The Public Option: Why Open-Source AI Needs More Than Just Code
AI-generated image · US National Wire
As proprietary giants like OpenAI automate the engineering process, a new non-profit effort backed by $400 million seeks to map and build a truly open alternative to the corporate AI stranglehold.
The current trajectory of artificial intelligence is one of extreme concentration. While the industry celebrates leaps in productivity, the underlying architecture—the weights, the datasets, and the infrastructure—remains largely locked behind the gates of a few well-funded corporations. To break this proprietary stranglehold, we don't just need more open-source libraries; we need a well-funded public option for AI.
**Opinion: The Case for a Public Option**
In my view, the only way to ensure that AI weights actually belong to the people is through a concerted, heavily funded effort to create a public alternative. When the tools of creation are owned by a handful of firms, the direction of the technology is dictated by corporate roadmaps and profit margins. A public option provides the necessary counterweight, ensuring that the foundational building blocks of the next industrial revolution are treated as a utility rather than a proprietary secret.
Recent developments highlight both the necessity of this shift and the daunting scale of the challenge. On one hand, we see the sheer efficiency of the proprietary model. Reporting from OpenAI, published April 27, reveals the launch of Symphony, an open-source spec for Codex orchestration. Developed by Zach Brock, Victor Zhu, and Alex Kotliarskyi, Symphony transforms project-management boards—specifically the tool Linear—into a control plane for coding agents.
OpenAI reports that by treating Codex as a teammate and using Symphony to orchestrate agentic work, some of their teams saw a 500% increase in landed pull requests within the first three weeks. The system allows agents to run continuously, pulling tasks from a tracker, restarting if they crash, and even filing their own follow-up issues for refactoring or performance improvements. This level of automation effectively removes the human bottleneck of "context switching," allowing engineers to manage complex infrastructure migrations or React upgrades by simply defining dependencies in a directed acyclic graph (DAG).
However, this efficiency is a double-edged sword. As OpenAI demonstrates how to build an "agent-friendly repository" where every line of code is generated by AI, the gap between those who own the models and those who merely use them widens. If the most efficient way to build software is through a proprietary orchestrator and a closed-weight model, the open-source community risks becoming a secondary ecosystem of wrappers rather than a primary engine of innovation.
This is where the concept of a "public option" becomes critical. As reported by Simon Willison’s Weblog, a non-profit organization known as "Current AI" was founded during the AI Action Summit in Paris in February 2025. Current AI is positioned as a global partnership dedicated to building a public option for AI, and it arrives with serious financial backing, including $400 million already committed.
To understand where the open-source community stands against the proprietary giants, Current AI recently released its "Gap Map v0.1." This index serves as a census of the current open-source AI landscape. According to the data cited by Willison, the Gap Map identifies 421 in-depth products produced by 228 different organizations. These are broken down into:
* **266 software tools and libraries** * **85 models** * **50 datasets** * **20 hardware projects**
These products are organized across three layers of the stack: infrastructure, model components, and product/UX. While these numbers show a vibrant ecosystem, they also reveal the "long tail" of the industry. Willison notes that there are another 24,400 artifacts in the open-source AI ecosystem that remain uncategorized and unscored, awaiting further research.
For those interested in the raw data, Current AI has released the underlying information under an MIT license via the currentai-org/os-ai-map GitHub account, consisting of 1,184 YAML files, notebooks, and schemas. The project is currently tracking 16,185 GitHub repositories.
The contrast is stark. On one side, you have the corporate efficiency of Symphony, which allows OpenAI engineers to make significant code changes from a phone in a cabin on poor wifi because the agentic orchestrator never sleeps. On the other, you have the fragmented, though growing, effort of the open-source community to map its own existence.
If we want a future where the weights are downloadable and the technology is transparent, the $400 million committed to Current AI is a start, but it must be viewed as the baseline. The goal is not merely to match the productivity of proprietary systems, but to ensure that the infrastructure of intelligence is a public good. Without a well-funded public option, we aren't building an open future; we are just renting one.