@kunchenguid I agree with you “chat isn’t all we need”. We need apps that build dynamically as we interact with our agents. The agent builds the app that contains the agent: An Emacs (Lisp, Smalltalk, etc) experience.
Before we use to have system design problems, the difference was that as the solution grew slowly we had time to address while we were building. Currently, problems are created as fast as llm writes code.
LLMs still produce bugs, but those bugs are different than what they used to be. It’s less off-by-ones and more about system design, ui usability, missing broader context. Some kinds of coding has been solved, but not all.
While models continue to improve, adversarial code review has been an incredibly powerful tool to catch many of these kinds of bugs.
It can be as simple as a one line prompt - “use a dynamic workflow to adversarial test every edge case in an iOS simulator”, or use Claude’s built in /code-review (or /code-review low, /code-review medium, etc.)
@_Felipe This seems inevitable to me. While browser use improved a lot in the last months, I expect that Web version of Github will continue to be developed for humans. So, they will always have to provide support to humans.
@glcst No one really is worthy, but it isn't about what we are or aren't. Jesus will kindly ask to visit our homes. If we welcome Him inside, repent of our sins, and recognize Him as our Savior, salvation will be there, and we may rest knowing that one day we will be with Him forever.
@_Felipe I usually do /clear and execute the plan. While models nowadays are really good on following your last (updated) requests/decisions, I feel that results are better when you start from a clean session.
@andy_matuschak I noticed the value of contextual backlinks when linking some of my question notes. The answer (or at least better understand the problem) emerged from the explanation between them.
@andy_matuschak Are tags also notes as in your examples of Peas? Or do you keep them in a different level? This idea of representing terms and tags as notes seems so powerful. Even though I don’t know how to explain why, I feel that the answer is related with code as data.
@fnando Agreed. One thing is talking about abstraction layers and other is implementing a simple one that later can scaffold in something valuable. Other aspect of your tip is that later if the api needs to change you can see do that easily (assuming that you use Typescript or Flow).
@fnando The point isn’t what to use for actually implementing those function, but the possibility of having these layers defined since the beginning (and the long-term value of having done the job a bit here and there). Nice tip!