Week in Tech #8 — Walmart, Google I/O, Meta, and a Y Combinator talk all pointed at the same thing last week. None of it is about how much AI you can afford.
A lot of retailers reported earnings last week, and the headlines mostly led with sales. I want to point at a different part of the same reports, because it changes what a small business probably ought to be doing on Monday morning. I sell rugs through some of these companies — Walmart, Target, Wayfair, the off-price world — so I read the quarters partly to understand who I’m working with and where the floor is shifting. And last week, the floor moved.
Once I held the retail numbers up against a Y Combinator talk I’d been chewing on, the whole thing snapped into one sentence I keep coming back to: your company already has a brain — it just keeps forgetting. Everything below is a footnote to that. None of it requires you to be big.
The Walmart quarter nobody framed correctly
On the surface, Walmart’s Q1 was a good retail quarter. Revenue up 7.3%, U.S. comps up 4.1%, global e-commerce up 26%. Fine. But sales isn’t the story. The story is what kind of company is now growing fastest inside Walmart.
Advertising grew 37% globally. Walmart Connect, the U.S. ad arm, grew 44%. Marketplace was up around 50%, its best run in ten quarters. And the CFO said it directly on the call — advertising and memberships are now roughly a third of Walmart’s earnings. That’s not the company most of us first started selling into.
Why should that matter to a small operator and not just to Walmart’s shareholders? Because every one of those high-margin engines runs on a loop. A search term feeds the ad targeting. A transaction feeds the inventory routing. A delivery feeds the next promise of “under three hours,” which pulls the next order, which feeds the loop again. Walmart didn’t hire its way to this. It turned its own operations into a machine that learns from itself. Hold onto that.
The talk that explains the quarter
The talk is Tom Blomfield’s “How to Build a Self-Improving Company with AI.” His first move is to throw out the framing almost all of us use. We say AI will make the team 20% more productive. He says that’s bolting a faster engine onto an old car — you’re still driving the old car. The real shift isn’t productivity. It’s capability. Building a company that can do things it structurally couldn’t do before.
The mechanism is a loop with five moving parts. Something senses what’s happening — customer emails, support tickets, cancellations. A policy layer holds the rules for what the AI decides on its own versus what needs a human. A tool layer actually does the work. Quality gates catch the mistakes before they ship. And a learning step closes it: the system notices where it fell down, asks why, and feeds that back to the front. That’s the whole difference between a chatbot and a loop. A chatbot answers the question you asked. A loop notices where the work broke and is better the next time.
Two ideas from it have been rattling around my head all week.
The first: your real asset is your context — your data, your customer history, the operating know-how you paid years to earn — and the software sitting on top of it is disposable. You protect the knowledge. You throw the dashboards away and rebuild them as the models get better. The second is his line about money. If your API bill doesn’t make you uncomfortable, you’re not doing enough. The question stops being how many people you can hire and becomes how much useful machine work you can safely run. There’s a companion talk from Diana Hu with a phrase I keep stealing: tokenmaxx, don’t headcountmaxx. Same idea, blunter sticker.
The dark mirror
Before this turns into cheerleading — look at Meta. Last week it started cutting around 8,000 jobs, roughly a tenth of the company, and killed thousands more open roles on top of that. Not because the quarter was bad. Because of the bill. Meta’s 2026 AI spend is guided toward $125–145 billion — several times its entire payroll. One write-up nailed it: the layoffs aren’t the cost-cutting, they’re the financing. “Burn tokens, not headcount” sounds great when you’re five people in a room. At Meta’s scale it means real people losing jobs so a data center in Ohio can switch on. I don’t think you get to quote the upside honestly without sitting in that for a second.
What this means if you don’t work in tech
Here’s where it lands for an actual business. A rug company in Easton, Pennsylvania, say.
A home-goods business runs almost entirely on knowledge that lives nowhere you can point to. Which loom runs which quality. Why one size keeps coming back. What a customer needs to hear before they’ll commit to an 8×10. Which colorway moved last spring, and why. That’s not trivia. Blomfield’s whole argument is that this — the scattered, undocumented stuff — is the actual strategic asset, and most of us let it evaporate the day the person who knew it leaves. You’ve felt it. Everybody has. The knowledge walked out the door when she quit. What’s new is that AI is the first tool that genuinely fixes it, and the first step is almost insultingly low-tech: write things down somewhere a machine can read them.
Which is the bridge to the other big story from last week. And it isn’t optional anymore.
Your storefront has to be legible too
While the retailers reported, the way people find products kept moving under everyone’s feet. Google launched “Buy for me” checkout — an AI that completes the purchase on a merchant’s site for you. ChatGPT, Perplexity, and Amazon’s Rufus are all sprinting at the same thing; Amazon says Rufus alone drove nearly $12 billion in incremental annualized sales. Adobe clocked AI-source traffic to U.S. retailers up almost 400% year over year, and it converts better than any other channel. McKinsey thinks agentic commerce could redirect $3–5 trillion globally by the end of the decade.
Now the part that should change what you do Monday. An AI agent doesn’t browse your beautiful product page. It parses it. It can only recommend your rug if the price, the availability, the return policy, the dimensions, the material are all sitting there as clean, machine-readable structured data — not trapped in prose, not in a PDF, not baked into an image. This is what’s replacing SEO; people are calling it Answer Engine Optimization. If your knowledge isn’t legible, you’re invisible — to the buyer’s agent, and to your own systems.
Which is Blomfield’s principle again, just turned to face outward. Make everything legible. Inside, so your own systems can learn from it. Outside, so the agents can find you.
A design footnote I can’t shake
One more from Google I/O, because I spent real time with it: Material 3 Expressive, the biggest update to Google’s design system in years. Google says it’s the most-researched design work they’ve ever done — tens of thousands of participants, dozens of studies — and it leans hard into motion, richer color, bolder type, a much bigger library of shapes.
Why does a design system belong in a business newsletter? Because it’s the same idea as the company brain, pointed at your interface. A set of consistent, documented rules every screen pulls from, so the whole product stays coherent as it spreads across phones and watches and cars and glasses. For a small operator it’s a gift — a free, well-documented visual vocabulary you can borrow so the thing you build looks native instead of homemade. Consistency is just legibility wearing different clothes. Machines one minute, humans the next.
The one move that costs nothing
If you take a single thing from this issue, take this. What’s slowing your business down isn’t a lack of access to AI. It’s that your best knowledge isn’t written down anywhere a machine — or honestly, your own future self — can use it. Walmart spent billions making its operations legible to its systems. You don’t have billions. You have a workbench and a quarter’s worth of know-how in your head that nobody else can see.
So before you automate anything, capture. Write down the decision and why you made it. Record your version of office hours. Get the supplier rules and the size-question patterns out of people’s heads and into one place someone — or something — can actually read. That’s it. That’s the move a textile importer, a plumber, and a dentist can all make Monday morning, with no code and no API bill.
Your company already has a brain. Stop letting it forget.
— Adem
Sources I leaned on: Walmart Q1 release · Walmart 8-K (SEC) · Target Q1 · TJX Q1 transcript · Blomfield / YC · Meta layoffs + capex · Google I/O 2026 · Material 3 Expressive · McKinsey on agentic commerce · Adobe AI-traffic data · structured data / AEO
Hey, I’ve Been Thinking… 
Leave a comment