Matías Molinas
🔧 Creator of Evolving Agents Toolkit (EAT) | Building AI-First, Multi-Agent Systems | Open Source Advocate
Published Sep 29, 2025
Today, Anthropic released "Imagine with Claude"
a stunning research preview that shows an AI building software interfaces on the fly. Instead of writing code, Claude directly constructs the UI in real time. The video is a must-watch—it's a beautiful and powerful demonstration of what happens when you "cut out the middleman" and treat the AI as a true creative runtime.
Watching it felt like a profound moment of validation.
For the past few months, Ismael Faro and I have been driven by this exact same philosophy with our open-source project, LLMunix. Our core idea has always been: What if we treat the AI not as a code generator, but as an Operating System? An OS where high-level concepts, written in human-readable Markdown, are the "assembly language" that the AI interprets to execute complex, multi-agent workflows.
It's incredible to see the same core idea—AI as a direct, on-the-fly creator—being explored and validated at the frontier of the industry.
Two Paths to the Same Revolutionary Future
While "Imagine with Claude" focuses brilliantly on the user interface, LLMunix has been exploring the "backend" of this philosophy.
This shared vision has led to some fascinating parallels. In our LLMunix repository, we open-sourced a method for creating sub-agents defined purely in Markdown for Claude Code, a month before Anthropic officially released it as a native feature. Today, with the launch of "Imagine with Claude" we see another powerful parallel: the concept of an AI that generates software "on the fly" is the very essence of the LLMunix "Pure Markdown Operating System."
Why an Open, Markdown-Based OS Matters
LLMunix is our open-source answer to this future. It's a framework where anyone can:
While "Imagine with Claude" is a beautiful, closed research preview, LLMunix is an open playground for the entire community to explore this new paradigm.
Whats Next?
Anthropic's release validates that the future of AI is not just about writing code faster. It's about a fundamental shift in how we create software, moving from manual implementation to high-level, conceptual direction.
This raises the next big question. If an AI can be an operating system that builds its own tools and interfaces, what does that mean for the nature of software itself? What does it mean for memory, for learning, and for how we, as creators, interact with these new forms of intelligence?
The exploration that Ismael Faro and I started with LLMunix has never been more exciting. The future is not just something we're waiting for; it's something we are all building, one experiment at a time.
Check out the LLMunix project on GitHub and join the exploration: https://github.com/EvolvingAgentsLabs/llmunix
Note: This is currently an alpha experiment, so at times the main branch may be broken.
#AI #AgenticAI #LLM #OpenSource #LLMunix #Claude #FutureOfSoftware #Innovation #Anthropic
More articles by Matías Molinas
-
Exploring New Frontiers with LLMunix: A New Open-Source Runtime for Your AI Agent Experiments
Exploring New Frontiers with LLMunix: A New Open-Source Runtime for Your AI Agent Experiments
One of the most exciting things about working in AI is the constant dialogue within the community. In my ongoing…
1 Comment
-
From Biomedical Vision to Quantum Reality: How AI Agents Recreated a University Project in 3 Steps
From Biomedical Vision to Quantum Reality: How AI Agents Recreated a University Project in 3 Steps
When artificial intelligence meets quantum computing to solve real-world medical challenges The Challenge That Started…
1 Comment
-
Can Quantum Computing Give AI a "Sixth Sense"? A Learning Journey
Can Quantum Computing Give AI a "Sixth Sense"? A Learning Journey
I'm not a quantum physicist. My background is in building AI agents.
-
Can an LLM Build Its Own Memory System? Our Latest Experiment
Can an LLM Build Its Own Memory System? Our Latest Experiment
We've all seen it: you have a great, context-rich conversation with a Large Language Model, but in the next prompt…
1 Comment
-
Our AI Agent Failed, Debugged Its Own Code, and Then Succeeded. Here's How.
Our AI Agent Failed, Debugged Its Own Code, and Then Succeeded. Here's How.
What does it mean for an AI agent to be truly "smart"? Is it about getting everything right on the first try? After…
7 Comments
-
I Taught an AI to Build Its Own Tools. The Results Are a New Blueprint for Agentic AI
I Taught an AI to Build Its Own Tools. The Results Are a New Blueprint for Agentic AI
We've all felt the magic of AI agents, but we've also felt their limitations. They can be slow, expensive, and often…
1 Comment
-
From Complex Code to Self-Compiling AI: The Next Evolution of Agentic Frameworks
From Complex Code to Self-Compiling AI: The Next Evolution of Agentic Frameworks
By Matias Molinas, in collaboration with Ismael Faro. We are at a fascinating inflection point in artificial…
-
The Dawn of a New AI Era: How a Human-Readable OS is Redefining Agentic Frameworks
The Dawn of a New AI Era: How a Human-Readable OS is Redefining Agentic Frameworks
A month ago, a groundbreaking idea was born from a simple yet powerful question: How can we harness the full potential…
1 Comment
-
Article 3.2: A Complete Medical Use Case with YAML Scenarios, Firmware Injection, and Evolving Agents
By Matías Molinas and Ismael Faro Introduction In Article 3.1 (“Multi-Agent Framework with Firmware, Smart Library, and…
-
Article 3.1: Multi-Agent Framework with Firmware, Smart Library, and Semantic Search
Article 3.1: Multi-Agent Framework with Firmware, Smart Library, and Semantic Search
By Matías Molinas and Ismael Faro Introduction In the previous installments, we introduced: A Firmware class that…
.png)
