The Rise of the Anti-API Brand

In which we consider the possibility that making your brand harder for robots to buy might actually be the power move.

Okay, I need to talk about something that's been rattling around in my brain for weeks. And no, it's not about another AI tool promising to "revolutionise" your product listings. It's about a quiet split happening in retail that almost nobody is discussing yet—and by the time everyone notices, the brands that figured it out early will be miles ahead.

Here's the setup: the dominant conversation in ecommerce right now is about making everything machine-readable. Clean APIs. Structured product data. Agent-optimised catalogues. Basically, rolling out the red carpet for AI shopping bots and hoping they pick your stuff. And for a lot of products, that's exactly right.

But there's a counter-argument forming that I find way more interesting: what if some brands' greatest advantage isn't how visible they are to machines, but how meaningfully they connect with humans through channels machines cannot replicate?

This isn't some Luddite fantasy. It's strategic. And a recent McKinsey analysis of what they're calling the "agentic commerce automation curve" backs it up with real data. (The kind that makes you sit up straighter at your desk. Or in my case, slouch differently on my couch at 11 PM. Same energy.)

Two Economies, One Retail Industry, Zero Chill

McKinsey's research projects AI agents could mediate $3–5 trillion of global consumer commerce by 2030. That's a number that makes you count the zeros twice.

But buried in that headline is a more interesting finding: automation doesn't climb uniformly across all shopping categories. It accelerates in some and deliberately plateaus in others. And that plateau is where things get spicy.

Follow the automation curve to its logical end and two distinct economies form:

The Agent-Selected Economy: Commodity shopping. Household essentials, basic electronics, groceries, anything replenishable. AI agents handle research, price comparison, substitution logic, and execution. Your role as a consumer shrinks to setting parameters and occasionally checking whether the robot bought the right paper towels. Shopping becomes a background process—as silent and automated as your thermostat.

The Human-Selected Economy: Luxury, fashion, milestone purchases, gifts—anything where identity and emotion are inseparable from the transaction. Here, delegation stalls. Not because agents lack capability, but because the human involvement is the product.

It's not a clean binary today and may never be. But every brand will increasingly need to know which economy it's competing in, because the competitive logic of each is fundamentally different. If you're optimising for the wrong one, you're bringing a spreadsheet to a feelings fight.

The Identity Plateau (Or: Why No One's Letting ChatGPT Pick Their Watch)

McKinsey describes delegation plateauing where "identity, aspiration, and regret risk" are high. Let me unpack that, because it's the most interesting part of this whole thing.

Imagine a consumer who happily authorises an AI agent to maintain their household supplies under a monthly budget. Swap out-of-stock paper towels for an approved substitute? Sure. Optimise delivery windows? Go for it. That's level 3 or 4 delegation, and it works because the stakes are low and the value is pure efficiency.

Now imagine that same consumer shopping for a watch. They might ask an agent to analyse resale value across brands, compare movements, verify authenticity, and locate global availability. The agent crushes all of this. But when it comes to the actual purchase? The consumer takes over.

Not because the agent can't execute the transaction. But because the act of choosing is where the meaning lives.

If an AI agent purchases a $10,000 watch based on optimal resale algorithms, the watch becomes an asset on a balance sheet. If the consumer spends months researching, visits the boutique, handles the piece, and finally commits—the watch becomes a story. The difference has nothing to do with the object and everything to do with the process.

This is the identity plateau: the point at which further automation doesn't add value—it subtracts it. (I spent a genuinely embarrassing amount of time thinking about this at 1 AM on a Tuesday. My partner has stopped asking what I'm reading, which feels like progress.)

Friction Is Dead. Long Live Friction.

In the agent-selected economy, friction is the enemy. Every click, every comparison, every moment of cognitive load that can be eliminated is an opportunity for AI. Agents win by making shopping frictionless. We know this.

But here's where things get counterintuitive: in the human-selected economy, friction might not be an inefficiency to optimise away. It might be the mechanism through which objects acquire emotional weight.

The effort of searching. The ritual of trying on. The sensory experience of a fitting room. Shopping with someone whose opinion matters to you. The deliberation. The anticipation. That feeling when you finally commit and your heart does the little thing where it's simultaneously excited and slightly panicked about the credit card statement.

These aren't bugs in the customer journey. They're the features that justify the price premium.

And as agentic commerce makes commodity purchasing increasingly seamless, the contrast with these tactile experiences only sharpens. In a landscape where nearly everything can be acquired through frictionless machine mediation, the experiences that can't be mediated start to feel more meaningful, not less.

Which means premium brands might not just tolerate the inefficiency of human interaction. They might invest in it deliberately. Scent, lighting, texture, the energy of a physical space, one-on-one conversations with someone who understands the craft. The kind of stuff that makes an AI agent go "I have no protocol for this" and politely wait in the car.

The Regret Paradox

In commodity shopping, agent error is a minor irritation. Wrong dish soap arrives, you note it, the agent adjusts. Tight feedback loop, trivial stakes. This is exactly where delegation thrives.

In identity shopping? Completely different. An AI agent can verify a dress is 100% silk, matches measurements, falls within budget, and arrives before a specific event. What it cannot verify is whether that dress makes you feel powerful walking into a room. Whether the colour sits right against your skin in the venue's lighting. Whether it communicates the version of yourself you want to project that evening.

These are what I'm calling subjective specs—criteria that are real, consequential, and completely internal. They shift with mood, context, and social setting. They can't be structured into product attributes because they don't exist as stable data points. They exist as feelings. (I know. I just used the word "feelings" in an ecommerce strategy newsletter. My attribution-model-obsessed past self is cringing.)

The result is a paradox: the categories where getting it wrong matters most are where AI is least equipped to prevent mistakes. Consumers will use agents as research filters but retain control over the final commitment. Not because they distrust the technology, but because outsourcing a decision that goes wrong here produces a uniquely painful regret: knowing you let a machine make a choice that was yours to make.

The Case for the Anti-API Brand (No, Not Like That)

Okay, here's where this takes its most speculative turn.

McKinsey warns that products "semantically opaque to machines risk becoming invisible in agent-mediated flows." For commodity brands, this is existential—if your detergent lacks clean data and exposed APIs, agents route demand elsewhere. Harsh but fair.

But for brands at the identity plateau, the relationship with machine legibility gets more complicated than a simple visible-or-invisible binary.

The argument isn't that luxury brands should retreat behind velvet ropes. (Though the image is amusing.) Agents need to surface products during research, or brands lose consumers before the human interaction stage. The opportunity is about where in the journey the brand invests its differentiation.

A brand maintains full machine legibility at the discovery layer—structured data, availability, spec comparisons—while simultaneously building experiential touchpoints that are inherently non-algorithmic. Private appointments. Artisan demonstrations. Curated in-store events. Commissioning processes that require extended dialogue between maker and buyer.

The structured data gets you onto the shortlist. The human experience closes the sale. And the gap between those two moments—where the agent hands off to the human—becomes the most strategically important real estate in the entire customer journey.

Over time, brands that do this effectively develop what we might call an anti-API identity. Not literally removing their APIs (please don't @ me), but becoming known as brands whose real value lives in spaces machines can't access. The API gets you to the door. What happens inside is deliberately, irreducibly human.

The Bottom Line for Sellers

If you sell commodity or replenishable goods: make yourself as agent-readable as possible. Clean structured data, exposed APIs, predictable fulfilment, transparent substitution policies. These aren't nice-to-haves. They're the baseline for remaining visible when AI agents are the primary shoppers.

If you sell in identity-driven or high-consideration categories: you need machine legibility at research and discovery—agents must surface your products during comparison, or you lose the consumer before they engage. But you also need to ask where the human experience creates irreplaceable value, and invest disproportionately in those moments.

If you operate across both economies: you may need dual approaches. An agent-optimised layer for your commodity catalogue, a human-optimised experience for your identity products. Treating both identically may increasingly serve neither well. (Yes, more work. I don't make the rules.)

The future of commerce isn't fully automated. It's selectively automated. And knowing where the machine should step back so the human can step forward might be the most important strategic capability in retail for the next decade.

Now if you'll excuse me, I need to go buy a watch. By myself. On purpose.

• Amazon is reportedly weighing a $50B investment in OpenAI, a move that could lift its valuation toward $830B and subtly reshape cloud power dynamics despite Amazon’s heavy backing of Anthropic.

• Ten charts cut through AI hype, showing explosive adoption, capital concentration, weak short-term ROI, and early job disruption, reminding us this is still a build phase, not payoff time.

• Power users and small teams thrive with agentic tools, while enterprises locked into weak Copilot setups risk falling behind fast.

• Ahrefs finds AI visibility now favors YouTube mentions, freshness, and “best of” lists over backlinks or long content, while Google still dominates traffic and ChatGPT leads AI referrals.

• Markets reeled as an Anthropic automation tool triggered a $285B stock selloff

• ElevenLabs says voice is becoming AI’s next major interface, moving control away from screens into wearables and daily life, while raising new concerns around privacy and surveillance.

📺 Smartly: Achieve better marketing campaign performance and business results with a single workflow using AI.

⚙️ Optimove: Data consolidation platform offering a Customer Data Platform (CDP) for unified, actionable customer insights across various sources.

📹 MeetGeek: Automagically record videos, transcribe, summarize, and share insights from every meeting to any tool.

👨‍💼 Aesop - Your AI-powered SaaS to find smart, distinctive names and the stories they tell – for your new brand, business, product, or project

🖼️ Lummi - Free stock photos made and curated by AI artists.

😍It was great, all I needed

🙂Good, not great

🙄It sucked