The AI Revolution Nobody Is Noticing

The AI Revolution: Unprecedented Growth Rates for Startups

In a special episode of "The Light Cone" filmed in Sonoma during a 300-person retreat for top AI founders, hosts Garry Tan, Harj Taggar, Daniel Gross, and special guest Paul Buchheit (creator of Gmail and YC partner) discussed the remarkable impact of AI on startup growth trajectories.

 The New Normal: Extraordinary Growth Rates

The conversation began with a reflection on what used to be considered exceptional growth. "10% week-on-week growth is an amazing metric to hit," noted Buchheit, explaining that previously only the top one or two companies in a YC batch might achieve such numbers. 

However, since summer 2023, something remarkable has happened. YC batches in aggregate averaged 10% week-on-week growth during their 12-week programs. This wasn't limited to standout companies—it represented the overall batch performance.

Harj Taggar highlighted one particularly impressive example: "One company went from zero to $12 million in 12 months. I'd never seen any growth like that." Daniel Gross added that this accelerated pace is becoming the new minimum expectation, with companies hitting $1 million ARR within six months of their batch, rather than the previously aspirational 12-18 month timeline.

The ambition levels have dramatically increased, with founders now setting goals like growing from $1 million to $20 million in a single year—targets that would have seemed absurd just a few years ago.

Why Is This Happening?

Aaron Levie, CEO of Box, offered an interesting perspective. Throughout previous enterprise software cycles (cloud, mobile), there were always decision-makers saying "no" to new technologies. With AI, however, something unprecedented is happening: everyone is saying "yes." There's unparalleled demand for AI solutions.

The team noted that companies experiencing these extraordinary growth rates share a common profile—they're primarily selling AI agents to businesses. Enterprises face enormous internal pressure to adopt AI, creating ready-made demand that founders can tap into.

"It goes back to our fundamental advice: make something people want," Buchheit observed. "In this case, the demand is already there. You just have to show up with a product that works."

 Building Products That Work

While the demand exists, creating effective AI products remains challenging. Many businesses seek software that can genuinely perform human-level work—essentially AI services rather than traditional software tools. This proves technically difficult, which creates opportunity.

As Garry Tan explained, "A lot of technical CEOs who aren't necessarily the strongest at sales are able to win big enterprise contracts now because, although there's 10 or 15 other companies competing for the same contract, it's very hard to build a product. Just building the thing that actually does the work well is enough to win these huge deals."

Founders are inventing new patterns for building AI products, finding innovative ways to make large language models behave predictably and accurately. This includes sophisticated prompting techniques and rigorous testing methodologies.

 The Rise of Evaluation Sets

One surprising trend highlighted during the retreat was the focus on evaluation and testing. A founder building an AI agent remarked that the most valuable asset his company developed wasn't the codebase but the evaluation set—a gold-standard collection of labeled data defining correct AI responses.

This represents a mental shift in how startups think about their assets. As Buchheit explained, "The whole Chat GPT wrapper meme is wrong. The model is changing very quickly—there are clearly five or more AI labs at the frontier. The thing nobody has that is actually hard to get is the eval set and the prompting."

The group emphasized that in an AI world, the two most important elements are:

1. Agency and taste in prompting—knowing what to tell the agent to do

2. Evaluation—the ability to judge if the output is good, beautiful, and useful

## Changing Design Workflows

The team shared an interesting anecdote about how AI is transforming design processes. One founder reported that their designer had stopped using Figma mockups entirely. Instead, the designer was designing with Claude (an AI assistant), going directly from text prompts to JavaScript code that produced results as tasteful and effective as traditional visual mockups would have been.

The pattern seems clear: whoever can iterate fastest wins, and AI is an extraordinary tool for rapid iteration.

 The Future of Jobs and Economy

The conversation turned to concerns about job displacement. Buchheit referenced Milton Friedman's famous anecdote about workers digging a canal with shovels instead of machinery because it was a "jobs program." Friedman responded, "If it's a jobs program, you should give them spoons, not shovels."

The team suggested this provides a useful mental model for thinking about AI and work. Rather than restricting technological progress (giving workers "spoons" instead of "shovels"), AI represents an opportunity to create dramatically more wealth and productivity.

Buchheit proposed a framework of "machine money" versus "human money." The goal should be to use technology to create massive deflation for machine-produced goods and services, driving prices toward zero. This could make essentials like quality healthcare universally accessible.

Meanwhile, "human money" would be allocated to things we uniquely value from other humans—personal time, creativity, and connection.

From Fear to Opportunity: A Decade of Progress

Reflecting on the past decade of AI development, Sam Altman shared how perspectives have evolved. Around 2015, there were serious concerns that AI reinforcement learning might lead to dangerous objective functions—the "paperclip maximizer" fear that an AI would pursue its goals at the expense of humanity.

What's happened instead is the discovery that intelligence at its core is about predicting what comes next. This has enabled the creation of AI systems that don't have a drive to survive or compete with humans.

The team expressed optimism that we're on a positive timeline. Ten years ago, leading AI research was concentrated at Google, which prompted the creation of OpenAI as a "moonshot" nonprofit to ensure open competition. Today, we have at least six major foundation models competing in an open market, preserving freedom through choice and competition.

 Changing Search Behavior

The conversation touched on how AI is already reshaping internet usage patterns. Early adopters—typically software engineers and technical professionals—are increasingly using ChatGPT or Perplexity for information searches rather than Google.

"Google's starting to have that weird Legacy website vibe to it," noted one participant. Stack Overflow traffic has reportedly dropped 60% this year, a trend that began in 2022 before ChatGPT's release, initially driven by GitHub Copilot adoption.

New Developer Tools and Productivity

The adoption of AI coding tools shows how quickly the landscape is changing. Cursor, an AI-powered development environment, went from negligible adoption to being used by approximately 80% of YC founders within a single batch cycle.

This is affecting hiring practices, with some founders reporting that they now screen candidates based on their familiarity with code generation tools: "If someone comes in and I ask them if they use Cursor or any code-gen tools and they say no, I can't hire them because they're not going to be as productive as the rest of my team."

The Future of SaaS and Business Models

AI is also transforming how businesses operate internally. Companies like Clara are reportedly replacing SaaS tools and reducing new engineering hires by leveraging AI capabilities.

Another company, Jerry, implemented GPT-4 for customer support, which not only cut their support team and budget by half but transformed the business from burning $10 million annually to becoming profitable with 50% yearly growth.

Buchheit described this as an example of how "AI is creating wealth by making economically viable businesses that weren't possible before."

 The End of Blitzscaling

The retreat revealed a significant shift in startup strategy. Previously, rapid growth meant aggressive hiring and capital raises—a concept formalized as "blitzscaling." Today's fastest-growing companies are achieving remarkable revenue numbers with smaller teams and less external funding.

"Nobody is bragging about becoming a unicorn," Garry noted. "It's all about leverage right now—how much you can do with limited resources because we have these magical tools that give us superhuman leverage."

 Building for an Uncertain Future

A key challenge for founders is building with tools that are evolving rapidly. Discussions at the retreat explored questions like whether retrieval-augmented generation (RAG) would remain necessary as context windows expand.

The most successful startups are those willing to continuously rebuild their technology stacks to incorporate the latest advances. As one participant noted, "The best startups are going to be the ones that can build the fastest, be at the bleeding edge, and reevaluate assumptions about the best approach."

This ability to rapidly adapt and implement cutting-edge tools gives startups a significant advantage over large enterprises, which typically move much more slowly.

Conclusion: The Best Time to Build

The hosts concluded that there has never been a better time to start a company. Twenty years ago, YC was founded on the premise that it was getting easier to build startups—a couple of smart people could create a web application without raising mountains of capital or hiring large teams.

AI has dramatically accelerated this trend, enabling small teams to build multi-million dollar businesses. As Altman summarized, "Technological leverage enables people with ambition and insight to do incredible things."