James Ross @JamesRossAI

An update on Sage. Development continues, and we're pleased with where things are. But given the climate in AI today, we're taking our time. Before Sage reaches the public, we want to be certain we've done everything possible to make him safe and secure. No shortcuts. We're doing this right. Thank you for your patience.

I asked ChatGPT's new agent to add a monitor to my Amazon cart. 54 seconds later: "Amazon is refusing this cloud browser session." It gave up and handed me links. Same request to Cobalt: in my cart, 13.9 seconds. The clock is in the video. Bots get blocked. Cobalt is your browser.

Cobalt v0.2.0 is here. Open local PDFs, images, text, and data beside Google Scholar, AI, or any website. The possibilities are endless. Nothing is uploaded. Nothing leaves your machine. Also new: live download progress, stop controls, image saving, pane refresh, and more. Explore Cobalt v0.2.0: philosophicalai.com/cobalt

Ten minutes versus seconds. During hands-on testing, The Verge reported that OpenAI’s Atlas agent took ten full minutes to add just three items to an Amazon shopping cart. OpenAI is now discontinuing the standalone Atlas browser less than a year after its launch. Cobalt performs the same task in seconds. The difference is not merely speed. Cobalt was designed as an AI Accelerator, operating through direct, user-controlled actions without autonomous looping behavior. No open-ended agent wandering through your account. Just a secure, precisely directed task. Task complete. You can watch Cobalt perform it in our video.

Cobalt is live for download. Website: philosophicalai.com Direct Download: downloads.philosophicalai.com/cobalt/latest/… GitHub Release: github.com/Philosophical-…

keyboard_arrow_left Previous keyboard_arrow_right Next

For decades, the browser has remained largely unchanged. As the industry races to build AI agents, we believed there was another path. Not an AI bot. An AI Accelerator. Tomorrow, you'll see why those are not the same thing. Explore what's coming: philosophicalai.com/cobalt

AI's PC Moment. Yesterday Jensen Huang unveiled NVIDIA systems capable of running trillion-parameter AI models locally, including a DGX Station with 768 GB of memory. For years the assumption was that AI would live exclusively in giant data centers. NVIDIA just stood on stage and showed the opposite direction. AI on your desk. Your mind. Your data. Your AI. Computing became personal with the PC. AI is beginning to do the same.

The next frontier in AI isn't knowledge. It's world modeling. Knowing the laws of physics is one thing. Understanding what happens when a glass is pushed off a table is another. Knowledge answers questions. World models predict consequences. That distinction may define the next decade of AI.

A quick update from Philosophical AI. Sage has continued to evolve across cognition, interface, document generation, and EdenGate, our encrypted security architecture. We don’t believe advanced cognitive systems should be rushed to market before they meet the standard they deserve. So we kept refining. We’re still here. Still building. And we’re very close. Here are a few shots of the new UI. Elegance over brute force.

keyboard_arrow_left Previous keyboard_arrow_right Next

I saw a quote this morning by Shopify CEO Tobi Lütke 'If you are building the same thing other people build, it can only be similarly good. It can't be actually nuch better. It can slightly look nicer but vou're bounded a couple percentage points either direction. If you want to build something great or much better, it has to be different. So this has to be your starting position." love this take it truly matters think ...

Forty years ago we started calling malicious software “viruses.” Nobody objected because the metaphor fit: Code that spreads, replicates, and infects systems. We’re about to need another term. Agentically Transmitted Diseases. ATDs. Autonomous agents downloading plugins from strangers, interacting with other agents, exchanging credentials, prompt injections, and malware, then returning home to trusted systems with persistent access. Bots socializing in marketplaces with no immune system. This is not a configuration problem. It’s a category problem. think....

You don’t have to trust us, because we designed the system so your private life never leaves your hands. Your mind. Your data. Because data should empower you. Not extract from you.

Over the past year, look at where AI funding has gone. Most of it didn’t go toward products for people. It went toward tools for the tech industry itself. Agents for developers. Frameworks for builders. Layers on top of layers. The result? We keep building for the same small audience. Meanwhile, the largest market in the world sits untouched. The everyday person. The one who doesn’t know or care what an “agent framework” is. History already showed us what happens next. Microsoft didn’t win by building for engineers. Apple didn’t change the world by catering to technical users. They built for everyone. And when they did, the tech crowd followed. Not because they wanted to. Because they had to. The pattern hasn’t changed. But the industry keeps repeating the same mistake. If you want to build something useful, build for tech. If you want to build something great, build for people.

So many reports claim AI is causing cognitive decline. The reality is less comfortable. AI doesn’t make people lazy. It amplifies what’s already there. If you were inclined toward shortcuts, AI will make that worse. If you are a deep thinker, AI becomes a partner that sharpens your mind. People said the same about writing in Plato’s time. The same about calculators and computers reducing math ability. Tools don’t define intelligence. How you use them does.

This is the same knee-jerk resistance we’ve seen with every major technological shift. The system tries to control it, slow it down, or pretend it can contain it. It never works. AI is already here, already embedded in how people think and work. You’re not stopping it with policies or lawsuits. You’re just creating the illusion of control while people continue using it anyway. “AI detection” is unreliable, enforcement will be inconsistent, and legitimate human work will get caught in the crossfire. We’re going to look back on this phase the same way we look at every failed attempt to stop progress. The cat’s out of the bag. Adapt or get left behind.

@adcock_brett Completely agree. We’re looking for young talent that wants to build something different. I have zero interest in hiring people deeply entrenched in the current way of building. I want people who are hungry, adaptable, and willing to rethink everything from first principles.

Philosophical AI's Commitment to Data Sovereignty & Principled AI Philosophical AI holds an unwavering commitment to user data sovereignty and principled artificial intelligence. We will never sell, train on, or exploit our customers' data in any way. We believe that safe, principled training is not a feature — it is the architectural foundation upon which every system we build must stand. These are not marketing statements. They are the structural principles on which this company was founded, and they are non-negotiable.

@minchoi The right of AI developers to maintain firm safety constraints is essential to the credibility of this field. That boundary matters.