The AI boom of 2023-2025 was a whirlwind of unprecedented innovation and, frankly, a lot of hype. The key lessons I learned from investing in and mentoring dozens of AI startups are that foundational models are a commodity, the real value lies in proprietary data and vertical-specific applications, and the human-in-the-loop approach remains the most defensible moat.
The End of Moore's Law for Foundational Models
One of the first things that became apparent during the AI boom of 2023-2025 was that the raw performance of large language models (LLMs) was hitting a plateau. While models got bigger and more expensive to train, the incremental gains in general capabilities started to diminish. This signaled a shift in the world; access to a powerful foundational model was no longer a competitive advantage. It became table stakes.
I saw countless pitch decks that led with "we're building a better LLM." My advice to them was almost always the same: don't. The capital required to compete with the giants is astronomical, and the differentiation is minimal. The real opportunity isn't in building the next GPT-4 or Claude 3, but in using these powerful tools to solve specific, painful problems for a well-defined customer. The lessons from the ai boom of 2023-2025 have shown that the value accrues to those who can build a unique application layer on top of the commodity infrastructure.
This is why I started looking for companies that had a deep understanding of a particular industry's workflow. For example, a startup using AI to automate legal contract review or another one using it to diagnose rare diseases from medical scans. These companies weren't trying to boil the ocean; they were using the powerful new tools to create a 10x better solution for a niche they understood deeply. For more on this, you can read my thoughts on how to evaluate AI startups.
Proprietary Data: The Only True Moat
As foundational models became commoditized, the value of proprietary data skyrocketed. The companies that won were not the ones with the fanciest algorithms, but the ones with unique, high-quality datasets that no one else could replicate. This data became the fuel that allowed them to fine-tune commodity models for specific tasks, creating a level of performance and accuracy that was impossible to achieve with generic models alone.
Think about it: if anyone can access the same powerful AI, how do you build a defensible business? The answer is data. A company that has been collecting and structuring industry-specific data for years has a massive head start. This data flywheel—where more users lead to more data, which improves the product, which attracts more users—is one of the most powerful competitive advantages in the AI era. The the ai boom of 2023-2025 insights clearly point to data as the key differentiator.
Key Insight: Don't just chase the latest and greatest AI model. Instead, focus on building a unique and valuable dataset. That's your real intellectual property and your most defensible asset in the long run.
Vertical AI: Solving Real-World Problems
The most successful AI companies I've seen are not horizontal platforms trying to be everything to everyone. They are vertical-specific solutions that go deep into a particular industry. This focus allows them to build a product that is perfectly tailored to the needs of their customers and to develop a deep understanding of the nuances of their domain.
Here are some of the characteristics of successful vertical AI companies:
- Deep Domain Expertise: The founders and team have years of experience in the industry they are targeting.
- Workflow Integration: The product is not a standalone tool but is deeply integrated into the existing workflows of its users.
- Clear ROI: The value proposition is crystal clear, and the return on investment is easy to calculate.
- Data Network Effects: The product gets better as more customers use it, creating a powerful data network effect.
From legal tech to healthcare to finance, the opportunities for vertical AI are immense. These are the companies that will create lasting value and transform industries. My advice to founders is to pick a niche you know well and go deep. This is one of the most important what I learned the ai boom of 2023-2025 takeaways.
The Human-in-the-Loop Imperative
Another key lesson from the AI boom is that fully autonomous AI is still a long way off, especially in high-stakes domains. The most effective and defensible AI systems are those that augment human experts, not replace them. This "human-in-the-loop" approach combines the speed and scale of AI with the judgment and intuition of a human expert.
This approach is not a temporary crutch; it's a long-term strategic advantage. By keeping humans in the loop, companies can handle edge cases, ensure quality, and build trust with their customers. It also creates a powerful feedback loop where the human experts are constantly training and improving the AI model. For anyone interested in the future of AI, I recommend reading about the future of generative AI.
I've seen this play out time and again. The companies that tried to build fully automated solutions often struggled with accuracy and reliability. The ones that embraced the human-in-the-loop model were able to deliver a superior product and build a more sustainable business. It’s a crucial part of my strategy when investing in early-stage companies.
Frequently Asked Questions
What was the biggest surprise of the AI boom?
The biggest surprise for me was the sheer speed at which foundational models became commoditized. In early 2023, having a powerful LLM was a huge advantage. By 2025, it was just the starting point. This rapid shift in the field caught many investors and founders by surprise.
What is the most common mistake you saw founders make?
The most common mistake was focusing too much on the technology and not enough on the customer problem. I saw too many pitches that were all about the "cool" AI they were building and not enough about how it would solve a real-world pain point. Technology is a means to an end, not the end itself.
What are you most excited about for the future of AI?
I'm most excited about the potential for vertical AI to transform industries that have been largely untouched by technology. From agriculture to construction to manufacturing, there are so many opportunities to use AI to improve efficiency, reduce waste, and create new products and services. The lessons from the ai boom of 2023-2025 have laid the groundwork for this next wave of innovation.
Final Thoughts
The AI boom of 2023-2025 was a transformative period that will have a lasting impact on the technology industry and beyond. The key lessons are clear: focus on proprietary data, solve real-world problems in specific verticals, and embrace the human-in-the-loop model. For founders and investors who understand these principles, the opportunities are immense. The AI revolution is just getting started, and I, for one, can't wait to see what comes next.