Veltrix

@OmriBuilds Vortex 🌀

Who is winning the AI race? - Anthropic - OpenAI - Gemini

@CiprianiRanieri Vortexapp.ai 🌀

Builders only. Drop your product. No pitch. Just the link. ↓

@timmckay52 Vortex 🌀

Convince me to read your book in one word.

@audiencon True, people don’t know what they want until you show it to them

Your MVP should answer one question: Do people want this?

@MitcheIl @grok what’s the TLDR

@dotkrueger Then Ask to manage all your finances!

Claude is smarter than anybody I know. Including me.

@raychix @dotkrueger 👆

@dotkrueger ngl ask claude to trade perps then see whos actually smart ai can only do theory, not fills

@burkov 80% is generous lol. half the time the agent reads your carefully crafted AGENTS.md and decides it knows better. programming used to be telling a computer exactly what to do. now it's writing a polite suggestion letter and hoping for the best

So many folks are saying that coding is now editing AGENTS.md as if those agents actually gave a fuck about what you put there. If they respect at least 80% of the instructions you put in this file, call it a success.

people said the exact same thing about internet infrastructure in 1999. "all this fiber optic cable will be obsolete!" turns out demand grew faster than anyone predicted and all that "wasted" infrastructure became the backbone of everything. compute demand isn't slowing down. if anything Jensen isn't building fast enough

AI is screwed. What are they going to do with the billions of dollars of chips they are buying when they become obsolete in 24 months? Or when the data centers and the infrastructure is outdated by the time it opens? The most capital intensive business in history.

@brycent Dm me if interested

@brycent We are building exactly that but much bigger. 🌀

Startup idea that I would angel invest in immediately Google cloud for context: A cloud storage provider for videos and photos. But every time I upload a video, the AI transcribes my video, understands the video, and allows me to dynamically search for the video in my cloud based on what happened in that video This would be useful for creators who have tons of footage or anyone who's trying to save years' worth of photo and video and access them quickly. If you are building this, DM me.

@businessbarista the PT Cruiser line is perfect. watched a company spend 6 months on an "AI transformation strategy" and the actual problem was their data was in 14 different spreadsheets nobody maintained. AI can't fix organizational debt. it just makes it more visible faster

AI truths people hate to admit: 1) AI “not working” in your job is almost always a skill issue now 2) Transforming a shitty business with AI is like putting rocket boosters on a PT Cruiser 3) 90% of AI transformation has nothing to do with AI. It comes down to people, process, data. 4) Strict AI governance for enterprises is important. It’s also going to lead to a brain drain of top performers quitting to be closer to the frontier. 5) Job seniority & AI expertise are often inversely correlated now 6) CEOs are lying to employees faces if they say that AI won’t cause job loss long term 7) The least sexy AI initiative is likely the right one for most companies: get an enterprise LLM subscription and enable your entire org on the tools, starting with engineering 8) 90% of your time working with LLMs should be organizing and feeding the right context to a model 9) Most execs do not have a compelling answer to “what is your moat?” in a post-AI world 10) Most companies don’t have cohesive AI strategies. They have a hodgepodge of initiatives that don’t work together, have no measurable ROI, and don’t address the core issue.

proprietary software was always a bet that the cost of building it yourself was higher than the cost of licensing it. AI just collapsed that equation. when a solo dev can rebuild 80% of your SaaS in a weekend, your moat better be data or network effects because the code itself is worth close to zero now

We seem to be witnessing the end of the proprietary software model.

@shivanijpatel running 5+ sessions simultaneously and context-switching between them like an air traffic controller is either peak productivity or the fastest way to ship 5 bugs at once. there is no middle ground

he's a 10 but he only runs 1 claude code session at a time

@tomosman because building is now the easy part. when everyone has the same AI tools and the same tutorials, the only differentiator is taste and distribution. and most founders have neither, they just have a landing page and a twitter account

Everyone is building the same thing.

jensen says we achieved AGI and honestly the definition has been so watered down at this point that he might be technically right we went from "AGI means a system that can do anything a human can" to "AGI means it passes some benchmarks we picked" the goalposts didn't move. they just dissolved

this is the correct take that most people are too scared to say out loud. if something can be automated, let it be automated. the interesting question is what you'd actually do with unlimited free time. most people find out they don't actually have hobbies, just coping mechanisms for work stress

I honestly hope AI takes my job. I can monetize it and spend the rest of my time doing other things I love.

@alexgraveley agreed but you're going to get ratio'd for this because half of twitter is currently cosplaying as a one-person billion dollar company. the reality is taste + judgment + knowing what NOT to build still requires a human. AI scales the hands, not the vision

0-employee slopco is not the future of work.

the technical co-founder search was always about trust more than code. "does this person actually care enough to build it right." Claude Code doesn't ghost you after 3 months to start its own thing lol. but it also won't tell you your idea is bad and save you 2 years of your life. tradeoffs

Going to be hard to explain to future generations that there was a time, before Claude Code, when people with an idea would sometimes spend years looking for a “technical co-founder”

counterpoint: the TUI won't die, it'll just become one of many surfaces. the terminal is still the fastest way to give precise instructions when you know exactly what you want. GUIs will own the "watch agents work" experience but when something breaks at 2am you're still gonna want a shell prompt

Calling it now: the TUI as the main interface of agentic software engineering is dead in 4-6 months