When I built menugen ~1 year ago, I observed that the hardest part by far was not the code itself, it was the plethora of services you have to assemble like IKEA furniture to make it real, the DevOps: services, payments, auth, database, security, domain names, etc...
I am really looking forward to a day where I could simply tell my agent: "build menugen" (referencing the post) and it would just work. The whole thing up to the deployed web page. The agent would have to browse a number of services, read the docs, get all the api keys, make everything work, debug it in dev, and deploy to prod. This is the actually hard part, not the code itself. Or rather, the better way to think about it is that the entire DevOps lifecycle has to become code, in addition to the necessary sensors/actuators of the CLIs/APIs with agent-native ergonomics. And there should be no need to visit web pages, click buttons, or anything like that for the human.
It's easy to state, it's now just barely technically possible and expected to work maybe, but it definitely requires from-scratch re-design, work and thought. Very exciting direction!
If you want to become a better software engineer, read these famous posts from top companies (OpenAI, Airbnb, Stripe, Figma, Netflix, Meta)
I spent hours curating top posts so you don't have to:
I rarely use ChatGPT anymore.
Instead, I have a simple script to access the API directly. Every developer should consider doing the same.
This approach is cheaper for me. It's also more flexible and easier to run different experiments.
There are advantages and disadvantages to this method.
First, when I mentioned this yesterday, some people said the API doesn't keep the context of a conversation. This is inaccurate. As you'll see in the code, I keep an ongoing dialogue with the API and retain the context. My solution is rudimentary but is good enough for me. You can improve it if you need to.
Regarding pricing, I usually use GPT-3.5, which is very cheap. I can't imagine getting closer to $20/month using this model. For certain things, I use GPT-4. Slower and more expensive, but get better results.
The API might be more expensive if you are a heavy GPT-4 user. Don't fall for the one-time bill someone had one time and think that will happen every time. ChatGPT Plus is $20 monthly, regardless of how much you use it. That adds up quickly.
The main downside of using the API is losing access to plugins. If you rely on those, you must stick with ChatGPT's GUI.
On the other hand, you gain access to "function calling" through the API. OpenAI released this feature a few weeks ago, and it's a game-changer if you want to build anything serious with these models.
Of course, whether or not this is better for you depends on how you use the system today. I wouldn't recommend that my mom does this, but developers will probably have a better time and appreciate the flexibility.
The code in the screenshot is just a preview of the code. Link to the full notebook is in the next tweet.