oleosjo

@_joshideas @0xkozue Fair enough, thanks for the reply.

It's the features, the SDKs, how you organize and word your system/user prompts, build MCP tools for specific use cases, creating a 'spine' that represents a workflow that can be created dynamically. Things like that. One of the most important things is how well you use that system prompt.

Anyone else found that working with AI agents is more mentally tiring than just doing the work yourself?

@MattPenny99 That’s not true for me personally at all. The feel and design of the app is actually a key deciding factor in if I’ll use it. If it looks like slop I won’t enjoy it as much (or template-y back in the day) And I imagine there’s at least a percentage of people that feel the same.

Design does NOT matter in software If your SaaS looks like this, clearly AI built, it makes no difference People care if your product works, and solves a real world problem That’s it. Now design on the marketing site is another topic, there it matters. But spending time changing the design of the product rather than - marketing - speaking to customers - building desired features Is pure copium

@jasonfried This is how you build your moat and fort. Maybe add in capture alligators?

Before shipping, chop wood, carry water. After shipping, chop wood, carry water.

@CasJam For me it worked much better when I could slowly build and think through each abstraction layer at a time as I built it. A big markdown file, some hope and dreams, and my lucky rabbit’s foot is the environment I have to define the problem in now. Skill issue, I know.

Stop asking "Can AI build this?" Start asking "Can I specify this?" AI took away the syntax barrier. Not the hard part: defining the problem, the constraints, and the milestones clearly enough for an agent to build it. That's the skill now.

@_joshideas @0xkozue Pray tell, what is their language?

@0xkozue No. Once you understand their language, it's pretty easy

@zekramu Omg you still use JJ? There’s PP now

it blows my mind that people are still using git.

@ThePrimeagen 10x developer naturally. AI 2xed that. It also negative 4xed on that. So I dunno where I’m at now.

I am so thoroughly convinced that anyone who thinks AI 100x's their output is a liar or a lunatic. You are telling me you can make 1 years worth of decisions in 3.65 days? Let alone describing those accurately and coaxing the result from the AI... (1.8 days european time)?

@hschmaiske The few people that have continued to invest in those “old school” skills AI has allegedly replaced will be at a marked advantage as time goes on compared to those who have invested only in “learning AI.” Could be wrong but I guess we’ll see who gets “left behind.”

People are stopping to think and letting AI do everything. They don't even read the AI output anymore. The bar is low... very low.

@teodorio 1. Train on entire internet 2. Think you have control over a prediction engine derived from said data 3. Cry when the truth sets in

After using Opus 4.8 extensively I assume there is some panic internally at Anthropic as it seems they are defaulting to cheap post-training tricks to get the models to "push" back. Which means that the biggest failure mode of wrong token paths out of which the models lack the meta rationality to pull themselves out is a huge and non trivial issue.

@KostjaPalovic And now you’re not allowed to take your time with the details of a quality codebase you personally like because that’s “wasting time.” The LLMs can write code that’s the equivalent of that hollow prose with a bunch of em dashes and you’re now required to say “good enough.”

AI took away all the fun of coding and gave us the worst parts of it: code reviews and writing specs.

@0xIlyy Zellij > tmux

Please, stop running your agents in tmux. I just found out about Zellij and terminal multiplexing finally makes sense.

@caps_raunak When given the chance, most people will just go ahead and shoot up the steroids to get ahead, with little thought about the nasty consequences coming for them later.

AI is killing the grind. You work hard for years to get good at building apps. Make mistakes, fix them, build taste. Now one prompt and it's done. No sweat, no learning. Everyone shares it like it's amazing. But we're making real skills cheap and fake. Stop smiling and think about it.

@haider1 I can say “LLMs are so stupid,” and “LLMs are amazing,” very confident that both are true.

person: "LLMs are so stupid" > me: which model are you using? > them: free GPT-5.5, 16k context, no tools > pastes no context, wants perfect code > refuses to try current models: "just hype" > tries a modern LLM once > "oh… ok, this works" every single time...

@weswinder @dreamsofcode_io Convenient

@dreamsofcode_io i ain't no snitch

i looked at 12 different x accounts that constantly trash vibe coding almost none of them have shipped a product with paying users interesting pattern

@mattshumer_ @OpenAI Woah I bet you have something mind-blowing to show for it!!

Light work @OpenAI Apparently, I’ve used 3x the tokens of OpenAI’s highest user in just the last 17 days

@GaneshKompella I guess so. I was going off of this statement: "The gap between engineers is no longer who types faster." But I think in the end we agree. I just was stating that fast typing was never what made you a good engineer even in those ancient times before AI.

@jonathanmcdill You failed to understand what the post was about

The gap between engineers is no longer who types faster. It's who can hold a clean problem statement while Claude Code does the typing. Cursor and Claude Code don't reward fast hands. They reward sharp specs.

@pothuLabs Don't wake them up to this yet! I enjoy building things in a predictable "slow" way that improves, not erodes my skills. It's like this fun secret while everyone else is letting their brain rot, getting so "good" at playing the casino.

Is it just me or is vibe coding way less efficient than doing it the old fashioned way

@maxedapps It's like a savant toddler came in the room and started screaming and wrecking things but we have to put up with him because "he's got so much potential."

Supply chain attacks, broken code everywhere, mass layoffs that are justified "because of AI", stupid / dangerous AI-powered customer support (hi meta!), AI gurus selling you their latest skills and "must use workflows" ... glorious times :)

@MillionInt To figure out what is truely important one must do unimportant things

Everything in life follows from repeatedly answering a simple question: "what are the most important things for me right now?"