Dev

“Meaning is not something you stumble across, like the answer to a riddle or the prize in a treasure hunt. Meaning is something you build into your life.”

@thatssodhawal @mavehealth @Google @ufc @ycombinator @BlumeVentures Looking to try out soon!

We've raised $2.1M to fix your focus. Our wearable headset @mavehealth improves attention & stress regulation in just 20 minutes a day for users at @Google, @ufc, @ycombinator. Backed by @BlumeVentures, alongside existing and new investors. Order now at mavehealth.com 🇺🇸🇮🇳

@sama what stood out to me using Codex is reliability. it doesn't try to be clever or "complete the vibe". it actually tries to complete the task correctly.

The Codex team are hardcore builders and it really comes through in what they create. No surprise all the hardcore builders I know have switched to Codex. Usage of Codex is growing very fast:

@gdb the interesting part is how fast "tokens" are becoming the new unit of compute. a year ago 5T/day would have sounded absurd.

gpt-5.4 has ramped faster than any other model we've launched in the API: within a week of launch, 5T tokens per day, handling more volume than our entire API one year ago, and reaching an annualized run rate of $1B in net-new revenue. it's a good model, try it out!

AI didn't remove software engineering. It removed the illusion that typing was the hard part.

@Yuchenj_UW --auto-approve-life-decisions

--dangerously-skip-permissions is how I run Claude Code also how I run my life

@steipete @notsunsakis @bcherny vibe coding might get you to a prototype product market fit usually comes from weeks of hitting your head, doubting, iterating

@notsunsakis @bcherny don't call it vibe coding - that's associated with yolo i smash head on keyboard, not thinking, engineering, building, testing, debugging, iterating. agentic engineering, or just...coding. We move faster, but it's still hard.

Has anybody vibe coded their way into product market fit? From the top of my dome: Claude Code: founded in 2024 as an internal tool by @bcherny who prototyped it at Anthropic and within a week nearly half of developers there were using it. Long considered their moat, eventually it was decided to release it to the public. After being so successful internally, this dogfooded product had already found a market fit among the devs and it spread like wildfire. OpenClaw: This tool by @steipete initially named ClawdBot, then at Claude legal department’s request renamed to MoltBot, and eventually OpenClaw along with acquisition by OpenAI and MoltBook by Meta, was another success story that was made real by granting AI executive power, just like Claude Code. These two vibe coded tools share a lot of similarities - they are AI harnesses that give the models that run them agentic power, in other words, permission to edit files autonomously. Any other vibe coded tools found a product market fit?

@ThePrimeagen 30 minutes a day for a year is ~180 hours. That’s the difference between "interested in something" and "good at it."

it's crazy how many people think that 30 minutes a day isn't enough to learn anything. Who told you this? Unknown knowns are wild. People just ingrain hurdles in their head about the universe because some thought leader spoke into their life in the distant past.

LLMs made code cheap. Cheap code changes how systems evolve. The new skill isn’t writing software. It’s preventing unnecessary software from existing.

@KevinNaughtonJr knowledge work, now with extra politeness

it's hard work sitting on my laptop all day politely asking my computer to do my entire job for me

@chamath there’s a reason big tech built strict code review cultures they were never limited by typing speed they were limited by judgment LLMs lowered the cost of writing code they didn’t lower the cost of running systems

true

@EricRWeinstein @claudeai so basically it learned you but decided it knows better

One of the things I *love* about @claudeai CoWork mode is that it learns me. My actual voice. Who I am and what I expect. As a mind. As a coder. As an author. As a partner. And then, it just TOTALLY disregards that very personal knowledge base to do whatever the fuck *it* wants.

@Hesamation the window market just got disrupted

holy shit window manufacturing stocks crashed right after this Claude announcement.

Here are 10 ways to get more good ideas: > listen to podcasts > try more things > go for walks > try more things > get 8hrs sleep > try more things > try more things > try more things > try more things > try more things

@PeterDiamandis Reminds me of that old joke. AI engineers keep building a better and better AI, asking the same question: “Is there a God?” Eventually, the AI answers: “There is now!”

@EXM7777 Exactly. Prompting is just the current UI layer. The durable skill is: - knowing what outcome you want - knowing when the output is wrong That’s judgment, not prompting.

i stopped writing prompts a few weeks ago and my output got better... all agents i used are trained on my prompting principles, started sending voice notes through telegram with raw ideas, and let the system handle the rest because sitting there engineering a solid prompt is just more friction the real skill is always upstream: knowing what outcome you want, understanding why certain structures get better results, building mental models for how these systems interpret instructions that part never goes away, and no agent can replace it because it's the judgment layer but everything downstream of that judgment... the formatting, the structure, the assembly that's just labor every tool eventually splits into a strategy layer and an execution layer, and the execution layer always gets automated first prompting just hit that split faster later this year models will get smart enough to interpret messy inputs without needing much at all... which means the window where even your agents need good prompting is shrinking the only durable advantage is the thinking that sits above the prompt

@GergelyOrosz Big companies kill products when they don't move the needle. Startups build them when they *are* the needle. Same product. Completely different economics.

As AI coding tools went mainstream, Amazon decided it’s not worth them supporting their Zoom clone, called Chime (that has paying customers!) And yet startups are assuming it’s worth rebuilding and supporting their own JIRA clones (with no paying customers) Who is mistaken?

Every major shift in computing moves the abstraction layer up. Assembly → C C → frameworks frameworks → AI The skill that matters is not the tool. It's knowing when the layer moved.

@thdxr AI adoption isn’t a strategy. it’s just a new input.

there's so much bro-economics going on right now "if the company doesn't use ai then they'll be less productive than their competitors and so they'll die" yeah imma go crush exxon mobil with the clanker crew and my 3 second attention span

@NeuraCraftAI @TheVixhal yes. execution gets cheaper, but judgment becomes the scarce resource.

@debuggerdev @TheVixhal this is the natural evolution of the stack. every time the abstraction layer moves up, the "how" becomes a commodity while the "why" becomes the premium. we're moving from being builders to being architects of logic. it's all about systems design now.

Computer science is gradually returning to the domain of physicists, mathematicians, and electrical engineers as large language models automate much of what we currently call software engineering. The field’s center of gravity is shifting away from manual code writing and toward deeper theoretical thinking, mathematical insight, and systems-level reasoning.