Jon Katzur

Recommending products @amazon. Prior founder at @proximodata, product @slackhq, data + eng @1010data

[Jon Katzur @jkatzur]

@far33d I think this flows the other way, where LLMs give languages an aggregation dynamic where you choose a particular language because of how good AI is at writing it. And then more people write in that language and the flywheel continues. https://t.co/lh2gJTj82W

[Jon Katzur @jkatzur]

@bernhardsson Atlanta somehow west of Detroit is the continental US version of this.

[Jon Katzur @jkatzur]

@bernhardsson And also feel comfortable hiring some mortgage services officers! The customer doesn’t care if it’s AI or human it just has to work. The X% that isn’t automated or accurate enough with AI still needs to work or the TAM shrinks dramatically.

[Jon Katzur @jkatzur]

@levelsio @chatrankai is working on exactly this

[Jon Katzur @jkatzur]

@jeff_weinstein Check out @chatrankai for this!

[Jon Katzur @jkatzur]

@Jason Check out @chatrankai

[Jon Katzur @jkatzur]

Obligatory short blog post thinking about this: https://t.co/0ES7S1vbpT

[Jon Katzur @jkatzur]

Friends have texted me this question before, so I have fun purely vibe-coding this up. My first foray into natural language only coding. Pretty fun!

https://t.co/Lv7LL0BkqG

[Jon Katzur @jkatzur]

@tangentstorm Great take. Love idea to use AI to create a local fork modified to your exact needs while reducing dependencies.

[Jon Katzur @jkatzur]

@jeff_weinstein @rauchg @vercel @OpenAI @Google @atlas You see https://t.co/059ik5c7xl? SEO tool for LLMs that my friends are building for folks looking to track and improve LLM referrals

[Jon Katzur @jkatzur]

Success early in your career is so much about building support for your ideas. You’re close to the problems and there’s less risk. Later it’s about making the correct decisions with uncertainty. Being too persuasive may be a net negative. You don’t want support for bad ideas!

[Jon Katzur @jkatzur]

@bradysays I think this tweet is what made me switch!

[Jon Katzur @jkatzur]

Are typescript and python the last languages?

https://t.co/lh2gJTj82W

[Jon Katzur @jkatzur]

@tangentstorm Yeah I’ve found hallucinations like this that almost would be good spec for features and very reasonable, but not actually working now, which puts you back in OODA loop. That said, I really noticed this with pandas last year and already seems to be improving.

[Jon Katzur @jkatzur]

@tangentstorm Good take. More like the AI version of Kotlin working on the JVM building on decades there with better syntax. Perhaps that means newer languages will have to work to be extra backwards AI compatible in the LLM layers to work well.

[Jon Katzur @jkatzur]

@bradysays Cursor was an easy transition for me as it’s just a fork and you can simply import your settings which includes all keybindings.

I find the compose workflow in cursor dramatically more powerful than the codegen in vscode when I was on it, but it’s been a few months

[Jon Katzur @jkatzur]

@bradysays 1. I almost just titled it typescript
2. Do you use cursor / windsurf? I really notice ease of typescript with cursor

[Jon Katzur @jkatzur]

it’s remarkable that it’s easier to get someone to give you money (angel invest) than to give them money (job offer). 20 year old me could not have comprehended this.

[Jon Katzur @jkatzur]

@bernhardsson I think this is an example of @benthompson ‘s aggregation theory at work with AI. The languages/frameworks with more indexable content will be better with AI causing developers to pick them causing reinforcement learning to make them still better

[Jon Katzur @jkatzur]

@patio11 Great example and framing. LLMs in best case give you conventional wisdom or even common sense in many fields. This can be super value - “common sense isn’t” - but obviously many domains where the conventional wisdom is wrong or not useful