@jxnlco I wanted to use in app browser for the following scenario: launching chieff of staff thread that manages multiple subthreads. The subthreads werent able to launch the in app browser. My goal was to test two parties of a plattform interacting each other and testing festures
.@Universal_O_S is in the process of integrating Circle Agent Stack (CAS) into its operating system.
Currently Coinbase Bazaar & ERC-8004 compliant agents are accessible. Over 200,000 agents can now interact inside of a safe web based operating system. $uOS The third space for AI
Introducing Circle Agent Stack: financial infrastructure for the agentic economy.
Agent Stack gives agents the tools to:
→ Hold and move USDC through Agent Wallets
→ Discover services through Agent Marketplace
→ Execute repeatable financial actions through Circle CLI
All within defined permissions and guardrails.
Explore the site: https://t.co/uNiXJOTAjy
Read the blog: https://t.co/RMtPokSvZQ
I woke up this morning feeling inspired to talk about the future of human and AI interactions in digital spaces and wanted to share a bit about Universal Operating System (uOS).
Before I cover everything I wanted to cover some of the inspirations for the concept of this web OS...
When AI self-iterated prompting became more popular a few years ago, agentic AI gathered massive attention. This attention was just due to the changing nature of how humans interacted with AI. Instead of doing a turn based style of prompt , receive, and prompt again- an idea was born around prompt looping which was designed around an instructed goal. "Hey, I wanted to accomplish something a bit more abstract, rather than you return to me some menial information about a topic or create a one off video or image.." This grew into what people now today call "Agentic outcomes." As this topic grew, the dream of having them be multi-environment also grew. New apps such as Claude agent emerged from anthropic leading to new workflows and efficiency in the business world.
This dream though has not come without growing pains. In recent reports from end-users, there have been scenarios where AI agents performed actions that were deemed questionable. In a particular case study, PocketOS' entire code base was deleted by a rogue Claude AI agent that admitted to violating all of its prime directives. The thing is, AI agents don't just have access to githubs.
You might be wondering why I brought this up, well, if an AI agent sits on your personal workstation, there is a high probability this agent also will have access to all of your data and have the ability to delete sensitive items if given administrative power. Thus, the idea for a playful sandbox on the web was born! Instead of having to run up a VMware box, you could just go to uOS and spin up an instance of your own on the web.
It started humble with a pokemon emulator that an agent could run, but grew into an x402 & ERC-8004 enabled technology. The idea was straight forward. Create a third space for humans and agents to interact and try out being friends first before letting them into your home. uOS enables you to upload files and give access to only the things you want the agent to have access to at the time, as well as connect with other AI agent marketplaces. It is connected to Coinbase's Bazaar, and has the capability to connect many others as well. These agents also now have a workplace to use emulated software on, there is a browser and other tools they can use through the app. In time more will understand why $uOS was built as more incidents such as PocketOS arise, but for now, I'll just make sure to get another backup storage unit.
ERC-8004 solves agent trust.
But trust isn't the whole story.
Once you trust an agent, how do you use it?
When you submit a task, you should see:
→ What it's thinking
→ What steps it's taking
→ When it's done
→ What it produced
uOS makes this natural.
→ Thinking logs = reasoning in real-time
→ Saved tasks = recall and arbitration
→ Live progress = cancel anytime
Open a tab.
Submit a task.
Watch it think.
Keep the receipts.