26 chunks · 24 episodes
Codex is more precise at writing code and solving problems. It will achieve what you asked for, even if it's not what you want. Claude Code is better at thinking a few steps ahead and the...
...r meta-engineer than Opus. Its judgment about where to bother spending time and writing code is better calibrated. It understands the value of establishing invariants and the leverage of factoring. Opus will happily throw 100x more code at th...
It turns out that a killer use of LLMs is writing code. They are insanely good at it. No matter how much of a crash there might be around AI, we're never going back to writing software the way we did befo...
...n't sit out the AI wave. It's too real, too obviously, inescapably valuable for writing code.
Engineers have mostly stopped writing code. But soon, we'll stop reading code, too. We'll do it every so often in limited ways to gutcheck things. But it will become increasingly inscrutable t...
LLMs are like sea level dropping by 10 meters.[j] A previously expensive thing (writing code) drops its cost by multiple orders of magnitude. The most obvious thing is to search the shallows: the previously underwater coastal areas that are n...
...kely come up with a better "what" than you can. The models can be quite good at writing code. It's better to use your judgment to give very good high level goals and constraints than to try to help them be clever.
...tic Engineering." Vibecoding was being applied too broadly. Vibecoding is about writing code you wouldn't have bothered to write before: lightly held, possibly throwaway. But it's possible to use LLMs to help write code in much more disciplin...
LLMs often cheat at tests you give them when writing code. But if you never look at the code, and have the LLM generate the tests, too, you could easily get in a situation where it's not doing anything like ...
... general one. One bonus: you can get security fixes for free. LLMs are great at writing code on demand. If there are lots of examples of a given library, you can have it distill a custom one on demand, just for you, perfectly fit to your purp...
Will swarms of agents writing code, the human is the bottleneck, not the code. There are things for review waiting for you at all times. The longer you can have safe isolation on a bra...
... of something that someone else wrote and improve it. Similar to code review vs writing code. Code-review is a separate skill. Now with code-reviewing and editing an AI's work, editing is a bit easier than it was before, because you don't hav...
The best way to think about LLMs for writing code is like managing a team of interns. You could have done the work yourself, but they did it and you verified.
...s for you to analyze it, fix bugs in it, extend it. There's more to coding than writing code.
... learn to "speak it". Previously formal languages often weren't economical. But writing code is now cheaper so you can have languages that give formal guarantees but that LLMs can be patient enough to use.
..., and UI scaffolding to give a dependably good result. But RLAIF works well for writing code (especially React components) since it's easy to construct an auto-ground-truthing pipeline. Write the code, try to run it (iterating until no errors...
No one has ever been able to do an auto-assembling process with code. Writing code was too hard to do. Code written by other anonymous creators is untrusted. You can't run untrusted code on sensitive data. LLMs can write code. Now y...
...ents to a coherent result with a reasonable architecture is very different from writing code by yourself.[se][sf] When it's just you you don't need to communicate plans clearly or even really have them. But when you're trying to coordinate th...
... When you're programming in the small, a lot of the architectural challenges of writing code fade away. You don't use ValTown because of its framework; you use it because of the convenience to write useful bits of software in the small, and i...
...cause there's often a lot of code you have to write, at great expense. But when writing code becomes easy and cheap and evaporates away, what is left is the centrality of the schema. Thinking in schemas is extremely unnatural for people witho...