It has been over a year since MoFA was founded. MoFA is an Agent Framework — our goal is to achieve Composition AI through MoFA. By combining the capabilities of developers, we aim to build Super Agents, putting AI into the hands of ordinary people and empowering them to do extraordinary things.
Reflection
Not long ago, Anthropic co-founder and CEO Dario Amodei said at the Davos World Economic Forum that coders as a profession have about six months left. The world is undergoing profound changes at an unprecedented pace. The only limit seems to be energy. Even foundational concepts need to be rethought with a skeptical eye.
In recent months especially, the programming capabilities of large language models, Vibe Coding, and AI software engineering have been nothing short of stunning. We realized that MoFA’s vision of empowering people is already being achieved by new AI software engineering. MoFA’s original aspiration remains beautiful, but its methodology — rallying developers to build small modules for an Agent Hub, then assembling Super Agents — now appears quite traditional and overly modest. The emergence of AI programming tools like Claude Skills has made the tools, sample applications, documentation, and even ecosystems that once required enormous effort and resources no longer necessary. People can now focus more on the core, without worrying about applications.
While new AI programming is revolutionizing traditional coding, new challenges are also emerging:
-
Quality Assurance Dilemma: When a system contains 100,000 lines of AI-generated code, how do you ensure quality? Traditional code review methods fall short.
-
Rapid Technical Debt Accumulation: The ease of development leads to architectural design being neglected, turning systems into black boxes that “work but are hard to maintain.”
-
Homogenization Risk: Everyone uses the same AI tools, and generated code starts to converge. Product innovation may be overshadowed by homogenization in implementation.
The New MoFA
With a selfless attitude, we strive to learn and adapt to the evolution of AI technology, seeking harmony between developers and AI.
MoFA is positioned as an Agent Framework. We’ve found that software frameworks still hold value. However, we must recognize that our framework is no longer written for humans — it’s written for AI. As seasoned architects say: “The best framework is one that’s invisible to the user.” With AI’s assistance, this has become a basic requirement. We even believe that in the AI era, good frameworks are more important than ever. AI-written software is still built on top of various frameworks. If even the lowest-level foundations of software are built by AI, wouldn’t AI be completely out of control?
Although not yet officially released, the new MoFA has taken shape. Its core is built on Rust (we call it mofa-rs), achieving ultimate performance with native multi-language bindings through UniFFI. It is a modular, extensible multi-agent framework built on a microkernel architecture. Staying true to our vision, the new MoFA supports the Dora-rs distributed runtime, cross-process and cross-machine Agent communication, distributed and edge computing, continuing on the path of Compositional AI.
On the new MoFA framework, we are experimentally developing applications such as MoFA.FM (an AI podcast platform) and MoFAClaw (a mofa-rs-based OpenClaw alternative). We hope to validate that the new MoFA framework can provide an efficient, stable, highly controllable, and observable foundation for multi-agent applications.
Although we can hardly navigate this leap in AI Coding technology with perfect grace, that doesn’t stop us from stumbling our way toward the future.