Deep

most people are using AI like a better Google. the ones pulling ahead are using it like a full team. researcher, engineer, writer, product manager. all running in parallel. that gap is only going to get wider.

@BretWeinstein knowing how to prompt is just knowing how to think clearly and communicate precisely. turns out that was always the skill. ai just made it impossible to fake.

Learn to prompt AI to code.

three months ago i was spending 2 hours debugging an issue that claude now solves in 4 minutes. i didn't become a worse engineer. i just stopped pretending that suffering through docs was the same as being good at my job.

unpopular opinion: the engineers who are scared of AI taking their jobs are the same ones who never wanted to do the boring parts anyway. the ones actually building with it every day aren't worried. they're too busy shipping.

@OfficialLoganK the ones that felt the most "human" and relationship-driven are probably first. nobody disrupts what they think can't be disrupted.

@thdxr the hoarding makes sense. everyone's betting the next 6 months decide who's 3 years ahead. nobody wants to be caught waiting for hardware when the moment hits.

basically everyone is telling us there's shortages in every component of deploying GPUs, even labor lot of nervousness and hoarding right now, some crazy stuff going on idk what things are going to look like in the next 6 months

@Codie_Sanchez the plumber who learns to use AI for quoting, scheduling and customer follow ups is going to eat everyone else's lunch while still being the only one who can fix the pipe.

What AI can do: -Write code -Do admin tasks -Write your emails -Review documents -Research and look at data What AI can’t do: -Shake someone's hand and close the deal -Fix your plumbing at 11pm on a Tuesday -Wire your electrical panel -Show up to a job site -Install a roof

@Codie_Sanchez @amasad the back office guy always had the domain knowledge. he just finally got the tools to do something with it.

Last year, I sat down with @amasad, CEO of Replit, and he told me a story I can't stop thinking about: One of his users is a finance guy at a VC firm. Not on the investing team, on the back-office side. He spent nights and weekends building his dream portfolio management platform on Replit. And he got $5M ARR in commitments before he quit his job... Domain knowledge + AI tools = a business hiding in plain sight.

@martin_casado the best prompt is often just thinking clearly before you open the chat. the tokens go down because the thinking went up.

Total token use as a measure of AI literacy is wrong headed. In my experience, after some baseline, more token use is inversely correlated with competency using AI.

@BoyanSlat curiosity was always the differentiator. ai just made it impossible to hide if you don't have it.

AI is putting a massive premium on curiosity. The kind of person that used to get lost in wikipedia rabbit holes is learning more than ever. Others are using it to outsource their homework.

@saranormous rubber ducking works because explaining your thinking out loud catches the gaps. silicon ducking just does it faster and talks back.

is talking to a model fine-tuned on your own thoughts (say, for investment judgement) a sign of madness? given this kind of iteration can help models, maybe it helps humans "silicon ducking" instead of rubber ducking

@garrytan the morning session hits different because your head is still thinking in systems, not tasks. by afternoon you're just reacting.

Weird realization: The best AI coding is in the morning when you are fresh from a night full of dreaming about latent space. Sleep early. Wake up early. The best ideas are in the morning. It's not just about raw token maxxing. It is about teaching the machines the right abstraction that comes out of your own personal experience and the synthesis that comes from a good night's sleep.

@NoahKingJr didn’t ruin it. just made the gap between people with something real to say and people with nothing to say much more obvious.

Do you think that AI ruined the Social Media?

@rezoundous mostly trust the result. but i read it enough to know what to blame when it breaks.

Do you understand the code or just trust the result? Be honest.

@ThePrimeagen we really said “what if the secret sauce was just… a markdown file”

we are currently living in a world where there are some companies whose moat are literal skill files...

@dotkrueger when the reasoning becomes free, the only moat left is knowing what to build and who to build it for. that’s a very human problem

Claude Opus 4.6 level reasoning will be effectively free in 6m-1yr. The stuff that can be built on this platform will change the world, far more than the information revolution of 1995-2020.

@tlbtlbtlb the tasks we procrastinate on the most are usually the ones ai handles the easiest. that’s a weird thing to sit with.

Something Claude did in 20 minutes that I'd been procrastinating for months: "Please convert this project from next.js to esbuild & plain React". The project had wasm, canvases, MathJaX. It just worked.

@shreyacasmalert you’re basically using AI to think about how to think. that’s the part most people skip entirely.

Most people are using AI wrong: → ask question → get mid answer Try this instead: → ask AI to write a prompt → paste it in a new chat and suddenly the answers are 10x better! basic recursion at its finest 💁🏻‍♀️

nobody talks about how AI made the boring parts of engineering disappear. not the hard problems. those are still hard. but the 40 minutes spent writing boilerplate, the Stack Overflow rabbit holes, the syntax you can never remember. that time is just gone now. and i don't miss it at all.

@mitsuhiko at some point you stop being the programmer and start being the one setting the intention. the agent handles the rest. that's a weird identity shift nobody talks about.

Having an agent edit its own edit tool like and debug it is the most meta experience possible.

@alexcooldev RLS feels like magic until it doesn't work and you have no idea why. raw SQL with explicit checks is boring but at least you know exactly what's happening.

A big lesson when using Supabase: use it as a PostgreSQL database instead of relying on the RLS pattern with the Supabase client.