By Sumeet Singh

When a new product category is emerging (e.g., fintech in the 2010s and AI apps today), the common wisdom is that an application layer emerges and an infrastructure layer emerges. I would argue however that there are two distinct infrastructure stacks that come to life: “technical infrastructure” and “commercial infrastructure,” where the former makes products possible and the latter enables products to complete their job to be done in a 10x way and often leads to truly N of 1 companies. I believe the most valuable companies in this phase of AI will fit this Commercial Infrastructure DNA.

Technical infrastructure includes the platforms and architecture stacks required to actually build a product and get it live. For building a consumer fintech mobile app this meant working with iOS or Android compatible programming languages (Swift/Kotlin), dev environments (xCode/Android Studio), software dev kits (iOS SDK/Android SDK), UI frameworks (UIKit/Android Framework), and various backend systems.

For building a web-based AI agent/app/clone/companion, the emerging technical infrastructure stack includes deployment (Vercel), accessing models (OpenAI), LLM orchestration (Langchain), storing data in vector databases (Pinecone), and app logic (Next.js).

For both types of products, the pieces of the technical infrastructure stack are critical, in that the products could not be built without them. Companies included below are examples, not the end all be all.

Technical Infrastructure Stack Examples:

Commercial infrastructure on the other hand allows for products to actually deliver the job to be done for their customer, often leading to N of 1 companies. When fintech was first emerging as a category, no fintech app could deliver a 10x UX improvement from their incumbent financial services companies if they had to rely on users to deposit and confirm micro-transactions with their existing bank account. Enter Plaid, critical commercial infrastructure that enabled consumer fintech companies to easily access customer bank account data and take action on the accounts. Without taking action on the funds or accessing deposit data, these apps had no value (i.e., you can’t underwrite, invest someone’s money, optimize their portfolio, file taxes, etc.).

I’d argue this idea of commercial infrastructure is not new, but rather it has been repeated through history. For example, the creation of the Commercial Internet Exchange (CIX) made it easy for anyone to get on the internet in 1991 (via one single Cisco 7500 router) — enabling web browsers like Mosaic to actually deliver their UI to the masses even though they had the technical infrastructure to build the product via the libwww library.

With a new AI application layer emerging in the form of agents, companion apps, vertical AI apps, etc., the following pieces of commercial infrastructure will help these apps deliver their own job to be done in a 10x way (e.g., have an agent take an action in the real world on your behalf or have your clone interact with a fan in a seamless way). Without a 10x user experience improvement enabled by commercial infrastructure, there would likely be too much friction for any of these new AI applications to get to scale, even if they can now be made possible:

• New authorization infrastructure will be critical for agents to be able to access and take action in various accounts internet users have, which is fairly long tail given how diverse internet usage is.
• For vertical AI apps, getting domain-specific context will lead to more accurate outputs, less hallucinations, and superior augmentation to general LLMs.
• We are on the precipice of the primary user of the internet changing from humans to computers via agents. In order for these agents to do their job to be done in a 10x way, developers will need better headless browser infrastructure that is faster, less brittle, able to read/write complicated websites (via React).
• For companion apps building relationships with humans for education, loneliness, therapy, health, etc. having a sense of history & memory via longitudinal user data will be critical in creating high fidelity experiences. These apps will also need to provide a feeling of being “more human than human” or having a soul, to ensure high fidelity interactions and relationships.

For commercial infrastructure companies enabling new product categories, I am looking for the following DNA, using the current AI wave as an example:

• The the underlying market being served by these picks and shovel plays need to be undergoing an exponential amount of growth to support a venture-scale, commercial infrastructure business → AI is experiencing explosive growth (180M MAUs for ChatGPT, 100M+ MAUs for Character.ai, 1M MAUs for Godmode, all within 12 months of launch)
• The product category needs to emerge on the back of either a technology innovation or regulatory change (which can lead to explosive growth), and perhaps to a lesser extent macro/demographic shifts → AI products that can interact with the real world are emerging on the back of LLMs :: fintech products emerged due to the Dodd-Frank regulation
• A part of the stack that is critical in nature to actually complete the job to be done in an incredible way → not just a nice to have but a must have (compare the combined EV of embedded fintech APIs for things like tax, accounting, insurance, etc. vs. that of Plaid)