This video uses Y Combinator statistics and examples of AI-native companies to look at changes in founder composition, how startups grow, product and pricing policy, and how organizations operate. Thanks to AI, small teams can now do much more, but the video points out that this makes the ability to judge what to build, the work of adjusting role boundaries, and the ability to resolve conflicts among team members all the more important. Ultimately, the key is not using AI for its own sake but correctly finding the problems worth solving and effectively mobilizing both people and AI.

1. "AI Does It for You, So Why Aren't You Doing Anything?"

The video opens with a remark from Y Combinator CEO Garry Tan. It's a provocative question: AI can do so much for us and our ability to build products has greatly improved, so why are people still not trying anything?

"AI now does almost everything for us, so why aren't you doing anything? What's your excuse?"

The hosts see this as a natural thing for YC to say. They explain that YC has recently seen changes in what founders and startups look like and in revenue growth, and Garry Tan was summarizing those statistics. Building on this, the video examines whether AI-native startups can actually make money and whether a company can be run with only a small number of people.

2. Fast Revenue Growth and Changing Founders, Seen Through YC Statistics

The YC program runs for about three months, and the companies in a batch continue their businesses after the program ends. The hosts discuss a statistic showing that the median monthly revenue of recent YC batch companies at the end of the program rose from about $8,000 to $20,000. They stress that because this is the median rather than the mean, the figure isn't inflated by a handful of outsized successes.

They also mention a company that reached $1 million in revenue within three months. Cases like this used to sound like legends, but now generating revenue very quickly right from the start of a company is actually happening.

"The legendary stories of startups succeeding fast right out of the gate are actually happening."

Founder composition is changing too. The share of solo founders has grown sharply, and many founders now start alone and bring on a co-founder later. In the past, splitting equity 50/50 between co-founders was common, but as more founders accomplish a lot on their own before bringing in colleagues, equity is increasingly split according to role and contribution.

"There's just so much one person can do on their own now."

The classic co-founder pairing used to be someone good at talking and business paired with someone strong in technology and development. But as AI fills in much of the building and development, a question arises: "Then why do you necessarily need a co-founder?" The talk of YC turning into something like an accelerator specialized for solo founders ties into this change.

3. Founders in Their 40s and "Knowing What to Build"

The video highlights as an important change that founders in their 40s are performing better than expected. The explanation is that getting older doesn't automatically put you at a disadvantage in starting a company; long experience can help you know better which problems are worth solving.

"The most important skill isn't knowing what you can build, but knowing what you should build."

With experience, it can become clear what you've always wanted to try and which attempts you regret putting off. That kind of conviction also helps when using AI, because getting good results from AI requires not just issuing instructions but providing context that clearly explains the situation and the goal. The point is that someone who has led a team is more likely to understand what needs to be done and what the goal of the result is.

So in the AI era, experience in defining problems precisely and setting direction may become as important as the ability to handle technology directly.

4. From Software to Hard Tech: "From Bits to Atoms"

Next, under the phrase "from bits to atoms," the video covers the growth of hard-tech fields beyond software, such as robotics, manufacturing, defense, and silicon photonics. It mentions that the share of robotics-related startups at YC has grown from about 1% in the past to around 6–7%, and that more founders hold PhDs.

The founders YC used to spotlight fit the image of young students who had dropped out of Stanford or MIT. Now, more founders with strong technical foundations are emerging to tackle problems that require complex technology and specialized knowledge. Fields like robotics, which touch the physical world and are hard to implement, are also becoming a stage for startups.

Hardware businesses have traditionally been seen as requiring large upfront capital and many people, and as hard to scale quickly the way software companies do. But as AI helps with many tasks, initial entry costs and staffing needs can shrink. The video even mentions a traditional metal manufacturing company growing as fast as a software company.

That said, the hosts don't give a definitive answer as to why hard-tech startups in particular are increasing in the AI era. They wonder whether it's because software is easy to copy and thus harder to differentiate, while hardware is hard to enter, but they don't draw firm conclusions from the statistics alone.

5. Quietly Growing Data Companies and AI Training Environments

It's easy to assume startups usually get lots of attention in the press and on social media, but the video notes that there are also companies growing quietly while earning far more than their fame suggests. In particular, companies that supply data needed for AI training and companies that build reinforcement learning environments are multiplying.

"There are companies nobody really knows about, and then one day you find out they're making a ton of money."

Scale AI is mentioned as an early example of this kind of data company. A business area that used to have no clear category has now settled into a category called "data vendors," and companies supplying the data and environments needed to train and improve AI models are forming an industry of their own.

This shows that startup opportunities don't have to be found only in highly visible consumer apps or famous products. Businesses that supply the foundations that make the AI industry possible can also be large markets.

6. Boldly Shutting Down a Product and Finding Founder-Market Fit

The next example is Lightfield, featured by a16z. The company originally ran Tome, an AI service for creating presentation decks, which reached 25 million users. But internally, the team wasn't satisfied with the product, and in the end they shut the service down.

They judged that even with many users, if the team that built the product didn't believe in its actual quality, users could leave in the long run. Because AI has made building new products easier, it has become possible to change direction toward a problem you can build better for and genuinely want to solve, rather than clinging to a product just because it gets a response.

"If it's a product we don't really love, won't those people eventually leave too?"

Here the hosts explain the difference between PMF (Product-Market Fit) and Founder-Market Fit. PMF refers to a state in which a product is genuinely needed by and fits well with the market. Founder-Market Fit refers to whether the founder's experience, interests, and strengths fit that market and problem.

One approach is to pick a good market first, build and launch a minimum viable product, and keep pivoting if there's no response. Another is to start in a field the founder has long cared about and built expertise in. For example, a PhD who has researched robotics for years can build a company grounded in a deep understanding of and passion for robotics.

As AI lowers the barriers to entering specialized fields, founders have a bigger opportunity to build products directly in fields they already know and love. The argument is that rather than forcing yourself into a market that seems lucrative, choosing a problem you understand well and sincerely want to improve lets you create a better product and a more compelling story.

7. AI Back Office and Pricing Gone Awry

Lightfield later built back-office tools that help run a company. In the past, people manually entered purchase records and transaction information into Excel or fixed tables to manage them. But if AI can gather data from multiple sources and automatically create the ledger, the very approach of people first defining a table's structure and then filling it in can change.

The video expresses this idea as "intelligence over schema." Here, a schema means a framework people set up in advance, like a table or format for organizing data. The point is that AI can now take over the intelligent categorization and organization that previously required expensive people and systems.

But separately from building the product, how to price it became a big problem. If you charge by number of employees, the cost of providing the service can spike when customers use it more than expected. Conversely, if you charge by usage, customers may not use the product enough because they worry about cost. In the real case described, the company nearly went bankrupt because its pricing structure couldn't handle the usage.

"Users used it so much that costs exploded, but when we charged by usage, nobody used it."

With AI products, usage can vary between users far more than with traditional software. The video gives the example of token usage differing by up to 10,000x between the hosts, and points out that a difference in usage doesn't directly mean a difference in productivity. Existing services that added AI features, like Slack and Notion, also have to figure out how to connect usage, value, and cost.

Ultimately, in the AI era, usage-based pricing, subscription fees, the value customers receive, and the cost of running the service can fall out of alignment. That it has become hard to set prices the traditional SaaS way is a practical problem facing both AI startups and their customers.

8. AI-Native Organizations Where Role Boundaries Blur

Lightfield's way of working differs from traditional job divisions. They first lay out the work to be done and prioritize it by importance, and then team members divide up whatever is needed, whether development, customer support, or product improvement. It's an approach that tries to blur the "this is my job, that's your job" boundary as much as possible.

"We completely eliminated the boundary between my work and your work."

The video explains that this doesn't simply mean everyone does everything haphazardly. Even if each person has their own expertise, the idea is not to ignore problems in other areas because of job titles, but to handle what's needed directly or create a draft and collaborate.

It also discusses a case where a product manager used AI to create a design draft and asked a designer to review it. This drew pushback along the lines of "Why are you doing my job?", but the hosts say there's no need to see the issue simply as encroaching on someone's territory. If the draft's quality fell short, that may be a problem with the output and the communication. Conversely, if someone actually produced a good draft, there's no need to block it just because their role is different.

However, because handing over a draft can feel like pressure to "now finish this quickly," paying attention to the intent of collaboration and how it's communicated remains important. Even though AI has increased the chance that one person can perform several roles well, not everyone can do everything well. Focusing on what you're good at and sharing the hard parts with others is also necessary.

This leads to a debate about whether eliminating roles outright is the answer. Short tasks can be divided flexibly among several people, but work that runs for weeks or months may need an owner and accountability — in other words, ownership. In a small organization where every member treats the company's success as their own, boundaries may matter less, but as headcount grows, aligning everyone in the same direction becomes harder.

9. Small Teams, Customer Acquisition, and the Survival of Traditional SaaS

The hosts say their own team also has few core members and operates by sharing a variety of work rather than strictly dividing roles. There's also the view that even as they grow, they might prefer forming small teams around products or business units rather than expanding one large organization. Companies like Telegram that deliberately keep headcount low are mentioned as well.

In the Lightfield case, it's noted that the people responsible for customer acquisition, combining sales and marketing, numbered only three. This means significant investment and growth can be achieved with a small team. In acquiring customers, too, a strategy of identifying early the companies likely to grow quickly may matter more than chasing only the customers generating the most revenue today.

Opinions differ on whether traditional SaaS companies will be replaced by AI-native ones. One side believes old-style software will disappear and AI-based alternatives will take its place; the other believes incumbents can survive if they adapt well to the change. The hosts' summary is fairly realistic.

"Traditional SaaS will survive if it adapts well, and disappear if it doesn't."

10. The Pace of AI Progress, Role Conflicts, and the People Needed Going Forward

There is uncertainty about whether AI will keep advancing at its current pace. Still, the hosts think AI's impact tends to be underestimated. Just a few years ago, there were skeptical views that large language models would remain at the level of answering questions, but the situation changed dramatically as AI evolved into agents that carry out real tasks.

"Even if models don't get any better from here, just actually applying the technology that's already out will change a lot."

Even as new models appear quickly, it takes time for the technology to be used in companies and everyday life. But since the models themselves keep advancing, the argument is that big changes can happen even from technology that hasn't yet been fully applied. Given how many companies are now worried about GPU shortages, they say the decision to invest early in AI compute may also have been right.

In this transition, the most expensive capability is ultimately deciding what to build. Even if it means throwing away what you built before, you have to find the next product customers truly need. As AI lowers building costs and reduces headcount needs, founders must focus on knowing what to build and implementing it well.

Within organizations, however, role conflicts can grow. If a designer uses AI to build in a single day a feature a developer said was difficult, it can be seen as demonstrating a new possibility, but it can also be taken as a threat to the existing owner's territory. Even if you know that blurring role boundaries can be productive, not every member embraces that change comfortably.

"People may be getting angry because everyone feels threatened."

Especially from an employee's perspective, a change in one's role may not be a simple work adjustment but a matter of livelihood and job security. Environments like the US, where workforce adjustments are relatively easy, and environments like Korea, where organizational change and layoffs are not easy, may call for different responses. A realistic concern emerges: if attempts to retrain existing members to use AI don't work out, yet they can't easily be replaced either, the conflict can drag on for a long time.

That doesn't mean leaving an organization to start a company is always the answer. If a company that looks inefficient from the inside keeps surviving, it may have strengths not easily visible from the outside. So people who can solve organizational problems will be valuable talent, but they note that actually reconciling people's interests is not easy.

Finally, the video poses a question to founders and employees alike.

"Do you know what you should build?"

"If you know that, you'll also know what to do — but before that, you have to develop that perspective."

They also say there's no settled answer to what kind of people companies should hire in the AI era. In the past, specialized roles were clear: if you were good at development, you were hired as a developer; if you were good at design, you were hired as a designer. Now, people who cross roles to solve problems may be needed, but that means companies must keep asking whom to hire, and individuals must keep asking what capabilities to develop.

11. AI-Native Companies Finding Answers Through Experimentation

The video concludes that, looking at statistics and examples together, changes in startups are really underway. Solo founders and middle-aged and PhD founders are increasing, cases of generating revenue in a short time are appearing, and companies are emerging that boldly shut down products and pivot. In organizations, role boundaries are weakening, and acquiring customers with small teams is becoming possible.

Even so, these changes don't immediately amount to a right answer applicable to every organization. Pricing is still hard, ownership and ways of collaborating vary by situation, and conflicts can arise over roles that AI has changed. The hosts add that while their thinking has gradually come together through covering AI-native companies several times and putting the ideas into practice themselves, they don't already know the answers.

"When you actually try it, your thinking changes little by little, and what's clear gradually starts to come into view."

Ultimately, the core of the video is not that AI solves every problem. It's that finding what to build, applying the changed technology to real products and organizations, and dealing with the conflicts and cost problems that arise along the way are the tasks that remain for founders and organizations in the AI era.