There is a question I have been hesitating over for nearly a year: whether to hire a front-end JavaScript developer. This is because a vast number of framework structures have begun to converge at this single point.
Then, last weekend, I stopped hesitating—because I found it: Claude 3.5.
In just over an hour on a bus, step by step, it built a full version of a Hold'em poker game.
From the plain version
To a usable version that some friends commented already looks quite good.
And even a server version (for training models) that I had been thinking about for a while but never had the time to implement.
It feels amazing to be able to think like a programmer without having to touch too much unfamiliar front-end code. I’ve always known I’m someone with relatively many ideas, and I know that the way to solve my problems is through matching execution power and restraining excessive impulsive thoughts.
Now, there seems to be a new solution, and for the 'low-code' approach we’ve been building, there’s an entirely different path forward.
In comparison, the siren example below is quite primitive, but it represents the first step many programmers—including myself—took thirty years ago from a purely text-based world toward 'multimodal': making a machine emit sound.
Now that ideas can take flight again, I’m going to ask for more—for instance, refactoring our entire current data analysis platform or even exploring the possibilities of the 3D world.
Alternatively, I wanted to test both the model's comprehension and execution, so I tried giving it just two pages of instruction manuals.
And it 'built' a camera!
All of this gives me enough motivation to continue enhancing 'productivity.'
What’s certain is that I really don’t need a JavaScript developer anymore. But another question follows immediately: if we simply divide a project into product managers (or business) and developers, how will their relationship change after the arrival of models like Claude 3.5 (which I believe is just the beginning)?
It seems that as long as you have enough ideas, you can quickly implement at least a prototype. But will developers say that moving from prototype to production has too many technical hurdles and that such a solution is impossible?
It seems that by leveraging the model’s knowledge base and comprehension first, it can answer about 70-80% of business logic even for unfamiliar domains. Then developers can quickly finish a business prototype, but will PMs say that landing this in actual business is simply impossible?
Interesting. Perhaps a 'one-person company' is a good solution.
By the way, one thing likely won't change: we still need to produce things that are 'beautiful.'