I have today figured out what new AI world would look like for software engineers, specifically for juniors and middle ones: They are going to babysit agents, build tools for them (using ai), etc. Seniors would determine a vector and what to build. Essentially everyone involved would need to build and ship a software factory model where software produces software. Honestly it is not that different from what it was before.

HO. LY. SHIT.

I just went to Walmart. I said, I’d like to buy these oranges. Seems simple right?

They said, “Let me weigh this.”

HO. LY. FUCK.

The register WEIGHS the oranges. 🤯 This way if you get two small ones or one big one, you pay the SAME. PRICE.

This. changes. everything.

I can’t even fucking believe the reality I’m living in. I’m literally fucking dissociating from how crazy this is. I literally am scared and can’t tell where I am. Please fucking help me

it seems people become much more self-absorbed and not willing to listen to criticism, i feel it is mostly because of ai that enables all kind of psychosis. funny but i did a small “research” in the university and studied how texts that are created in a special way (ie horoscopes) would turn people to believe all kind of shit and it feels it is exactly the same situation. outcome of that research was that only the craziest cynics would survive, second tier people (like me) would need year+ for mental training to withstand. everyone else just got oneshotted. frankly it was correlated to intelligence of a person but really poorly it wasn’t that smartest would survive they have higher chances but not that much really.

i bet many projects will be built in next months around social + ai, but it will be so overcrowded so no one would care and they all will die. just like “personal wiki” or “todo” products are dead (except incumbents).

Still want to build a small harness similar to claude managed agents one, where you can upload envs, apis, etc and then combine agents for specific task, but i am torn apart - should it be code first (ie store everything in github) or cloud first (ie do everything in the cloud). i kind of like the idea of a box that just sits there and provides a simple api for simple agents. Cloud first plus is that you just write prompts/connect tools/deploy functions and it is wokring. You can have your own versioning, experiments, forking anything really. With code first it is a little bit more natural but you have to diff on changes all the time which is annoying at minimum, but you can share and fork stuff, but i highly doubt that i want to share all my personal agents with anyone? why would i? something internal to my home network for example? i dont even need forks and merging.

Funny cloud vs code results and now ramp posts yet another nice thing in favor of essentially cloud first approach without code at all:
https://x.com/sebgoddijn/status/2042285915435937816

Trying new thing - method acting LLM, i am answering to people like i am LLM: for now i found two patterns that worth trying: answering without right context and not getting responsibility for an answer. It actually shapes in my head what LLMs are.

New definition of AGI is here

People clearly lost faith in themselves, like that they cant code efficiently anymore without LLM. Sure you can ship new things, but only built on solid foundation that you will design and maintain. You produce slop as a coping with disbelief in yourself, very few use ai to build something better. There is a growing herd of wanderers that are indeed build a great things with AI, optimisits. Pessimists are the one who will stay in permanent underclass, optimists will make money.

i was discussing robotics with friends like it was one of the things where you can commit for year or so and try to build something, and now it is like several people told me so. how does it work?

I think i can estimate if someone is smart by simply checking if they are reacting immediatelly to some non-trivial question or if they stutter and introduce some filler words before the answer.

Poland is now 1/2.5 of russia

Very nicely written text, haven’t seen much better long form posts on X

https://x.com/jayagup10/status/2052870394093408558

first in my bloodline to have MRR

Sometimes i like the models, they do crazy things. I was building a small shim for Amazon Bedrock inference and it didnt work for something, Codex decided to find where claude is installed went to its folder, grepped its compiled js files and found how claude code does this and replicated. Zero prompting.