How OpenAI Teams Use ChatGPT for Ongoing Work

At OpenAI DevDay 2026, Nick and Danielle from the developer experience team present examples of colleagues using ChatGPT for recurring work. These are company demonstrations and descriptions of their own workflows, not independent evaluations.

They start with the problem of scattered information across email, Slack and dashboards. They ask what work could be handed off and what a person would do with the time saved.

The presenters interviewed three internal users: Dani in developer social, Adrian in competitive research and Casey in people operations. Their jobs differ, but each uses an agent to gather context or maintain a shared resource.

They describe a simple event-planning example in which an agenda becomes a site that helps visitors find relevant sessions, and say an agent can also take on ongoing responsibilities.

Dani monitors developer responses to product launches and reports to stakeholders. The presenters say she uses an agent to gather posts and engagement data into a live site and content sheet.

The intended result is less time spent collecting signals and more time interpreting feedback and communicating with the developer community. She also has the agent watch selected Slack channels and flag urgent items.

Adrian’s competitive research team fields questions from sales, product and strategy. The presenters describe a plugin that encodes his research method, including gathering evidence, looking for patterns and challenging conclusions.

In one example, an investigation does not support a proposed explanation for a competitor’s growth. In another, the agent reviews sales-call transcripts to identify reasons for lost deals not captured in summaries. These are reported examples, not proof every conclusion is correct.

The team uses a site called Compete Corner to make research and sources accessible. An agent can take questions from Slack, investigate and put findings in a shared space for a human to review before the site is updated.

The presenters show an interactive visualization called Model Galaxy for comparing model performance, cost and latency. It is an internal demonstration; benchmark choice and underlying data affect any comparison.

Casey’s team brings meeting notes, papers and articles into a shared knowledge base about organizational change. The presenters describe a coaching site that uses that material to help leaders consider how teams might work differently.

Page instructions describe how agents should organize new research, and the knowledge base is updated as the team adds material. The value of the coaching output still depends on source quality and human judgment.

The presenters introduce ChatGPT Space as a place where colleagues and agents can work on shared pages. In their example, teammates review research together, ask questions and add sources while their individual chats remain private.

They compare page instructions to repository instructions: they provide working practices for agents operating on that page. They also describe plans at the time for more artifact types in spaces.

A suggested starting point is to turn an existing conversation into a page, supply relevant context, refine it with an agent and invite colleagues to edit the shared result.

The presentation introduces Dots as ongoing agents with a cloud computer and browser. Dani’s example assigns a Dot responsibility for monitoring a feed, maintaining a sheet and notifying her when attention is needed.

The presenters describe reaching the same Dot through different channels and giving it a name, appearance and responsibility. They distinguish features available at the event from messaging capabilities described as coming later.

They return to the three examples: Dani spends more time with developers, Adrian shares research more broadly and Casey makes expertise accessible outside individual conversations. These are the presenters’ accounts of their colleagues’ workflows.

One presenter says automating parts of technical onboarding helped create room to move into developer experience. The talk closes by asking listeners to identify a recurring responsibility and decide what they would do with more time.