George Sibble

OpenAI just stole our entire company’s premise. Been working on this for 2 years. Not sad. Flattered. Imitation is the most sincere form of flattery. End of the line for us though.

Call it Sibble’s law: The price of artificial intelligence tokens will continue lowering until they approach or are below the marginal cost of producing those tokens (electricity). There’s no profits to be found here.

@TheGeorgePu Founder events are for the performative people who want to be founders but not grind it out. I never go. They are useless. You should be focused on delivering product and generating revenue. Founder events do neither of those things.

Go to founder events. - Everyone's pitching. - Everyone's networking. - Everyone's 'building something big.' I leave exhausted. Realized: I don't actually like most founders. I like the 5% who are honest about struggling. The rest feel like performance art.

@FrancescoCiull4 I use Claude code almost exclusively. I am very happy. I will say there was a learning curve and I learn new things every day. It requires quite a bit of technical knowledge to use properly.

Based ONLY on your personal experience, what is the best AI model to code?

The only thing you need to know about women and what they think about sex is how every female comedian makes a successful career making disgusting, crass jokes about their vaginas.

Code beats drag & drop every time. I'm willing to die on this hill.

VCs: "wow look at all these cool AI vibecoding apps! Replit! Lovable! Bolt! $0 to $100m in 1-2 years! Yeah gross margins suck but so what! They're going to be huge.... oooh wait, look at all these awesome agent companies! $10M seed round pre-product? Sure why not!" OpenAI: "ummm hello?" Foundation models and systems of record are the only ones with a moat. Everyone else is a wrapper. It sucks because many have built really thoughtful, deeply complex products, but I’m not sure that matters anymore. OpenAI's agent build looks bad *today* because it's a prototype. But it’s a shot across the bow. They’re the chocolate *and* the wrapper. Who cares if the wrapper doesn’t have pretty colors? Everyone building AI-native software should treat this as the existential threat it is.

Everyone is talking about AI taking away knowledge roles but no one is talking about Figure and Optimus talking away physical roles. We should be.

My new custom water cooled Threadripper 9980x machine Love it.

@vasuman I built a multi-agent App ecosystem a year before OpenAI. odai.chat DM me.

There’s a lot of demand for AI application engineers. People who know which models are top of line for which use cases, understand how to prompt them and chain each one together, and know enough programming to build the software around it. There’s demand because this isn’t widely taught, meaning the only way to learn is by doing, and most are lazy. If this is you, let’s chat. Show me your work below, might want you at Varick.

@svpino We also solve this by securely storing oAUTH keys. odai.chat

Unfortunately, most MCP implementations I've seen so far rely on: 1. Exposing static API keys 2. Giving agents over-privileged access to resources Everywhere I go, I see one of the following problems: 1. Persistent credentials in config files so an agent can access resources using an MCP server. 2. Nobody is doing any session management or using fine-grained permissions with agents and MCP. 3.Nobody knows what agents accessed what data because there's no visibility into any of this. Security on MCP is a huge problem. One way to solve this mess: Check out Teleport. They make MCP deployments secure, compliant, and auditable. They are sponsoring this post. Teleport gives your code/agents the same first-class identity as a human user. Here is what you get: • Unique cryptographic identity for every MCP server and AI workload • Every single request will be authenticated and authorized • Ephemeral, just-in-time credentials • Granular, scoped permissions • Every action is tied to an agent identity and resource Imagine your AI assistant with access to query multiple databases. In a typical MCP setup, you'd provide this agent with full database credentials in an environment file. If that agent is compromised, the attacker now has access everywhere. With Teleport, that same agent: 1. Requests a temporary token only for a specific query 2. Operates under strict access policies 3. Leaves behind a detailed audit trail This turns MCP from a potential security nightmare to a completely safe and secure protocol. You can see more information here: fandf.co/46rhIn9

It’s amazing that Joker came out in 2019 and was a smash hit because it resonated so much about the downfall of society, specifically in cities. And now everyone pretends it’s not happening. So odd

we need to go deeper we need an abstraction for this, like a language to write this code in like a coding language

We’ve been building this for over a year and have a better UX odai.chat

@illscience We did it first. odai.chat

An open question was answered today - "what will the AI-native distribution channel be?" It looks like ChatGPT will be that channel with 800M active users + the Apps SDK. This is likely as important as Steve Jobs announcing the app store in March of 2008 ...

@VraserX It’s an open secret he’s a power hungry sociopath. We should all be very afraid.

You can hate him all you want, but the guy’s playing 5D chess. I’ve been saying this all along, Sam Altman isn’t just building AI, he’s building the infrastructure that every other player will eventually depend on. He turned OpenAI from a company that needed others into one that others now need. Genius move after genius move, Oracle, NVIDIA, AMD, all locked in, without draining a cent from current treasuries. You don’t have to like him, but you can’t deny he’s one of the smartest strategists in tech.

i have made fun of "ai consultants" in the past i must retract those statements normal people have no idea how to prompt, we are living in a bubble

@GergelyOrosz Literally me right now. odai.com

I really don't think most people appreciate the difference between shipping something in 6 weeks, beating the global competition (first to offer an agent workflow builder!), then fixing up small issues in 6 more weeks. Vs building for 6 months, and no one cares by launch...

Incorrect. It’s best at what it has the most data about.

@no_earthquake Or that there's the largest amount of useful training data about programming and the smallest amount about taste.

@simonbs @tweethue We built this first. odai.com And our UX is better.

Hit me up @tweethue if you want to work on this together 😄

I built my first ChatGPT app! 🚀 It connects with my Philips Hue lights so I can control them directly from ChatGPT. After logging in with my Hue account, I can view the status of my lights and control them with natural language. Here’s a short demo 👇