At YC W26 Demo Day, 199 companies gathered. Watching a pitch event of this scale reveals some clear patterns.

First, what surprised me most is the reality that AI is no longer just a category, but has become infrastructure. About 60% of the entire batch are AI-native, and an additional 26% are AI-enabled. In other words, 86% of the companies are centered around AI. But what's important is not just "we use AI," but what has become possible because of it.

The most interesting shift is the disappearance of the co-pilot concept. Last year's showcase showed about 4% of companies emphasizing a co-pilot framework, but this year it's only 1%. Instead, what’s emerging are AI agents claiming complete replacement. The mainstream pitch is, "Replace this high-paying job entirely and price it as part of the salary."

In terms of business models, 87% are B2B companies. Only 7% target consumers. This suggests the market is mature. That’s why the characteristics of fast-growing companies are clearly visible.

What do the companies that generate revenue the fastest have in common? They are founders who sell directly to their former employers or industry networks. Of the top 15 companies, 60% acquired their first customers through founder networks or YC networks. In other words, distribution channels weren’t an afterthought—they existed from the start.

Some founders say they "built the product after having 50 business conversations." They deeply understand customer needs before redesigning their products. Conversely, struggling companies tend to build first and then think, "How do we sell this?"

The importance of data is also striking. LegalOS achieved a 100% approval rate from 12,000 visa application data points. Every interaction with customers leads to product improvements. Without this, it’s just a tool.

The revival of hardware companies was impressive. 18% handle physical products, including robots and space technology. Companies founded by alumni of SpaceX and Tesla stand out. GRU Space plans to build the first hotel on the moon by 2032 and has received a $500 million letter of intent.

The most to avoid are un-differentiated agent infrastructure. Eight to ten companies are building monitoring, testing, and compression functions, but these will likely be natively integrated by foundational model providers. Also risky are AI-native services without data advantage. They generate revenue quickly but have the lowest defensibility, and core technologies can be replicated within weeks.

Founders’ backgrounds are also distinctive. About 60% are immigrants or foreigners, 86% are men, and 14% women. Many participants come from Stanford, MIT, Berkeley, etc., but the most successful are those with deep industry expertise. For example, a dentist developing surgical AI, or an aircraft maintenance supervisor creating machine tools. Success often comes from focusing on industries where you sell in a quiet, high-paying niche rather than at a cocktail party.

There are notable gaps in the landscape. No education companies, no government tech firms, and no mental health or fitness companies. Paradoxically, these are areas with the potential for the greatest future returns. Historically, the fields with the least funding have produced outsized returns.

Finally, five common traits of rapidly growing companies: they sell results, not tools; they charge from day one; their customers are in urgent situations, not just curious; founders build customer relationships before the product exists; and their MVPs are surprisingly simple.

The gap between the "build first, learn later" phase and the "build and expect" phase will likely be the main cause of most failures in these projects. That’s why choosing the right market and problem is the most critical initial decision.