The best early-stage investors going forward will build internal AI systems.

About a year ago, I started playing around with v0 and Cursor to build small products. I quickly realized that as someone with no prior coding experience, I can build full-scale products in the matter of days. It wasn’t as easy as I make it sound, but with the help of some friends in the engineering space, I built some pretty incredible things.

A few months ago, I brought this into my job and started building internal products for myself and teams at my company. Through this we have realized how the barrier to entry for building software has never been easier.

I believe over the next few years, we will see two things:

Companies will buy too many AI SaaS products. (This is happening now)
Companies will start building their own internal tools (Will start slowly, then accelerate) Visually, this will look like the chart below. The winners are going to be the companies that make building internal software easier (aka Composer).

This is an opinion I’ve held for a few months and continues to be re-enforced. Most recently, I was listening to Michael Lord Barton on the Sourcery podcast talk about how he is building AI systems in his current role (a trader at Coatue). He spoke of AI like a team of analysts that are working 24/7 to gather information. Removing the barrier of human labor when it comes to information gathering will allow him to execute at a more efficient level than he was prior.

Going forward, I think not only will the best companies build internal AI operating systems and tools, but also the best investors (private and public) will build internal systems that help them gain the extra edge on their competition.

Company Internal OS

Every company operates completely differently. I won’t speak on companies building internal AI tools much, but I will talk to some of the important aspects that go into how companies are building right now. This transition will happen slowly, but then all at once. It’s already starting to happen:

At my current company, I have spent time building small internal tools that have helped teams with company specific problems that no third party software provides. There have been no conversations about building a full company operating system, but the mountain that would need to be climbed is substantial.

There is first a conversation of tradeoffs. Is it worth having a team build this vs build client facing products? The second conversation is around data and context. An internal OS by nature nowadays would be extremely AI driven. This means a ton of data and API connections would be needed in order to piece together everything, connecting all these pieces would require heavy engineering hours.

As I said above, I am not going to talk in depth about company operating systems. I’ve built small internal tools, but nothing as large of a project as what we have spoken about above. I will spend more time around building investor operating tools as I have spent substantially more time there.

Investor Operating Systems

Public investing is a mammoth of an industry and every single hedge fund and bank has been leveraging machine learning and AI before ChatGPT was even released. With that, I will focus specifically on the private markets and more specifically, early stage investors.

Across all asset classes however, investors thrive on information. The more information you can gather on a company, an industry, or a specific problem, the better you can understand the long term opportunity that is in front of you. Partner a pile of information with a key insight, view, or opinion about how the world will work over the next decade or so and you can make some pretty incredible investments.

The first part of this, information gathering, is where AI can thrive.

Building Around This Notion

A few months ago, I had a friend tell me about how researching companies prior to YC demo day was a pain. 150+ companies in a batch and only a few weeks to find the ones you like, research them, and schedule meetings.

So, I sat down over a view weeks and built out BatchPro for him. The idea was to take every company in the current batch, surface the basic information, and then use AI to enhance the data around each company. For each company in the batch we generated over 10+ pages of research in just a few API calls.

Layered on top of this, we use AI to understand what your investment preferences might be and surface the top companies in the batch for you.

This same practice should be built on a firm-by-firm level.

The story here is all about information gathering. AI should be treated like junior level analysts. These analysts should work to tirelessly to gather information, build reports, give highlights, and should always be working. It is impossible as a human to talk to new companies, discover new opportunities, while also keeping up with how one portfolio companies industry is changing. When Barton mentioned building a “team of AI analysts”, this is how I imagine operating this inside of a venture firm:

Three main teams that would serve three distinct purposes.

Internal Process Management

This is firm specific but really would refer to any memo writing, back-office finance, or busy work that needs to be done throughout the year. These automation tools can be built pretty quickly with the help of tools like Replit or Retool and can be one-off in most cases.

Portfolio Company Management

Most firms are working with a portfolio of 25+ companies. Each partner is responsible for a select set of these companies, but it is hard to constantly be in the loop with every company. This is where a portfolio-CRM that is driven by AI analysts can help. One analyst can research competitors for a company, another can research industry wide news, a third can research headwinds for the company, and a forth can research tailwinds. These analysts can then run on a set schedule to be refreshed every few weeks.

You can then pair these analysts with summaries of internal notes and investor updates you receive from each company.

You now have an in-depth report on every company in your portfolio at any given time.

New Opportunity Management

This team of analysts is where firms can get creative. A better way to say “new opportunity” would almost be “sourcing” agents. A first step here would be to build the same portfolio company management analysts, but for companies that are entering the diligence process. The next steps would be to work to build agents that actively conduct sourcing activities.

This is a very open-ended problem for each firm and everyone has their own unique ways of sourcing companies. Whether it be scraping job boards, looking at the new YC batch, or scouring harmonic, all of these tasks can technically be done by an AI Analyst.

As you may have noticed though, I purposely didn’t call this third arm a “Sourcing Team”. Every firms sourcing methodology can be different from partner to partner, so it may not be in the best interest to build analysts around this. Building analysts to help with the diligence of a new opportunity? That’s a different story.

The best firms will move to this operating model

Over the next few years, firms will start to build bespoke software that lends itself to this operating model. You will continue to see multiple “AI for Venture” startups that claim to build this software for firms. In my mind, this is a waste of money. Firms should build software themselves because they know what to build in order to best serve their own goals and objectives.

The best firms will continue to build these internal operating systems as the models keep getting better and better. These firms will be the next decade’s winners.