The supply side of AI
Issue 285: What you sell in a marketplace changes with every platform shift
Every major distribution shift creates a new marketplace. And every new marketplace starts with the same question: what’s being supplied?
The answer is never static. What fills the shelves of a new platform economy always begins with the most obvious artifact—the thing that looks most like what came before—and evolves as builders discover what the platform actually makes possible. The interesting part isn’t the first wave of supply. It’s the second and third waves, when the marketplace starts to reflect the platform’s true nature rather than its predecessor’s shadow.
We’re at this inflection point with AI. The marketplace is forming. The supply side is being defined in real time. And if history is any signal, what we think we’re selling now is not what will matter in two years.
Distribution creates the marketplace
The CD-ROM was one of the first moments where software had a physical supply chain. You walked into a store, picked a box off a shelf, and took it home. The supply was literal: packaged software, printed manuals, jewel cases with cover art. Distribution was constrained by retail space. What got shelf placement won. The marketplace favored products that could justify their physical footprint—large applications, encyclopedias, games. The constraint shaped the supply.
The App Store changed the equation entirely. Suddenly, distribution was infinite. No shelf space. No packaging. No retail relationship required. The supply side exploded: from a few hundred apps to millions. But more importantly, what counted as supply changed. It wasn’t packaged software anymore. It was lightweight, single-purpose utilities. A flashlight app. A fart button. Things that would never have justified a jewel case now had a viable path to distribution. The marketplace didn’t just grow—it redefined what a product could be.
Then came integrations. Platforms like Slack and Shopify built ecosystems where the supply wasn’t standalone software at all—it was connective tissue between systems. The marketplace shifted from products to capabilities. You weren’t buying an app; you were buying what an app could do *inside* something else. The value moved from the surface to the seam.
Each of these shifts followed the same pattern: a new distribution mechanism emerged, the marketplace initially filled with familiar-looking supply, and then the supply evolved to match what the platform uniquely enabled. The CD-ROM marketplace looked like a software store. The App Store marketplace eventually looked like a services economy. The integrations marketplace looked like infrastructure.
What you supply will change
This is the part most people miss when a new platform emerges. The initial supply—the thing that fills the marketplace first—is almost always a translation of what existed before. Templates on Webflow. Themes on WordPress. Presets in Lightroom. The first instinct is to package expertise in a form that’s familiar.
Webflow’s marketplace is a good example of this evolution in practice. It started with templates—complete, pre-designed websites that you could purchase and customize. This was the natural first supply: it mirrored what existed in the WordPress and Squarespace ecosystems. Then the marketplace expanded to include libraries—reusable layout components and design building blocks. Then apps. Then expert services. The supply went from finished artifacts to modular capabilities to human expertise. Each layer reflected a deeper understanding of what the platform actually needed.
This isn’t unique to Webflow. It’s the pattern. Figma’s community started with UI kits and grew into plugins and now developer-facing tools. Shopify’s ecosystem went from themes to apps to entire fulfillment networks. The marketplace always migrates from static assets toward dynamic capabilities.
The lesson is that if you’re building for a marketplace, you shouldn’t only optimize for what’s being sold today. You should be watching for what the platform makes newly possible—and positioning for that supply before it’s obvious.
Supply for AI
So what does the supply side of an AI marketplace look like? We’re early, but the categories are starting to take shape. And as with every previous platform shift, the first wave of supply looks familiar. The next waves won’t.
Skills
The most immediate form of AI supply is packaged expertise—skills, prompts, workflows, and instructions that make a model better at specific tasks. This is the template era of AI marketplaces. Just as the early App Store was full of simple utilities, the early AI ecosystem is full of system prompts and agent configurations. They’re useful, but they’re the starting point, not the destination. The interesting evolution will be skills that are dynamic rather than static—expertise that adapts to context rather than executing a fixed script.
MCP access
The Model Context Protocol is emerging as the connective layer for AI, much like APIs became the connective layer for integrations a decade ago. MCP servers are supply. They give agents the ability to interact with systems—databases, design tools, communication platforms, infrastructure. The MCP marketplace is already forming: hundreds of servers, each exposing a different capability. This is the integrations era of AI, and it’s happening fast. Companies like Google, Microsoft, and Stripe are already shipping managed MCP servers, treating them as first-class distribution for their services. The supply here isn’t software—it’s access. The product is the connection itself.
Data and graph
The most valuable AI supply might not be capabilities at all—it might be context. Structured data, knowledge graphs, organizational memory, proprietary datasets. As models become more capable, the bottleneck shifts from what the model can do to what the model knows about *your* specific situation. The marketplace for context—curated, trustworthy, domain-specific data—is barely visible right now, but it may end up being the most important supply category. This is the equivalent of moving from apps to infrastructure: less visible, more foundational.
Insights
Beyond raw data, there’s a layer of supply that packages interpretation: benchmarks, evaluations, quality signals, and taste. As AI output becomes abundant, the scarce resource is knowing what’s good. Curation becomes supply. Evaluation frameworks become supply. The ability to tell signal from noise—to separate useful output from AI slop—is itself a product. This category doesn’t have a clean analog in previous platform shifts, which might be exactly why it matters.
The pattern holds
Every distribution shift creates a marketplace. Every marketplace starts with familiar supply. And then the supply evolves to match what the platform uniquely makes possible.
We’re in the template era of AI marketplaces right now—skills, prompts, pre-built agents. It’s useful. It’s also temporary. The next wave will be about access and connectivity: MCP servers, protocol-level integrations, agent-to-system bridges. The wave after that will be about context and judgment: proprietary data, knowledge graphs, curation, and evaluation.
If you’re building for the AI supply side, the question isn’t what’s selling today. It’s what the platform makes newly possible that wasn’t possible before—and what form of supply that implies. The CD-ROM gave us packaged software. The App Store gave us lightweight utilities. Integrations gave us connective tissue.
Here’s my prediction: context becomes the new code.
The models will commoditize. The interfaces will converge. The skills and prompts that fill today’s marketplaces will be table stakes within a cycle or two. What won’t commoditize is proprietary context—the structured knowledge, domain-specific data, and organizational graphs that make a general-purpose model genuinely useful for a specific situation. The companies and builders who accumulate context, who build rich and trustworthy data layers, will hold the real moat in the AI era. Not because they built better software, but because they built better understanding.
Every previous platform shift relocated value from the obvious layer to a deeper one. Software moved from the box to the app. Apps moved from the surface to the integration. And now, the integration is moving from the connection to the context behind it.
In five years, the most valuable listings in an AI marketplace won’t be agents or skills or MCP servers. They’ll be curated knowledge graphs — proprietary, domain-specific datasets that turn a general model into an expert on your industry, your customers, your supply chain. The company that builds the Bloomberg Terminal of AI context, the Shopify of structured domain knowledge, will own the supply side the same way AWS owned infrastructure for the last era. That’s the marketplace worth building for.