Investment Thesis: AI Automation Extension Marketplaces
Rob May is someone I respect deeply. Rob has invested in over 100 AI companies, has built and exited his own SaaS company and has a sixth sense for where technology is headed. Therefore I am excited to share his recent article on a new AI Business Model, which he has termed FAEM. Check out his substack Investing in AI for more exciting AI insights.
“Today I want to write about an emerging AI business model that I really really really like. I’ve seen 3 companies doing it and I’m sure many more are going to figure it out as well. During the thesis development meetings at HalfCourt Ventures, we decided to call it FAEM - Focused Automated Extension Marketplaces. I’ll explain what that means but also, if you have a better acronym let me know and I”ll update this post and credit you.
Here is how FAEM works. First let’s talk about the power of human-in-the-loop. We loved the HITL business model in previous years because it gives humans some final oversight of AI outputs, and helps collect more proprietary data for model training from the human input. The problem with it though, is it limits scalability and can sometimes lose much of the benefit of automation.
FAEM is a business model that takes less and less human-in-the-loop over time. It starts with an AI tool that needs some customized inputs. Let’s say its an image generation tool that needs a text prompt. It turns out some people are better at prompts than others, so a little over a year ago we started seeing things like prompt marketplaces. Instead of writing your own prompt, you could find prompts other people had written and put those into the tool as-is, or with your own tweaks.
This business model of having a marketplace of user generated content is powerful and can often be a great source of defensibility. Zapier is one of my favorite examples from the last wave of technology companies. By allowing users to create Zaps, the usefulness of the core Zapier platform just kept increasing in value. Now lets take the marketplace idea a step further and apply it to AI companies, and lets use it to replace the human-in-the-loop.
FAEM business models are marketplaces for small extensions that allow for more automation on a core AI platform. We say small extensions because if the extensions were large and complex, they should probably be part of the core platform. Small extensions allow for lots of variability. And if the platform can support lots of useful small extensions, it becomes defensible. Why? Because no extension represents a large enough proportion of the use cases to incentivize a competitor to replicate it.
For our acronym though, we chose “focused” over “small” because everyone wants FAEM and no one would like to be the SAEM.
The first concrete example I saw of this was Screens. The way it works is, Screens has a platform that checks legal documents for certain criteria. Then they have a marketplace where lawyers can write a Screen to run on the platform. For example, here is a screen for customer contract checks to run in M&A diligence. The screen is written by a human lawyer, then ingested into the Screens platform to check the customer contracts you upload, and make sure they meet the criteria specified by the screen.
This is an interesting twist on the human-in-the-loop model because the human isn’t in every execution loop, but their work is. It’s the best way I’ve seen to combine human and machine interactions.
There is some nuance here in the business model though that I want to point out. Just having a UGC marketplace does not make you FAEM-worthy. FAEM business models meet the following criteria:
There are thousands of possible human inputs to the AI platform.
Those inputs are complicated enough that having one person write them well one time is valuable.
The AI automation platform takes care of a significant portion of the work post- marketplace extension.
So, for example, if you have a marketplace but writing that marketplace extension is 85% of the work and the AI is 15%, its still too human focused to really benefit from the AI. What we are looking for in FAEM companies is that the AI is really powerful and beneficial but, the human inputs to it might also be complex or variable enough to support a marketplace, and that the lack of human inputs could otherwise be a barrier to usage. We believe this creates a virtuous cycle of extensions that make the platform better, which leads to more defensibility over the long term.
To look for FAEM opportunities, think about areas where AI could automate much of the process but still needs a lot of human expertise at the top. We’d love to see a product management / automated product builder type of company with a FAEM approach.
We’ve seen two more companies already deeply using the FAEM model, which we will stay mum on for now as we are considering investments. If you have a business that can be FAEM-ous, please reach out as we would love to talk to you about it.”
Executive Summary
The AI landscape is evolving rapidly, with new business models emerging to fill gaps in automation and scalability. One of the most exciting models we've identified is FAEM — Focused Automated Extension Marketplaces. This model addresses the inherent limitations of human-in-the-loop (HITL) AI systems by reducing ongoing human intervention while still leveraging human expertise in the early stages of platform development.
FAEM companies combine the defensibility of user-generated marketplaces with the power of AI automation. They allow users (often domain experts) to create extensions, scripts, templates, or configurations that plug into a broader AI platform. These extensions reduce the complexity for end users and expand the platform's capabilities exponentially. The marketplace creates a flywheel effect, where each new extension increases the platform's utility, leading to more users, more contributions, and deeper defensibility.
This thesis outlines why we believe FAEM companies will be a significant part of the next wave of AI success stories, key criteria for FAEM classification, examples of FAEM models in the wild, and our investment strategy moving forward.
The Problem with Human-in-the-Loop (HITL) Models
Human-in-the-loop (HITL) models were initially seen as a bridge between manual human workflows and fully autonomous AI systems. While effective, HITL has significant drawbacks:
Limited Scalability: As the volume of AI tasks grows, HITL scales linearly with human labor, reducing cost efficiency.
Bottlenecks: Human oversight slows down workflows that could be fully automated.
Training Data Dependency: HITL models often require constant human correction to improve the AI, which increases operational costs.
High Labor Costs: In domains like legal, medical, or product development, human expertise is costly, making HITL models economically unviable for many industries.
Solution: FAEM solves these issues by reducing human involvement in each execution loop while still incorporating human expertise into the platform in a one-time, up-front capacity. The key difference is that the human's role becomes a content creator instead of an ongoing operator.
What is FAEM?
FAEM stands for Focused Automated Extension Marketplace. These marketplaces enable experts to build, distribute, and monetize AI-driven "extensions" (also known as "apps," "skills," or "templates") that augment a core AI platform's capabilities. Users purchase and apply these extensions to improve or tailor the platform's performance to specific tasks.
Core Characteristics of a FAEM Business Model
User-Generated, Reusable Extensions: Users (or domain experts) create extensions (e.g., templates, prompts, rules, "screens," scripts, or configurations) that plug into a larger platform.
Marketplace Economics: The extensions are sold or shared in a marketplace, creating network effects where more extensions lead to more users and more revenue opportunities.
AI-Driven Automation: The extensions enable deep AI automation where the core platform can now perform tasks that previously required human expertise.
One-Time Human Input, Infinite AI Execution: Instead of constant human oversight (HITL), FAEMs only require human expertise once — when the extension is created. After that, the system runs autonomously, eliminating human bottlenecks.
Small, Focused Extensions: Extensions are lightweight and focused on specific use cases. They do not compete with the core platform, but they enhance it, making it modular and defensible.
Why FAEM is a Breakthrough Model
FAEM leverages the principles of modularity, reuse, and defensibility, much like the App Store model but applied to AI automation. Here’s why it’s a breakthrough model:
Flywheel Effect
More Extensions = More Value: Each new extension increases the utility of the platform.
More Users = More Extensions: More users create more extensions, drawing in additional users.
More Utility = Higher Switching Costs: Once a user relies on specific extensions, it becomes difficult to switch to a competitor that doesn’t have the same selection.
Platform Defensibility
The long-tail nature of FAEM extensions ensures no single extension accounts for too much value, discouraging competitors from copying the marketplace.
Unlike traditional SaaS, which can be replicated, a platform with thousands of niche, user-contributed extensions becomes a "networked fortress" that competitors can't easily replicate.
Monetization Opportunities
Revenue from Marketplace Sales: Platforms can take a percentage of all extension sales (like Apple’s App Store).
Subscription Model for Access: Users may subscribe to a platform for access to premium extensions.
Recurring Revenue from Enterprise Use: Companies may pay a license to use the entire platform and its extensions.
How to Identify a FAEM Opportunity
Not all marketplaces qualify as FAEMs. To identify potential FAEM investments, companies must meet the following criteria:
FAEM in Action: Key Examples
Here are some emerging FAEM players and how they operate.
Why We Are Bullish on FAEM
Massive Total Addressable Market (TAM)
FAEM models can emerge in almost any sector where AI needs human input. Examples include:
Legal Tech: M&A contract diligence
Sales Enablement: Custom sales playbooks
Enterprise Software: Product management tools
Finance: Risk management screens
Healthcare: AI-driven patient screening templates
Horizontal Expansion
A single FAEM marketplace can expand horizontally into new sectors. For instance, Zapier has Zaps for every SaaS company. If a FAEM company builds extensions for one vertical (e.g., M&A diligence), they can expand into compliance checks, HR onboarding, or regulatory audits.
Network Effects Create Moats
The more extensions that exist, the harder it is for a competitor to replicate the offering. As the network grows, so does user lock-in.
Investment Strategy
We are actively seeking early-stage FAEM companies that meet the following criteria:
Extension-Driven Model: Companies with a clear marketplace of user-contributed extensions.
Lightweight Extensions: Simple, modular tools that provide significant utility.
Automated Core AI: The platform does most of the work autonomously after the extension is added.
Industry-Specific Niche: FAEM models work best when focused on a specific industry with complex, domain-specific inputs.
Go-To-Market Strategy: Ability to quickly onboard customers to both use and create marketplace extensions.
Risks
Over-Reliance on Creators: If extension creators leave, marketplace supply may dry up.
Winner-Take-All Marketplaces: The most successful extensions may capture a disproportionate share of platform revenue.
Marketplace Liquidity: Getting enough creators and users to engage early is challenging.
Conclusion
FAEMs are a groundbreaking evolution of AI business models, offering scalable automation while capturing human expertise. Companies like Screens and PromptBase are just the beginning. We believe FAEM-driven businesses will generate defensible moats, outsized returns, and a new generation of market leaders. We’re actively seeking investments in this space.