Companies thinking doubling the token limits on the engineering team's plan will double their output are the same people thinking that doubling the number of engineers on a team will double the team's output

Open source models are so much cheaper to run than frontier models, once the subsidy of tokens stops they will get much more important

looooopmaxxing all day, everyday

ai dev tools giving subsidized access to models is both the greatest and worst possible thing we all know it won't last forever right?

we should build more deterministic workflows into our agents to reduce costs and improve accuracy is something I’ve been saying for 3+ yrs now

AI is great at coding but makes the same architectural mistakes that humans often make, which is not strange as the models are trained on huge datasets of human mistakes

PROGRAMMING IS NOT DEAD My openclaw agent got stuck after an update, I tried to let it fix itself but ended up reinstalling nvm myself

Companies thinking doubling the token limits on the engineering team's plan will double their output are the same people thinking that doubling the number of engineers on a team will double the team's output

Most coding interviews today are broken. Leetcode challenges made no more sense the moment AI became part of software development. And even before, online coding assignments were never a good indicator of someone's actual job skills.

Good developer with AI = better developer Bad developer with AI = worse developer Average developer with AI = hit or miss

Online coding assignments were always stupid. When interviewing I cared much more about someone's process than their actual coding skills. If you can think in systems and problem solve, you will fix any challenge with or without AI

I just feel sad for the people who have spent the last 3-4 yrs learning something they cannot put in practice, while they thought they would get a solid career out of it. Instead, they need to pivot fast to become experts in using AI on top off these newly learned skills

Would it matter if we prompt agents to generate secure code, when there's an almighty LLM (like Mythos) finding all our security gaps and fixes it for us?

I've seen people replace their own critical thinking skills with AI, that's not progress

Oh man, calculators made me so lazy in high school. the moment we were allowed to use them I basically stopped calculating any numbers in my head

Can you believe it has been 3+ years since this tweet?

The fact that Anthropic is hiding Mythos makes me want it even more

Tokenmaxxing has become the new MRR Every developer is just showing off at this point

junior slop coder vs senior

It can happen to the best of us

When do we get to see how AI CEOs use their products?

anyone feeling like this?

The fastest way to break an AI agent? Give it 50 tools.

parameters are like horsepower

MCP conference anyone?

Sunday: restored my chatgtp plus account, being able to use both codex and oauth in tools like openclaw is huge

It’s sad to realize some of my non-tech friends who say “I’m too deep into AI” have no clue what the world looks like in a few years

Who is building in this space?

This is impressive! Vibe coding is turning technical founders back into developer mode

Workflow of a senior vibe engineer