OpenAI’s own tools now generate about 80% of the code the company produces, co-founder Greg Brockman said this week. In the same remarks, Brockman estimated the firm is 70-80% of the way toward achieving artificial general intelligence — a milestone that would let AI match or exceed human capability across most economic tasks.
The code-writing shift
Brockman’s disclosure offers a rare look at how deeply OpenAI relies on its own technology. The company builds large language models and other AI systems; those same systems are now doing the heavy lifting for internal software development. Engineers still review and tweak the output, but the bulk of new code comes from the machines.
That 80% figure is a sharp increase from just a couple of years ago, when automated code generation was still experimental. It reflects a broader trend in the industry: firms from Google to Microsoft are betting that AI can handle routine programming tasks, freeing human developers for higher-level design work.
The AGI timeline
Brockman’s AGI estimate is the most specific the company has offered publicly. He put progress at “70-80%” without giving a precise target date. OpenAI has long stated its mission is to build safe AGI, but the path has been opaque. The co-founder’s comments suggest the company sees the finish line as closer than many outsiders assumed.
AGI is a disputed concept. Some researchers argue it’s a decade away; others say current models already show signs of general intelligence. Brockman’s confidence aligns with OpenAI’s aggressive rollout of products like GPT-4 and DALL-E, which have pushed the boundaries of what AI can do.
Sovereign AI and national policy
Brockman also touched on “sovereign AI” — the idea that nations need their own AI infrastructure to maintain control over data and decision-making. He said the concept is reshaping national policy as governments race to build or buy domestic AI systems. Countries like India and Japan have announced plans for national AI models, and the European Union is crafting regulations that could force non-European firms to tailor products for local markets.
For OpenAI, sovereign AI presents both an opportunity and a constraint. The company could license its models to governments, but it may also face restrictions on where data can be processed and how models are trained. Brockman did not outline specific deals or proposals.
The intersection of AI code generation, AGI progress, and geopolitics is likely to dominate the next wave of industry discussion. OpenAI’s own tools are already shaping the code that will power future systems — including, potentially, the AGI itself.
