OpenAI DevDay 2026 Introduces AI Agents, Collaborative Spaces

OpenAI introduced 'Dots,' always-on, long-running agents designed to operate across various platforms including ChatGPT, Codeex, and Slack.

These agents can function on their own virtual computer or access local devices to execute tasks, effectively serving as a user's digital delegate capable of proactive work. Currently, users on Pro, Business, and Enterprise plans have access to a single master Dot.

Dots feature unique interaction methods, including voice calls where the agent 'rings' the user, providing a more intuitive way to receive updates or task completions.

Each Dot is equipped with a dedicated virtual machine pre-loaded with a browser and developer applications, bridging the gap between cloud-based AI and local task automation, with integration into Slack allowing for persistent, team-based AI assistance.

The speaker noted initial confusion regarding the overlapping functionalities of Dots, standard ChatGPT threads, and Codeex threads, indicating that while the underlying capabilities are strong, the information architecture is still evolving.

Despite this, Dots demonstrated effectiveness in practical tasks, successfully handling online shopping, coding, and schedule management.

Proactive notifications, such as those for calendar conflicts, highlighted the agent's potential value in streamlining daily operations.

The speaker characterized Muse as more consumer-friendly and approachable, largely due to its cute avatar and intuitive interface.

The AI model powering Muse shows promise in memory handling and smart task performance, though a formal review is pending further testing to assess its exact utility within a professional stack.

The speaker generally prefers deploying specialized agents for specific tasks rather than relying on a single 'one-size-fits-all' agent.

Spaces provides a shared environment within ChatGPT, enabling humans and AI agents to collaborate on documents, artifacts, and slides.

The platform includes built-in data controls, permissions, and sharing capabilities tailored for business needs, with the speaker noting that it addresses a long-standing demand for integrated agent-human workflows in an enterprise setting.

The speaker demonstrated how their 'Dot' agent automatically created a 'personal scratch pad' within Spaces, facilitating collaborative to-do lists.

This feature allows users to directly '@mention' ChatGPT within documents to generate real-time updates, supporting a 'stream of consciousness' style of work that aligns with individual organizational preferences.

A key new feature allows users to bundle connectors and plugins, such as Snowflake, directly into their Sites, streamlining access to external data sources.

When a site is shared, the platform automatically enforces individual user permissions, ensuring that recipients only view data they are authorized to access, which addresses a significant enterprise challenge in securely deploying internal tools.

OpenAI Sites aims to become a centralized source of truth for enterprises, offering a structured environment for prototypes and reports.

The platform emphasizes key competitive factors such as data accessibility, robust governance, granular permissions, and ease of sharing, with a deeper dive into these features expected with the product lead in the near future.

This pricing positions Soul as a more economical workhorse for routine engineering tasks, with the host highlighting its reliability for everyday development workflows.

A key differentiator for this API is its native vision processing capabilities, which were previously absent in competitors like Jev.

The Decisions API has demonstrated utility in tasks such as identifying optimal frames for video thumbnails to avoid awkward poses and performing accurate object recognition.

This high-performance tier is now available in work codecs and through the API, with the host suggesting its use for demanding tasks such as clearing extensive bug backlogs or re-architecting complex systems.

While acknowledging that some latency still exists, the speaker views this capability as a significant step towards future high-frontier intelligence and its potential for truly real-time applications.

The speaker showcased a 3D application, 'Little Starship,' where user prompts instantly transformed the environment, such as adding a bathroom or shifting to a tropical island.

Live physics modifications, like disabling gravity, could also be implemented dynamically. However, running this demo for 30 minutes incurred a cost of approximately $97, underscoring the current high expense of real-time AI rendering.

Key announcements for developers included the new GPT-6 Soul model, lauded for its speed and affordability, and the Decisions API with its novel vision capabilities.

The event also highlighted the Astra UltraFast platform for its significant performance boost, alongside discussions on agent-based platforms, rumored plugin monetization updates, and the introduction of a new Pro 500 ChatGPT plan. The speaker encouraged developers to explore 'Dots' for new experiences and leverage 'Spaces/Sites' for enhanced team collaboration.