years, for my generation, was slower, more expensive. A privilege, not an expectation.
While all this was happening, we had music and culture to keep us going. Disconnecting from the always on, always available reality of today. Experiences were visceral and lyrics and sounds bring us back to moments in time. So as I guide you through my AI, MCP, ACP, AEO, GEO ecommerce journey, I’m taking a trip on a memory bliss. Soundtracking todays moments through lyrics of yesteryear to maybe ease the anxiety. I am not sure if this is for you, or for me. Entangled in this was my own awkward developmental stages, mishaps and triumphs.
Today’s Brand leaders grappling with AI and tech are forced into it, like it or not. Their answer:
Definitely Maybe — Oasis, August 1994
Slip inside the eye of your mind, you know you might find,
A better place to play.
You say that you’ve never been,
But all the things that you’ve seen,
Slowly fade way. (October 95, before you lynch me)
Thirty-two years ago this month. I was 13 and had completed primary education and building up to my secondary cycle. I spent the previous 12 months wearing an over the head brace tightener, almost identical to Lisa Simpson. It required a minimum of 12 hours of torture but by this stage my overbite was so bad I could eat an apple through a tennis racquet. Heady times. I had not yet had a girlfriend or a first kiss. Maybe it was the braces?
I was getting good at sports and had a job for the summer, so things were not all bad. Britpop music was emerging and dotted among the sports posters on my wall, were Spice Girls - mostly Baby Spice.
Oasis and Blur were dominating the UK scene and in the US Dave Grohl was a drummer, Rage against the machine were killing in the name of. Meanwhile Sabotage was a summer banger and Hip Hop was in its Golden era, forming its east coast/west coast divide. Warren G decided to Regulate. Biggy was Juicy and the Olympics Ambassador we know as Snoop Dogg today was drinkin’ Gin n Juice. Laidback.
Lyrics have always been a gateway to something maybe we could not articulate ourselves. A love message, angst or fear. The exact posture of every brand and retailer I speak to about artificial intelligence. The certainty and the vagueness arrive in the same sentence, usually the same breath. We are definitely doing AI. We are maybe piloting something in customer service. We are definitely going to be data-ready. We are maybe going to find out what that means.
It is not stupidity. It is me in 1994. Knowing what I want but equally knowing what I am now. It is the sound of an entire category being sure in public and unsure in private, and understanding at some level that the sureness is the product.
During those hedonistic years, brands still had the luxury of time and the luxury of a channel. Dominance meant owning a vehicle. MTV was a vehicle. A back page was a vehicle. Culture was something a brand could sit inside and shape, at a pace that allowed for a meeting on Tuesday and a decision by Friday. Music was the barometer and the barometer moved weekend to weekend, which felt impossibly fast at the time and now looks like a leisurely stroll through a shopping centre.
Things moved from weekend to weekend. Now they move from earnings call to earnings call, and the earnings calls are three days apart.
2026, the new 2007
This was the year the news cycle turned into dog years. Wars, straits, supply chain, oil price, cost of living and of course nothing trumps Trump. News volumes beat sanity. And in our wisdom, he market has decided that SaaS is dead. It is not dead. It is changing, quite significantly, and the distinction between those two statements is worth about a hundred billion dollars of misallocated capital. But the decision has been made in the way markets make decisions, which is loudly, in unison, and slightly too late to be useful. It has made it difficult for brands. Easier, in many ways for SaaS builders.
The consequence is that every software company in the category is now being told by two constituencies at once that it must have AI embedded. The first constituency is investors, for whom AI embedding is a proxy for survival. The second is the conference circuit, which has spent two solid years staging the same panel with different lanyards. You could attend every major retail technology event since early 2024 and construct a near-perfect timeline of the phrase “AI-native” migrating from the keynote to the breakout room to the trade stand to the apologetic footnote. Admittedly, it is exhausting to watch. I cannot imagine living it. The merry go round for employment continues to be one I do not want to be on.
I saw the sign
And it opened up my eyes, I saw the sign
Life is demanding without understanding
You won’t thank me for that earworm. That pressure produced velocity. It did not produce judgment, and it did not produce a pause for thought. The daily news cycle is in our hands, in our pockets, dominating our thinking and clouding our judgment, and the companies feeling it worst are the ones whose valuation depends on appearing to have already understood something nobody has understood yet.
I have been close to one of these companies recently. They are all Icarus but how else could they be right now. So here is the thing I have come to think, and it took me an embarrassingly long time to get to it.
SaaS was forced to show AI. Brands are forced to justify it.
Those are entirely different accountabilities and the industry has misread the second as slowness. When people hear you’re workin’ in AI in 2026 -
I’ve known a few guys who thought they were pretty smart
But you’ve got being right down to an art
You think you’re a genius, you drive me up the wall
You’re a regular original, a know-it-all
Oh-oh, you think you’re special
Oh-oh, you think you’re something else
Okay, so you’re a rocket scientist
A software company that ships a copilot in three weeks has satisfied a demand for a demonstration. That is not nothing. It is genuinely hard engineering under genuine time pressure, and some of it is very good. But the accountability was performative by design: the thing had to be seen.
A brand that spends six months asking who will use this, what it replaces, what breaks when it fails, and who owns the answer when a customer gets a wrong one, is not lagging that software company. The question is open for them all - they are waiting for the first fallers, the first winners. How do we avoid that and repeat this. I listen ALOT, more than you would think and i read enough.
Kelly Goetsch has become one of my go to for our industry. A very smart guy with an open attitude to tech. His kind and makes time for people. People skills are just naturally in him, not forced or learned. He has set a standard and followed others. He questions well, provides great ideas but he has made it ok for me to question the way I think about the work. The architecture is where I arrived. Because of Kelly.
Architecture goes beyond data readiness and I will get more into this. But now, we have the permission because we can learn quicker, change quicker and be wrong quicker. Structure and standards set the tone. Architecturally are we ready, culturally are we ready. Those two areas, for brands, more than for SaaS companies are symbiotic. They are linked to understanding the customer and shaping their experience of your brand proposition.
The failure mode is not slowness. The failure mode is when a brand looks at its inability to justify a chatbot and concludes that it therefore cannot invest in architecture. Those are unrelated questions and conflating them has probably cost the sector two years.
Public Gar and Private Gar
There is a play about this and it was written in 1964 by a man from Omagh. Another little town I know so well.
In Philadelphia, Here I Come!, Brian Friel puts two actors on stage playing one person. Public Gar is the version everyone else can see and talk to. Private Gar says the things Public Gar cannot, and he says them only to us. They share a life, a body and a history, and they never once look at each other.
Every brand doing AI right now is a Friel character. But then, so am I. Maybe you are too.
Public Gar is the press release, the innovation slide, the LinkedIn post about the pilot. Private Gar is the actual data model: the four systems that disagree about what a customer is, the pricing logic that lives in one merchandiser’s head and one spreadsheet nobody has opened since March, the process that only works because Sinead in the warehouse knows to check the second screen.
For thirty years those two could coexist, because nobody outside the building could hear Private Gar.
That has now changed, and the rest of this essay is about what changes with it.
“And all the roads we have to walk are winding
And all the lights that lead us there are blinding
There are many things that I would like to say to you”
Nobody under fifty needs the opening bars explained. That is the thing about a wonderwall: it is a place you are told to look for inside yourself, and the song is quite careful never to specify what you will find there. Wonderwall is where we spend our time creating, the answer starts further downstream. Or upstream, I never know with that expression.
I have found out what you find there. It is a mess.
What MCP actually feels like (my experience)
My own journey into this was not strategic. It was operational, which is the only way I know how to learn anything. It had to be. It does not imply the lack of strategy, but strategy involves diagnosing the problem.
Over the past year I have been connecting language models directly to live business systems. ERP, ecommerce platform, integration layer, all of it, through the connective tissue that has emerged for exactly this purpose. The technical term is Model Context Protocol. The practical description is that you stop pasting screenshots into a chat window and start letting the model ask the system itself.
The first time it works properly, it is genuinely startling. You can coordinate a better answer, because the model can consult three sources instead of one. You can interrogate deeper, because you are no longer limited to the question you thought to ask before you started. And you can formulate a plan that is grounded in your actual systems rather than in a slide about your systems, which is a distinction most consulting has been quietly avoiding for twenty years.
I have watched a defect go from a three-week investigation to a ninety-second query. Not a summary of a defect. The actual thing, with the actual record IDs, traced through the actual flow.
And it is far, far harder than it looks. It produces unnatural levels of anxiety because you arrive at the answert through means other than walking it through. That’s the next part. Validation through experience. As soon it works, it stops being a technology project and becomes a mirror. The mirror is daunting, like most mirrors to me in my 40’s.
Vin Private
2026, a typical news cycle: The channel is compounding quietly while the industry argues about agents. The boring number is doing the work. Also the mobile-in-store detail from the same release: phone use while shopping in a physical store rose from 30% of consumers in January 2024 to 42% in March 2026, and in-store review reading is up 30% since 2024. The shopper is already interrogating you mid-aisle. Your architecture is already being read. Good data alone won’t get you there.
WTAF
I brought the musical musings up to the millennium. The year Y2K. Remember the things that were supposed to happen. From Vanity Fair -
Will the millennium arrive in darkness and chaos as billions of lines of computer code containing two-digit year dates shut down hospitals, banks, police and fire departments, airlines, utilities, the Pentagon, and the White House? The nightmare scenarios are only too possible, ROBERT SAM ANSON discovers as he traces the birth of the Y2K “bug,” the folly, greed, and denial that have muffled two decades of warnings from technology experts, and the ominous results of Y2K tests that lay bare the dimensions of ticking global time bomb.
What is going to happen after the SaaS crash? The compounded human relationships become the place that we stay safe - our panic room if you will. Most people just want to do their job, not compete with the market, the market noise. The SaaS industry sold brands the very tools that fragmented their data, and is now selling them the AI that cannot read it. Every app install was a small act of decentralisation. A brand does not have an architecture. It has an accumulation, assembled from a hundred vendors’ architectures, none of them its own.
And that brings anxiety in large quantities. Is this discussion about my job or is it about the job I have done or my job in the future. All very confusing. I realise that to anyone, being asked by a consultant type about how they do things during an AI deep dive, it means only one thing - the pink slip.
Summer of 2000, I finished school as it was and was destined for college of some sort. Not having the discipline to think ahead, I did enough. I was playing sports at a good level and praying in between. I landed on the leg up programme in Dublin Institute of technology. The Certificate to get into the Degree course. It suited my new found love of the college social lifestyle. £1 per drink on a monday night in Dublin made rest of week something something. I was gearing up after my first year to go to Toronto for a summer. After my repeats I was good to go. My job - A rickshaw runner, running 15km per day with a rickshaw full of people. I learned how to pitch at couples leaving Blue Jay stadium and fending off 25 other callers just like me.
By 2002 Justin Timberlake had crossed a divide and turned up on MTV base, not somewhere he would historically have been found, but with Pharrell Williams as his muse and partner and now, as Robin Thicke would later put it we had Blurred Lines. Culture and subculture were mixing.
And when you’re out there without care
Yeah, I was out of touch
But it wasn’t because I didn’t know enough
I just knew too much, mm
Does that make me crazy?
Does that make me crazy?
Does that make me crazy?
Possibly
Vin Public
What you find when you look inside is a mess, and it is not because anyone was negligent. It is because nobody was ever asked to make it legible to a stranger.
Every system in your business was built to be operated by people who already know how it works. That is not a defect. That is how organisational knowledge functions and it is how it has always functioned. The warehouse team knows that location 21 is not real. Finance knows the report is right if you run it after eleven. Customer service knows which of the two order statuses to believe. None of that is written down anywhere, because it never needed to be, because the only people reading were the people who already knew.
A model does not know. A model is the most sincere reader your business has ever had. It believes your data.
Don’t go chasing waterfalls
Please stick to the rivers and the lakes that you’re used to
I know that you’re gonna have it your way or nothing at all
But I think you’re moving too fast
Which brings me to the phrase you have heard at every event for two years and which almost nobody has defined.
Data readiness.
Pop Quiz: How many of you SaaS companies or platforms have used data readiness in a marketing campaign in the last 12 months?
It is being sold hard right now, and it is being sold narrow. In practice it has come to mean product data: feeds, attributes, titles, images, the PIM, the catalogue. And that is being sold because product data faces outward, which makes it auditable, which makes it invoiceable. There is a service line attached to it. There is a deck.
Product data readiness is an entrée. It is real and it matters and it is nowhere near sufficient.
The fundamental question is architectural, and it has only one honest form: what does a machine need to know about us, and can it get that without a human in the loop?
That question covers your catalogue, yes. It also covers customer identity across four systems that each think they own it. It covers transactional truth, meaning whether anything in your estate can state with confidence that a given order actually shipped. It covers process: who is allowed to change a price, and how many systems find out, and in what order, and what happens when two of them disagree.
An AI that can read your product data but cannot tell whether an order shipped is a demo. It is a very good demo. It is not a capability. And, to be clear, you need this. We all need this. We should not ignore this.
Bad architecture used to be a private embarrassment.
It slowed you down internally. It made Tuesdays worse. It cost you margin in ways that showed up as a vague sense that things took longer than they should. But it was yours, and it was invisible, and the customer never saw it.
Connect a model to your systems and your architecture becomes the model’s understanding of your business. Every duplicate record, every process living in one person’s head, every place where two systems disagree about the truth: all of it becomes an answer, delivered confidently, in a tone of complete authority, to somebody who is not you.
AI does not reveal your data problems. It publishes them.
That is the opportunity and the weakness in the same breath. The gift and the curse.
Vin Private
The hype cycle around AI has done what hype cycles always do - generate fervour, start mini movements, brings investment but it splits and fragments industries. It quietly and rather permanently silenced retail media to be sitting firmly in the corner, right where Baby is not supposed to be sitting. But this AI hype cycle did bring some real change. Investments by governments, the emergence of sovereign AI systems and the creation of new wealth means that it is here to stay.
2007 was a year defined by Thomas Friedman as being the birth of the Super Nova for technology.
Born in 2007
Hulu
Zendesk
Dropbox
Fitbit
Kindle
Netflix
Zinga
Airbnb
The iPhone
Lamine Yamal
AWS was only a year old and Samuel Moores theory was appearing on TedTalks the world over. 2026 will be one of those watershed years. Fable, Opus 5 and whatever else is in our futures emerged and when AGI hits, well, it will have to be seen to be believed. Back to the plot.
Brands have to contest with unknowing how they did things. Only to build how they should. What has been helpful for me to think through were some odd b ut necessary steps. This pile of work makes me feel overwhelmed and underwhelmed in equal measure. Here I am using advanced technologies to bring defects and repairs to minutes and seconds, from days, weeks and months. To rewrite company policies that capture the real state of our present and allow us to plan for a better, more flexible but still more rigid future. Training models and training data Operating in a black box without fully getting why or who cares. But there are generations who surface their needs and wants in very different ways to us. We need to be aware of ourselves and what we need to lose soon.
Architecture is visual, it is learned and it comes from documentation. Lots of documentation. Version control is important and the ability to know what to change and ask why to get there.
Sidenote: I just got an email titled - Your next customer could be an AI agent.
Vin Public
And while the industry has been arguing about agents, the actual channel has been compounding quietly in the background.
The US Census Bureau published its second quarter ecommerce figures yesterday, 18 August. Ecommerce hit 17.1% of total US retail, up from 17.0% in the first quarter and 16.3% a year earlier. Seasonally adjusted, that is $340.2bn, up 3.8% on the quarter and 12.2% year over year, against total retail growth of 6.7%.
Online is growing at roughly twice the rate of retail overall. It has been doing this, unglamorously, through every hype cycle of the last decade.
But the detail in that same release is the one worth sitting with. Phone use while shopping inside a physical store has gone from 30% of consumers in January 2024 to 42% in March 2026. Reading product reviews while standing in a shop is up 30% since 2024. Price comparison is up 18%. Checking which payment methods a merchant accepts is up 45%.
The shopper is already interrogating you mid-aisle, with a device, using systems that read your data and not your marketing.
Your architecture is already being read. It has been for years. The only thing that changes with agents is the fluency of the reader.
If we are to believe everything we read this is our reality. But we know thanks to the Wachowski Brothers (the Matrix 1999) that not everything we believe is to be accepted without questioning. Load the fight scene:
There is no spoon.
The SaaS industry sold brands the very tools that fragmented their data, and is now selling them the AI that cannot read it.
Every app install was a small act of decentralisation. A perfectly sensible one, individually. Reviews app, subscriptions app, loyalty app, returns portal, review syndication, a second review syndication because the first one did not do Germany, an inventory tool, a bolt-on for the inventory tool. Each one solved a real problem in a fortnight and each one took a slice of your operational truth and moved it somewhere you do not control, in a schema you did not design, behind an API you have never read.
Fifteen years of that, compounding. All that compounding has led to unprofitbaility for the masses, a CVR stuck on 3% for the masses, a generation unconvinced and billions of shoppers who could not wait to not shop online after COVID. Yes, our system ain’t broken.
Which is why I keep I have come to learn: a brand does not have an architecture. It has an accumulation. It is assembled from a hundred vendors’ architectures, none of them its own, and the seams between them are held together by exports, spreadsheets, and a person that lef the business last year..
The vendors are not villains here. Each of them built something good. But nobody was ever accountable for the whole, because nobody sells the whole. And how they have pivvoted, adopted and moved on is remarkable. Even Jack Barker would point and clap. Ref from Silicon Valley FWIW. In Silicon Valley, the central conflict of the third season is platform versus box. Richard Hendricks wants to build the platform, the general thing, the substrate other things run on. And the pressure to abandon it and ship a physical appliance instead does not come from Hooli, the corporate antagonist. It comes from Jack Barker, installed as chief executive by Raviga.
By the investors.
You got me beggin’ you for mercy
Why won’t you release me
You got me beggin’ you for mercy
Why won’t you release me
I said release me
The box is never the enemy’s idea. It is the board’s. It is the sellable artefact, the thing with a price and a demo and a shipping date, and it exists because somebody needs a story for a quarterly meeting.
That is precisely what is happening to software companies right now. And it is precisely what is being passed downstream to brands, who are being sold boxes: AI features bolted onto tools they already rent, producing capabilities they cannot reconfigure, swap, or own.
If you buy the box, you have bought a capability. If you build the architecture, you have bought the ability to change your mind.
Another turning point, a fork stuck in the road
Time grabs you by the wrist, directs you where to go
So make the best of this test and don’t ask why
It’s not a question, but a lesson learned in time
It’s something unpredictable, but in the end is right
I hope you had the time of your life
So take the photographs and still frames in your mind
Hang it on a shelf in good health and good time
Tattoos of memories and dead skin on trial
For what it’s worth, it was worth all the while