About once a year the urge to write here comes back. The topic is always the same: artificial intelligence. But lately there’s been so much information that I didn’t feel like adding more to the pile.
Trying to keep up with every update is easy to turn into a trap — you read what’s new, and… you don’t do anything. Trying to keep up with everything ends with losing interest in doing anything.
For a long time I thought AI was underrated, not overrated. Now I see both are true at the same time. The magic button that “solves every problem and makes you money” — that’s the hype. Over the past year I’ve seen dozens of prototypes where it all looks simple: arrange the boxes on a diagram, and there’s your finished product.
In practice this almost never flies.
The good products I’ve come across always share two things:
- Clear metrics (evals).
- Very fast iteration.
The formula is simple: you don’t have to be the smartest person to build the best prompt or the best system. You have to understand the pain, figure out how to measure the quality of a solution, and check hypotheses as quickly as you can. The more experiments you run, the closer you are to a genuinely good product.
Not all that different from regular product development, is it?
Let’s try one more pass at Telegram — fewer news, more observations.