Investing in AI Startups: A Conversation with Warren Packard and Amit Garg

Stanford Angels and Entrepreneurs United (SAEU) hosted an online panel discussion in May 2025 on the topic of investing in AI startups. Warren Packard, a partner at the AI Fund, and Amit Garg, founder and a partner at Tau Venture, participated in the panel, which was moderated by SAEU co-president Sheila Proeve.

The Rise of AI

Warren said AI started as a component of what we are doing. Now, AI is the core of what almost everyone is doing, in terms of starting up companies. Further, he sees that AI can be a programming companion. The evolution in AI is becoming more intelligent, more interactive, more involved in the products. It has shaped startups and what they can do, the speed at which they can create and evolve and iterate their code, and even the fundamental architectures of the products.

Amit said AI as a modern concept has been with us for almost 70 years. What's different now is that there is a lot more data, computational power, models, and people who understand what AI is. The combination has meant the things that we hoped to do are now possible. Having an agent that does tasks for you, whether it's programming or reducing paperwork in healthcare, is a huge difference, with 10x potential savings or more. As an example, Iterative Health was able to go from the average of 75% detection to 99% detection of cancer by taking the video feed off the colonoscopy in real time, running it through their AI and saying, 'there's something here'.

The State of the Market

Warren said there are tailwinds with AI. Corporations across the globe are interested in jumping in the AI race, and really don't know how to do that. His firm is partnering to build companies together.

Amit said his firm is excited about applying AI to solve problems, so it is not investing in core AI. Core AI is the companies building the tools, like OpenAI, Anthropic, Google and Microsoft. His firm is investing in companies that take those tools and use them to solve the problems.

Identifying Opportunities - Which niches or categories show lasting promise

Warren said his firm can see the power of AI today and where it's going in the future, from a technology basis. Agentic AI is where individual language models take on specific personas and pursue specific roles in concert with other LLMs. There's also a trend within programming to leverage AI for creating amazing programs extremely rapidly. The final trend technology trend is image, audio and video models, and being able to interface with AI. It is about solving real world problems, not the technology itself.

Amit said there is a bimodal distribution. There are companies that are raising massive amounts of money without having much revenue because they have promise in their potential. There's a dichotomy of reactions around AI because some people see what is working and other people see why it's not working. The trend he has focused on is that AI is making an impact into verticals that have traditionally been very resistant towards change. Legal and gov-tech are sectors that are laggards that resist. They are adopting these technologies at rates that he has never seen before.

Early Signals that stand out when Evaluating an AI Startup for Investment

Amit said team is always the number one factor. You are investing in people. The second T is technology - how they're applying technology and building a moat. And the third is traction.

Moats or Models that Matter most in Building AI Companies

Warren said the team is absolutely a moat, as is proprietary data access to train models on or do evaluations. Beyond that it's execution. Execution is all about being able to deliver and being incredibly nimble because the environment is changing and adopting. Start-up should embrace AI as a tool set to develop products rapidly. If they can execute, they can grow their customer base. What's not a moat is AI technology.

Funding, Deal Velocity, Investor Selectivity, Round Dynamics and the Exit Environment

Amit said Carta publishes data on rounds, the sizes, the velocity. Fenwick, a law firm in the Bay Area, also publishes about the state of venture capital and startups. It's a bimodal distribution. A few startups are able to raise massive amounts of money before having significant traction. That's not for the 99.9%. The average seed in the Bay Area is $3-4 million, and valuations are usually 4-5x that amount. He sees three months as the average to be able to fundraise - one month to get to a partnership meeting, a month to get the term sheet and a month to close the syndicate. He recommends that founders build a spreadsheet with a hundred names and divide them into three tiers based on fit. Start with tier two, test the idea, pitch for a couple weeks, and then migrate to tier one. And never except for the very first round have less than six months of cash.

Warren said companies that are executing well should be able to raise a round of financing. If it's taken a long time to raise, they may not have hit the milestones that are required to get to that next round. Companies looking for pre-seed financing are typically seeing a million-dollar check, and then 6-9 months later, they're going out for their seed round.

As AI matures, is there more clarity on Exit Paths?

Amit said M&A and IPOs have been lower for two or three years. M&A is coming back slowly but surely. Acquisition is the result of a long-term relationship that converts at the right time. VC is looking for people who are building for 10 years and possibilities of 10x or more. If a founder is building a business to exit in two or three years, they should not come to VCs, They can go to other sources of capital: family offices, corporates, grants, loans or their own revenue.

Given the AI tool support for founders, are Raises getting Smaller?

Amit said you would think so, because as things get more efficient, founders need less money. They don't need to build AI tools. However, the round sizes have kept going up because there are other things feeding into it.

Warren said successful companies are able to raise more money to do more things. The entire financing is not necessarily going to be less. However, the pre-seed financing is coming down because of AI tools.

What Recommendations do you have for a Mid-career Person starting a Company?

Warren said whether you're early career, mid-career or late career, just embrace AI. Use the tools that are out there in your daily life. Code and get an app up. "If you're passionate about something, in addition to your family, go ahead and start it."

Seed Rounds are around $3-4 million. How does Funding Work?

Amit said that, typically, the lead is putting in about half of the round. The definition of a lead is not the amount. It's that they are setting the terms. Seed stage is all about collaboration. Seed is high risk, so investors want other good people in the deal. A syndicate has a lead taking 50%, a couple of follows, and then fills up the rest with angels. He encourages every founder to spend a minute on investor's websites to understand what their investment philosophy is and then ask questions to understand how much ownership they want, how much follow-on capital they will have, and what their strategy is for the next round.

How can newer Companies build Moats? Beyond core LLM, which Investible Segments look Promising?

Warren said the key is for entrepreneurs to realize what their advantages are. If they can get in front of customers and iterate quickly, they can outrun larger companies.

Amit said an example is Signos, which leverages a continuous glucose monitor, takes the data and understands what happens when people eat, drink, exercise or sleep. They got to hundreds of thousands of users with no marketing. They went directly to consumers first because it's a problem that many people are attuned to. They're making a pivot into B2B2C, which is going through providers and doctors. They got to a series B with that business model. "We see that a fair bit. Companies start with consumers and then eventually expand."

If you are a Stanford founder building in AI today, what would you Focus on or Avoid over the next 12-18 months?

Warren said he would focus on staying up with the state of the art in AI and staying informed. "Do what excites you. Solve a problem. Build a solution that has some real power."