Q&A with OpenAI President Greg Brockman about OpenAI's research direction, how far it can push Codex, closing Sora, betting on text vs. world models, and more
OpenAI is shifting strategies yet again. Here's the logic behind the latest moves and what they mean for the company's direction.
Context & Ripple Effects
OpenAI had already framed product and enterprise priorities urgently in Sam Altman’s December “code red” discussion of enterprise strategy and product ambitions. Brockman’s comments make the research choice consequential: the company is narrowing attention around text-based capabilities and coding while ending a named video product.
The decision also sits at the start of a broader organizational arc. Subsequent coverage described OpenAI’s restructuring alongside the Sora cut, followed by a plan to place ChatGPT, Codex, and the API in one core product team.
First-order effects
- Sora users lose access to the service, while OpenAI can redirect product and research attention away from that offering toward text-based models and Codex.
- Codex becomes a more central test of how far OpenAI can turn frontier-model research into a developer-facing product, rather than one experiment among several separate surfaces.
Second-order effects
- A narrower OpenAI roadmap increases the importance of coding and text-model performance for customers choosing its products; video and world-model work no longer has the same visible product outlet inside OpenAI.
- The later combining of ChatGPT, Codex, and the API under a core product team suggests that research priorities can increasingly be expressed through a shared product and developer platform rather than standalone services.
Third-order effects
- If this allocation persists, frontier labs may compete less on maintaining every model modality as an independent product and more on concentrating investment where distribution, developer adoption, and model improvement reinforce one another.
- The pattern points toward AI workspace consolidation: research portfolios become organized around a smaller set of integrated user and developer surfaces, though OpenAI’s stated text preference alone does not establish an industry-wide retreat from world models.
The trend: This is one data point in the consolidation of frontier-AI research and product portfolios around broadly distributed text, coding, and developer platforms.