AI-Native VCs: People, Automation, and Everything In Between

(Also available on LinkedIn)

More and more investors are looking for AI-native startups, which makes me wonder, what would AI-native VCs be like? Marc Andreessen predicts that venture capital may be one of the few jobs that will survive the rise of AI automation, arguing that VC work is more "art than science" and requires human judgment that AI cannot readily replicate. On the other hand, QuantumLight, a venture capital firm, claims to have built an AI system that can identify the next 10x companies. Their radical thesis is to remove humans entirely from investment decisions.

In this article, I’ll take a look at what the future might hold for “AI-native” through some numbers, industry tensions, and my own take.

Trends

Let's look at some numbers and major changes in the VC circle:

• Two Y Combinator Partners Are Leaving To Start A New AI-native Series A Fund

• 92% of VC firms now use AI to optimize due diligence, manage relationships, and discover new opportunities

• 68% of VC firms believe AI can significantly improve their investment decision accuracy

Since the release of Google's or OpenAI's deep research feature, everyone has become very familiar with how AI accelerates traditional processes in market analysis. We also have Harvey to help us review legal documents during due diligence. Here is a breakdown of how AI is affecting the pipeline:

Sourcing

Crunchbase has always been the most well-known startup database, providing VCs with the most up-to-date market information. Other well-known services, such as PitchBook, Tracxn, Harmonic, and others, can help VCs quickly identify potential startups for their own deal flow. From my perspective, AI-powered scraping of news sites benefits those services more than VCs. Do you need an internal system for auto-scraping deals? I would rather say no. To manage the deals, you do require a database, but not all of it. That is the true recommendation based on my own experience with designing a system that scrapes thousands of startups.

What I believe is the most important value in the sourcing phase is an agentic scout who can identify potential founders at the right time. It may not be the best time to start a journey if someone is talented but lacks ideas or co-founders to join their cool/weird projects in the garage, even if ML algorithms are able to predict that they will be the next Alexandr Wang based on their education or personality. The most difficult thing for an AI scout is immediately detecting the right moment when they have ideas, whether it's a tweet on X or a flash on Hacker News that gets flamed.

Screening

Most often, investors used to say, "We have tons of decks to review every month." But now, 42% of firms use AI for first-pass screening. Automated screening of thousands of pitch decks, enabling instant follow-up questions—no more manual reviews of 1K+ decks.

Time is limited, and investors always want to effectively prioritize deals without missing any opportunities. In this stage, you can simply use AI to summarize the form submitted by the founders, or NotebookLM to extract content from the GDrive slides. If you have an internal engineering team, a reasoning model or vector database can be used to score and rank startups based on your thesis.

Due Diligence

Every VC has a unique method for doing DD. McKinsey sees gen-AI as a “new diligence chapter” for PE/VC. If you are an early-stage investor or an angel investor, there is limited information or numbers available to help you calculate the financial projection or exit strategy. Then AI Agents for Market/Comp Scan will be more useful to you. If you're a multi-stage/late-stage VC, anomaly detection on data rooms could be a good way to discover informal information or assess risk.

Another gray area that I think everyone in this field tries not to bring up is founders' due diligence. Before the internet and social media era, we verified a founder's credibility by asking insiders. Then it is easier to get to know someone by scrolling through their online posts, essays, or videos. In theory and practice, it is very simple to use AI to accomplish this, but is it ethical? Once, I used the search API to grab a harassment event for a young founder who was an alumnus of a well-known accelerator. I was thinking about how we verify the accuracy of the sources. Honestly, this problem isn't new with AI; it existed before. How do we accurately judge and filter information sources? What AI helps us with is quickly gathering and consuming information, but the judgment is still up to human beings.

Tensions

Initially, this LinkedIn post inspired me to write this article. What caught my attention was the phrase "completely remove humans from investment decisions." I might have a basic understanding of how quants work in the public market, especially since quantitative finance was my dream field in high school. However, what I'm trying to figure out is, can it also work in the private market? Which will win the market in the next few decades? Machine-First or Human-First Methodology?

QuantumLight

QuantumLight isn't the only VC firm I've seen using AI or ML to make data-driven investments. I noticed a few agentic micro-VCs that use GenAI to select and manage investments more quickly. QL claim that the AI model (“Aleph”) analyzes over 10 billion data points and tracks more than 700,000 venture-backed companies worldwide. All 17 investments to date have been sourced and selected solely by the AI model, with no human overruling or gut-feel input, effectively eliminating human judgment from the investment process and outperforming top investors by 2X.

When you deep dive into the strategies, they are mainly targeting Series B and C rounds—companies that have reached product-market fit and are seeking growth capital. They also provide operating playbooks and hands-on support with proven systems for scaling.

Y Combinator

YC is undoubtedly the firm that sees the most talented young successful entrepreneurs in the SaaS eras. The latest RFS of Y Combinator from Jared Friedman is "Full-Stack AI Companies." Though aimed at startups, I believe it also applies to VCs. His advice is to either create day-one AI companies or sell AI tools that support current industries.

Still, when it comes to how they actually select startups, YC’s focus is still very much on founders. As Jessica pointed out in her public talk, the nine traits of super-successful unicorns all come back to founder personality—something you can’t easily capture with a model.

a16z

In this episode, Mark explained how the investors' patterns can be traced back to the old whaling industry. These investors would sit in coffee houses or pubs, and ship captains would pitch their projects, detailing plans to buy a ship, staff a crew, and go out to catch a whale. The financiers would then decide whether to back the captain by providing money for the ship and crew. If the ship didn't return, they lost all their money. If it returned with a whale, the captain and crew would receive 20% of the whale as "carried interest."

He concludes that any part of the economy involving high-risk, high-return endeavors, where there are more aspirants than funding, and which requires a multifaceted skill set, is "art" rather than "science." This "art" involves human relationships, psychological analysis, and navigating unpredictable outcomes.

Everyone was talking about how software is eating the world and disrupting the venture capital (VC) industry, but it hasn't had as much of an impact as we anticipated. How will AI affect VCs? Andreessen acknowledges that it hasn't happened yet, but it "could" still happen in the future.

Insights

Most of the time, I work closely with founders, and they often say my superpower is connecting them with the people they need, especially engineers, whether as co-founders or founding team members. Maybe this comes from my academic background, having researched mathematical models in psychophysics during graduate school. Or perhaps it’s my constant curiosity and habit of exploring my diverse network for the latest and most exciting technologies.

Because of this, I truly understand both sides of the table as mentioned above. Many people see “data” and “relationships” as being on opposite ends of the spectrum—one cold and structured, the other irrational and subjective. However, in network-driven industries, particularly venture capital, these two elements are not only compatible but also deeply intertwined.

Every relationship, every interaction, is a wellspring of valuable, unstructured data: from conversations and meetings to shared goals and ongoing needs. For example, I keep a personal network database, tracking my interactions with founders, mentees, and peers in overlapping fields. I log their goals, current status, and challenges. Whenever a need arises, I use an AI agent I designed to capture these updates and recommend tailored resources or connections.

This isn’t unique to me; more investors are recognizing that harnessing both relationships and data is key to adding value. As more top VCs shift to hands-on support, assisting founders in building their businesses, leveraging this combination of human connection and actionable information is rapidly becoming a key investor advantage.

Unlike before, this shift won’t just tweak the edges; it’s set to transform the entire landscape for investors, whether institutional, angel, or family office. Below are my thoughts on how you can adapt to the changes ahead:

Centralized Information Hub

Whatever your deal sourcing process or deal flow setup, having a central information hub is key in industries where information gaps make all the difference. It is not about using fancy tools or models; rather, it is about designing the relationship infrastructure based on your thesis and culture.

You don't need to fine-tune open-source models that run locally; the most valuable thing is that with reasoning models and unique prompts, you can identify noise and critical signals in a dynamic market, companies' moats, and founders' hidden superpowers. Find someone who is familiar with aggregating information from many channels, such as emails, Airtable, X, hacker news, and so on.

Another suggestion: when it comes to automation, it’s better to start with backend processes and address the frontend later. The first touchpoint is crucial for founders, and automating it too early—like using an AI screening interview—can make founders feel like they aren’t even worth a few minutes of real interaction. I recently heard from a founder who experienced this, and he told me it was discouraging. As someone who has always advocated for automation, I realized that, if we really want to automate, we should prioritize the back office before automating any founder-facing moments.

It’s about balance and mutual respect. For example, Bumble accelerated female engagement by providing template questions, making the process easier without losing the personal touch. I believe a similar philosophy could be applied to the automation pipeline: use technology to enhance rather than replace the human connection, particularly in those early, trust-building moments.

Human-AI Judgment System

One of my favorite books, The Art of Loving, offers a profound description of intuition and art. The book suggests that intuition and art are often built upon a deep understanding and extensive experience in a particular field. In other words, they are not as irrational or lacking in structure as many people assume. I think this is also why Mark describes investing in startups as an art.

In Naval’s sharing on how to angel invest (part 1, part 2), he uses the metaphor that investing at the seed stage is akin to playing the lottery. However, good judgment allows an investor to “get some of the winning numbers in advance,” thereby increasing their odds of success.

How do we design a Human-AI collaboration system to co-judge deals? If a model can focus on broad-scope judgment, objectively synthesizing and analogizing vast amounts of information. So, what is the role of humans? And which human abilities will still surpass AI in such a system?

Once, I was analyzing a highly competitive market, where dozens of startups were building almost identical solutions. Neither I nor any LLMs could truly differentiate them based solely on product websites; they all communicated the same value propositions. That’s when I realized the real distinction lies in dimensions you can’t capture from product pages alone. By digging into founders’ social media, I realized that with enough “data,” you can start to infer how a founder makes decisions inside the company, their communication style, and the kind of leadership culture they create. All of which ultimately shape what the company will become. It’s almost like seeing higher dimensions in a vector space beyond what’s on the surface. I think this might be what legendary investors refer to as “gut feeling.

It’s important to strike a balance: we don’t want to miss great opportunities because of our own biases during first meetings, but we also don’t want to overvalue deals submitted online simply due to outstanding founder backgrounds or exaggerated traction. I don’t have the specific guideline for this part, since so much depends on how you and your firm articulate your judgment. My only suggestion is to let the “silent intelligence” of AI have a voice in the decision-making process.

Adaptive Investment Architecture

In today’s environment, with so much capital and all the hype around AI, the landscape is shifting. Foundation-layer AI companies still need to burn a lot of money. But many new application-layer startups are reaching impressive ARR milestones much faster. Some even claim they have never needed to raise money from investors. This new dynamic is making competition among VCs fiercer than ever.

I've noticed an interesting trend in recent months: an increasing number of investors are actively seeking out AI startups, and some VCs are even pitching on stage about how they can assist founders. It’s almost as if the roles have flipped, with investors now trying to win over startups rather than the other way around. Also, more VCs have begun taking a hands-on approach, working closely with founders to help them build products. The line between accelerators and early-stage VCs is becoming increasingly blurred.

Early-stage VCs can benefit from building modular support systems—as demonstrated by YC’s extensive content library, video resources, and co-founder matching. Automated recommendation engines can help portfolio founders address knowledge or resource gaps, while 1-on-1 time is best reserved for high-level strategic work. This maximizes efficiency and impact across the portfolio.

When we take a look at the mega funds, it’s a whole different game. Instead of just writing checks and waiting for exits, they’re now launching and acquiring companies, then rewiring them with AI. In recent years, top Silicon Valley VCs have gone through a significant transformation, increasingly adopting private equity–style strategies. Firms like Andreessen Horowitz have registered as investment advisors and expanded into wealth management and company restructurings; General Catalyst now acquires and operates entire businesses, sometimes not even calling itself a VC; and others like Thrive Capital and Sequoia Capital have built new vehicles and evergreen models focused on long-term ownership and AI-driven value creation.

The VC world is seeing clear barbell polarization: mega funds are absorbing most of the LP allocations, while micro funds increasingly rely on HNWIs to survive. At the same time, data shows that top small, operator-led funds deliver IRRs 25% above established firms. Although some reports predict that 30–50% of small VCs could disappear.

Looking ahead, we can expect to see a rise in vertical AI agents, and micro funds are well-positioned to evolve into specialist funds that leverage the unique domain expertise of their GPs. To operate more efficiently, these specialist GPs can develop their own AI twins—training AI models on their accumulated knowledge to support more founders in hitting key milestones before bringing in mega funds for follow-on rounds. In this way, micro funds help mega funds by de-risking early-stage investments, ensure startups hit real revenue milestones, avoiding the trap of “vibe revenue” highlighted by Pat Grady at Sequoia AI Ascent 2025.

Fresh Lens

It’s not just about VCs, change is coming for LPs, PE, and the entire private capital ecosystem. As AI agents advance (while they may not be as smart as people imagine yet, today’s SOTA models are already far ahead of just two or three years ago), the agent layer itself could be restructured, fundamentally shifting the relationship between LPs and GPs in the decades ahead.

While fully bypassing professional agents is still challenging, it’s clear that LPs’ direct investment capabilities are strengthening. Large LPs can now leverage AI systems to enhance direct investment, reducing reliance on intermediaries—especially in mid-to-late stage deals where data is more transparent.

For LPs focused on financial performance, Sheikh Saoud Salem Al-Sabah, Managing Director of Kuwait Investment Authority (KIA), highlighted during the panel that LPs and family offices are shifting back toward active management. He advised all LPs and co-investing FOs to ask their GPs: “Show us the entry valuation multiple for every investment.” If the answer isn’t clear, it’s a red flag. The key is that sophisticated LPs and FOs need to be willing to look through, using data transparency to push GPs to improve. AI-native VCs can support this shift by making chain-of-thought and decision-making data transparent, enabling continuous learning and judgement optimization.

For LPs whose ambitions go beyond asset allocation and extend to enterprise transformation, AI-native VCs are becoming the neural system for both new and established companies. What you need is someone with a telescope—someone who can keep their eyes on the big-picture vision and make sure it aligns with your goals—while AI acts as the microscope, diving deep into the details that really matter.

The future belongs to those who can look both far and deep, connecting vision with detail. If you’re building the future of investment, seek the space where human insight, intelligent automation, and new possibilities converge.