AI Funds and Founders
What I look for in an AI founder after building seven AI agents of my own
The AI founders worth backing are the ones who can tell you what one task costs, who pays for it, and what happens when the model underneath them gets better. I know because I have built seven AI agents for my own businesses at Guz.ai, and every one of them taught me a question I now ask founders. This is the list, and the money context around it.
The money is real, and concentrated
As of the end of 2025, AI companies raised about $212 billion globally, up 85 percent from $114 billion in 2024, and close to half of all venture funding for the year, according to Crunchbase. About $159 billion of that, 79 percent, went to companies based in the United States. CB Insights put total venture funding at $469 billion for 2025, the highest since 2022, with 738 mega-rounds capturing $307 billion, or 65 percent of all venture dollars. Half the money in one category, and two thirds of it in rounds of a hundred million or more. That is a market where the median early-stage AI company is competing for attention with a handful of giants, and where an early-stage fund’s edge has to come from judgment, not access.
What building the agents taught me
Guz.ai started as a tool for my own work: a founder app that walks a consumer entrepreneur through a thirteen-stage journey in my voice, with my books and call transcripts behind it. Then came a press release ranker, a website ranker and a website builder, a sales agent, a video machine and an ad machine. Seven product lines, built by a small team on top of frontier models, scored against rubrics, used every day in my own companies. Here is what the work taught me that no pitch deck would have.
1. The model is not the product. The workflow is. Every agent I built got better when the model under it got better, for free. The value we added was the workflow around it: the research step before writing, the scoring loop that refuses to publish below a ten, the compliance pass that nobody can skip. A founder whose moat is the model will lose it at the next release. A founder whose moat is a workflow that customers run every day keeps it.
2. Every task has a cost, and the founder must know it to the cent. Tokens are a cost of goods. When I ask a founder what one completed task costs them in model calls, and what the customer pays for it, I learn whether they have a business or a demo. The ones who know the answer also know how the cost falls as they cache, batch and route to cheaper models. The ones who do not know are selling a margin they have never measured.
3. Evaluation is the product’s quality control, and most teams do not have one. In consumer goods, nobody ships a beverage without a spec sheet and a QC test on every batch. In AI, I meet teams shipping to customers with no written definition of a correct output. The founders I trust can show me the rubric their agent is scored against, the failure cases they collected, and what they changed when the score dropped.
4. Distribution still decides. This one I learned long before AI. I put a product into four thousand Walgreens stores without corporate approval, one manager at a time, and no amount of product quality would have done it for me. The AI founders who win have a specific answer to who sells this, to whom, and what the first hundred customers have in common. “Developers will find us” is not an answer, and neither is “we will add sales later.”
5. The founder has to be the first user. My agents are good because I use them every morning and feel every failure. When a founder cannot demo their own product from memory, or has never run it on a real customer’s data in front of me, I assume nobody else has either.
The questions I ask, in order
- What does one completed task cost you, and what does the customer pay for it?
- What breaks when the model underneath you improves, and what do you keep?
- Show me the rubric your output is scored against, and the last time it failed.
- Who sold the first ten customers, and what did those customers have in common?
- Run it for me now, on something real.
A founder who answers all five has a company. Most answer two. That gap is where an early-stage AI fund earns its return, and it is what I am building the fund around. The fund is not named here and nothing on this site is an offer of any security; this is how I think, written down.
Why an operator belongs in this market
Half the venture money in the world is going into AI, and most of the people deploying it have never run the kind of business the software is sold to. I have run the distributor, the brand and the public company that AI tools are now pitched at. When a founder says their agent replaces a sales rep, I know what a sales rep does on a Tuesday. That is the diligence edge, and I will keep writing about it on the AI funds and founders hub. If you want to see the agents themselves, the founder app is free at cpglife.com.
Frequently Asked Questions
How much venture funding went to AI in 2025?
As of the end of 2025, AI companies raised about $212 billion, up 85 percent from 2024 and close to half of all global venture funding, according to Crunchbase, with 79 percent of it going to United States companies.
What should an investor look for in an AI founder?
As of 2026 the five tests I use are the cost of one completed task, what survives a better underlying model, a written evaluation rubric with real failure cases, who sold the first ten customers, and a live demo on real data.
Is an AI company’s moat the model?
No. Every agent I built at Guz.ai improved for free when the underlying model improved; the durable value was the workflow, the evaluation loop and the distribution around the model, not the model itself.
Why does distribution matter for AI startups?
Because product quality has never sold itself in any category I have operated in; a product reached four thousand Walgreens stores through one manager at a time, and AI software reaches its first hundred customers the same way, through a named person selling it.