The AI product era is over before it began. Not because the technology failed—OpenAI's GPT models, Anthropic's Claude, and Google's Gemini have achieved remarkable capability milestones—but because consumers never wanted AI products in the first place. They wanted what AI products could do for them. That distinction, which Daring Fireball's John Gruber articulated in a widely-shared essay this week, reframes the entire platform wars narrative and exposes why every company racing to ship "AI-powered" features is building on sand.
Gruber's argument is deceptively simple: AI is enabling technology, not a standalone product. The analogy he draws to electricity is apt. Nobody buys electricity. They buy refrigerators, washing machines, and lights. The infrastructure is invisible; the outcomes are everything. The picks-and-shovels framing that venture capitalists have applied to AI infrastructure companies—Nvidia, cloud providers, foundation model APIs—captures this reality more accurately than the consumer product narrative that dominated 2023 and 2024.
The evidence is accumulating. Apple's "Apple Intelligence" branding quietly buried AI into iOS updates rather than launching it as a destination feature. Users gained writing tools, image generation, and Siri improvements without ever being asked to adopt a new product. Microsoft's Copilot integration into Office followed the same playbook: AI as upgrade to existing workflows, not a separate purchase. These companies understood what early AI startups learned the hard way. When Rabbit R1 and Humane AI Pin launched as dedicated AI hardware products, consumers rejected them. The Humane device sold fewer than 100,000 units before the company explored a potential sale. The Rabbit R1 became a punchline.
The counterargument—that ChatGPT, with over 180 million monthly active users, proves consumers do want AI products—misses the distinction between endpoints and destinations. ChatGPT is infrastructure for developers. Its API drives hundreds of billions of tokens through business applications. The consumer interface is the public face, but the value creation happens downstream, in the tools built on top of it. When Stripe integrated GPT-4 into its fraud detection, no customer needed to learn a new AI product. The improvement was invisible and seamless. That is the template.
This has profound implications for the platform wars. The battleground is not which AI product consumers prefer, because that market may never materialize at scale. The battleground is which company's infrastructure gets embedded into the products and services people already use. Enterprise software vendors—Salesforce, ServiceNow, Workday—have the most to gain, because they control the workflows where AI assistance becomes indispensable. Their customers will adopt AI through existing software updates rather than purchasing AI subscriptions separately.
The companies positioned to win are not the ones with the best AI models, though that matters. They are the ones with the deepest distribution into existing habits. Apple's 2 billion active devices, Microsoft's enterprise install base, and Google's search integration represent distribution advantages that no AI capability lead can overcome. Gruber is right: AI is infrastructure. And in infrastructure markets, distribution is king.
The startups still chasing consumer AI products should read Gruber's essay twice. The picks-and-shovels thesis does not mean opportunity has vanished—it means the opportunity is API-first, platform-first, and invisible to end users. The next great AI company may never launch a consumer-facing product at all.