From Dot-Com to Dot-Ei: How can we learn from the last technical transformation (and avoid making the same mistakes)

At the height of the Dot-Com mutation, adding “.com” to the company’s name was sufficient to send its stock price-even if the company did not have customers, revenues or path to profitability. Today, history repeats itself. “Com “for” artificial intelligence “, and the story looks frighteningly familiar.

Companies are racing to spray “artificial intelligence” on the stadium floors, product descriptions and domain names, in the hope of noise riding. As mentioned before The basic field nameThe registration of domains.

The late nineties made it clear: Use Horching technique Not enough. The companies that survived the collapse of the Dot-Com were not chasing the noise-they solved real problems and expanded their scope.

Amnesty International does not differ. It will reshape industries, but the winners will not be those who slap “artificial intelligence” on a broken page – they will be the ones who penetrate the noise and focus on what matters.

The first steps? Start a small, search for your wedge and your shoulder deliberately.

Start small: Looks for your toh before expansion

One of the most expensive errors in the Dot-Com era was trying to go very much Builders of artificial intelligence products Today it cannot be ignored.

Take Ebay, for example. It started as a simple online auction site for holdings – starting with something like PEZ distributions. The first users loved it because it has solved a very specific problem: it is connected to the amateurs who were unable to find each other in a non -connection mode. Only after controlling this initial vertical, EBay expanded to wider categories such as electronics and fashion, and in the end, anything you can buy almost today.

Compare that with WebvanStarting the Dot-Com era with a very different strategy. Webvan aims to revolutionize a grocery shop through online demand and the speedy delivery of the home – simultaneously, in multiple cities. Hundreds of millions of dollars have spent building huge warehouses and complex delivery fleets before they require a strong demand for customers. When growth was not achieved quickly, the company collapsed under its weight.

The pattern is clear: Start a sharp and specific user need. Focus on a narrow wedge that you can control. Expanding only when you have evidence of strong demand.

For AI products, this means resisting the desire to build “artificial intelligence that does everything”. Take, for example, a Truffie tool To analyze data. Do you target products, designers or data scientists? Do you build for people who do not know SQL, those who have limited experience or tanker analysts?

Each of these users has very different needs, functioning and expectations. Starting with a tight and well-defined collection-like technical project managers (PMS) with the Limited SQL experience who need quick visions to direct the product decisions-allows you to understand your user deeply, adjust the experience and build something really indispensable. From there, you can deliberately expand people or neighboring capabilities. In the race to build permanent GEN AI products, it will not be the winners who are trying to serve everyone simultaneously – they will be the ones who start young, and they serve an incredibly good person.

Own your data trench: Build an early driver

The small start helps you find the suitability of the product market. But as soon as the traction is gained, your next priority is to build the defense – and in Gen AI worldThis means having your data.

Companies that survived the Dot-Com boom not only captured users-they have acquired ownership data. Amazon, for example, did not stop selling books. They followed purchases and product views to improve recommendations, then used regional demand data to improve loyalty. By analyzing the purchase patterns across cities and postal symbols, they predicted the demand, storing the smartest warehouses and simplified shipping methods-the basis for the delivery of Prime for two days, and the main competitors were unable to match. None of them was possible without a baked data strategy in the product from the first day.

Follow Google a similar path. Each query, click and correct training data to improve search results – and then, ads. They only made a search engine; They built a reactions in the actual time that I constantly learned from users, which created a trench that made their results and aims to overcome it.

Lesson for Gen AI products builders Obviously: A long-term feature will not come from just accessing a strong model-it will come from building royal data rings that improve its products over time.

Today, anyone with adequate resources can convert a large open source language model (LLM) or pay to access the application programming interface. What is more difficult-much more valuable-is a high-signal and realistic user reaction collection that is over time.

If you are building the Gen AI product, you need to ask the critical questions early:

• What is the unique data that users will pick up with us?
• How can we design counter -feeding rings that constantly improve the product?
• Are there data on the field that we can collect (morally and safely) that the competitors will not enjoy?

Take duolingo, for example. With GPT-4, they have crossed Basic allocation. Features such as “explaining my answer” and playing artificial intelligence role to create more richer reactions for the user-not only capture answers, but how learners think and speak. Duolingo combines this data with their artificial intelligence to improve experience, creating an advantage of competitors that cannot be easily identical.

In the Gen AI era, the data should be a guaranteed feature. Companies that design their products to capture and learn from royal data will be those that survive and perform.

Conclusion: It is a marathon, not the enemy race

The Dot-Com era showed us that the noise fades quickly, but the basics carry. Gen AI mutation is no different. It will not be the companies that are chasing the main headlines – it will be what solves real problems, expanding with discipline and building real trenches.

The future of artificial intelligence will belong to the builders who understand that it is a marathon – and has grays to run it.

Kailiang Fu is the AI ​​product manager in Uber.