OpenAI DevDay 2026 live blog

* The event took place on September 29, 2026, in Fort Mason, San Francisco.
* A product named "Dots" was introduced as a personal agent, visually resembling Muse agents, with user-selected names and avatars.
* Dots are powered by the Astra model.
* ChatGPT Space was introduced for team collaboration, resembling shared artifacts like Google Docs.
* Holly Li demonstrated a Dot named "Dottie" interacting with tools like ChatGPT, Slack, and Teams to build an application.
* GPT-6.1 Sol was announced, aiming for near-Astra level intelligence at a lower price.
* Ultrafast was introduced, providing 8x faster performance, available in the API, ChatGPT, and Codex.
* Codex is now fully in the cloud, along with "Codex Security Cloud" offering automated scanning and patch generation.
* The session on Research acceleration involved Tejal Patwardhan discussing model use for optimizing computer use harnesses.
* A GPT-Live session demonstrated delegation where an agent handled image generation while continuing to listen.

The narrative centers on the shift from monolithic models to personalized, agentic workflows, exemplified by the introduction of "Dots" and "ChatGPT Space." This progression suggests a path where AI moves from being a static tool to an active collaborator, embodying user-defined personalities and connecting across disparate software systems. The emphasis on Astra as the core alignment mechanism highlights the industry's focus on grounding capabilities within a trusted framework, positioning alignment not just as safety but as a prerequisite for advanced functionality.
The emergence of performance multipliers like Ultrafast and cost-optimized models like GPT-6.1 Sol demonstrates an aggressive pursuit of efficiency alongside capability. This suggests a tension between raw intelligence advancement and the practical economic and speed constraints developers face in deployment. The move towards agentic capabilities, demonstrated by the integration of tools like Codex into personal agents, signals a pivot toward automating complex, multi-step tasks rather than simple query-response.
The security developments, specifically the internal work on "The Defense Factory" and the rollout of automated patching via Codex Security Cloud, reveal that trust and reliability are central to unlocking further deployment. The underlying implication is that as AI capabilities scale, the focus must shift from raw model performance to creating verifiable, auditable execution environments. The final reflections on the job market and the role of builders suggest a future where the skill lies not just in coding, but in architecting these complex, autonomous systems and ensuring their alignment across vast operational boundaries.