There was precious little of what GSD added in terms of value, other than the author saying it did some process stuff and blew through a bunch of tokens.

Anyone able to comment on how GSD improves CC over the base behavior? Can you get the same benefit by writing thoughtful descriptions of what you want it to build?

What would it take for Linux phones to gain the ability to run Android apps?

Congrats on launching!

How is this superior to an RSS reader?

The author managed to find the strangest people & phenomena in San Francisco and make it sound like they’re a complete picture of life there. But there are packed brunch spots and parks on sunny weekends that would disagree very strongly.

San Francisco is a tolerant place. Tolerance is how you get Juicero or Theranos and whatever Cluely seems to have pivoted to, but it’s also how you get Twitter, Uber, Dropbox.. and thousands of others.

So it is crucial to consider proportionality. Taking some bad with some good results in getting a little bit of bad and a hell of a lot of good. But if you aren’t careful, all you’ll see is the bad.

Yeah that was pretty much my thought throughout the piece.

It felt like the author was punching down, too. This Cluely founder seems largely unsuccessful and, as the boat guy says at the end, just a kid. A chud of a kid, but a kid nonetheless.

Where do you see money going to the good?

Superficially it sounds like this could create a bit more of a move toward doing compaction on some continuous basis, or compacting in batches once you hit the context limit, rather than starting fresh with a summary and system prompt..

Feels like high fidelity, fast compaction could be a path to “solving” long context.

How many of these could be code checks written by the agent? Have it walk the AST looking for the antipatterns - it ends up more like a custom linter, and no need to waste tokens making the poor model subjectively analyze its decisions.

My suspicion is you can’t expect LLMs to one-shot any sort of non-derivative work.

If you want better than the default outcome you have to take what it gives you and feed it back in alongside examples.

In backend dev they say, make it work - then make it fast - then make it cheap. It’s another way to say, no one will get it right first time because just getting anything the first time is hard enough.

I guess frontend would be something like, make it work; make it functional; make it beautiful.

> write the first test so the AI knows how they should be written, and which cases should be tested. Then I tell the AI each test case and it writes them for me.

This is too low level. You’d be better off describing the things that need testing and asking for it to do red/green test-driven development (TDD). Then you’ll know all the tests are needed, and it’ll decide what tests to write without your intervention, and make them pass while you sip coffee :)

> I don’t trust it yet is when code must be copy pasted.

Ask it to perform the copy-paste using code - have it write and execute a quick script. You can review the script before it runs and that will make sure it can’t alter details on the way through.

dang the top comments here are about the title being editorialized. Isn’t that against HN guidelines?

Either I’m going crazy, or I’m pretty sure I submitted this with the original title. Meaning the change was done by the mods.

Oh wow they’re really holding him to account by asking some interesting questions then letting him get back to it.

/s

Agreed - such useless pageantry. At least with meat, 'grilling' changes it.