From Dot-Com to AI: Lessons for Avoiding Past Tech Pitfalls

During the dot-com boom, appending “.com” to a company’s name could skyrocket its stock price, even without customers, revenue, or a viable business model. Today, the same frenzy surrounds “AI,” with companies eagerly adopting the label to capitalize on the hype.

Businesses are rushing to integrate “AI” into their branding, product descriptions, and domain names. According to Domain Name Stat, “.ai” domain registrations jumped 77.1% year-over-year in 2024, as startups and established firms alike scramble to align with artificial intelligence, regardless of genuine AI capabilities.

The late 1990s taught us that leveraging cutting-edge technology isn’t enough. The dot-com survivors weren’t chasing trends—they were addressing real needs and scaling thoughtfully.

AI holds similar promise to transform industries, but success won’t come from superficial branding. It will come from companies that cut through the noise and prioritize impact.

The key? Begin modestly, identify a niche, and scale strategically.

Start Small: Nail Your Niche Before Expanding

A major dot-com misstep was scaling too quickly—a lesson today’s AI innovators must heed.

Consider eBay. It started as a simple auction platform for collectibles, like Pez dispensers, solving a specific problem for hobbyists who couldn’t connect offline. Only after mastering this niche did eBay expand into electronics, fashion, and beyond.

Contrast this with Webvan, which aimed to overhaul grocery shopping with online orders and rapid delivery across multiple cities. It burned through millions on warehouses and logistics before proving demand, collapsing under its own ambition.

The takeaway: Focus on a precise user need and dominate a narrow segment before expanding.

For AI developers, this means avoiding the trap of building an “AI for all.” For instance, a generative AI tool for data analysis must target a specific group—say, product managers, designers, or data scientists. Are you serving SQL novices or experienced analysts? Each group has unique needs and workflows.

By focusing on a defined audience—like technical product managers with limited SQL skills needing quick insights—you can deeply understand their needs, refine the experience, and create something essential. Only then should you expand to related users or features. In the race to build enduring AI products, the winners will serve a specific audience exceptionally well, not everyone at once.

Secure Your Data Advantage: Build Lasting Defensibility

Starting small helps achieve product-market fit, but sustaining success requires defensibility—especially through proprietary data.

Dot-com survivors didn’t just attract users; they amassed unique data. Amazon, for instance, didn’t stop at selling books. It used purchase and browsing data to refine recommendations, then leveraged regional order patterns to optimize logistics, paving the way for Prime’s unmatched two-day delivery.

Google took a similar approach. Every search, click, and correction fueled better results and, later, ads, creating a feedback loop that strengthened its edge.

For AI builders, the lesson is clear: Long-term success hinges on proprietary data loops that enhance products over time.

Anyone can fine-tune an open-source large language model or access an API, but high-value, real-world user data is harder to replicate.

AI product builders should ask early:

• What unique data will our product capture from user interactions?
• How can we create feedback loops to continuously improve?
• Can we ethically and securely collect domain-specific data competitors can’t access?

Take Duolingo. With GPT-4, features like “Explain My Answer” and AI role-play generate rich user data, capturing not just responses but how learners think and interact. This data refines the experience, creating a competitive edge.

In the AI era, proprietary data is your enduring advantage. Companies that design products to learn from unique data will lead the pack.

Conclusion: A Marathon, Not a Sprint

The dot-com era proved that hype is fleeting, but fundamentals last. The AI boom follows the same pattern. Success won’t come from chasing trends but from solving real problems, scaling deliberately, and building defensible advantages.

The future belongs to AI builders who treat it as a marathon, with the discipline to stay the course.

Kailiang Fu is an AI product manager at Uber.