Quick Overview
This video is a product demonstration and engineering walkthrough presented by OpenAI engineer Ali Shobeiri. It introduces dots, OpenAI's persistent cloud agents, and shows how developers can integrate them into workplace tools and operational workflows. The presentation focuses on how engineers can use these agents to handle on-call triage and routine development tasks.
Key Points
- 1.OpenAI has developed dots, which are persistent cloud agents equipped with their own cloud computer, access to the user's local machine, and unique identities across workplace tools like Slack.
- 2.Users can integrate a dot into Slack through the Codex application, enabling team members to tag the agent in conversation threads to delegate tasks.
- 3.A dot can be configured through natural-language prompts to monitor communication channels, triage incoming pages, and autonomously initiate investigative threads.
- 4.During on-call incidents, the agent can independently analyze issues, isolate root causes, and run synchronized Codex threads accessible on both mobile devices and local laptops.
- 5.Beyond on-call incident response, dots can triage channel feedback and generate or iterate on pull requests directly within Slack.
Summary
Ali Shobeiri, an engineer at OpenAI, introduces dots as persistent cloud agents. He explains that each dot is built with its own cloud computer, possesses access to the engineer's local computer, and has its own identity across workplace services such as Slack. In the demonstration, Shobeiri opens the Codex application on his laptop, clicks on his dot, and connects it to Slack. Integrating the dot gives it an identity in Slack, shown on screen as alishobeiri-dot, enabling Shobeiri or his colleagues to mention the agent in any thread so it can pick up work.
To configure an automated workflow, Shobeiri navigates to the composer interface in the Codex application. He enters a prompt directing the agent to monitor a specific channel and triage any incoming page alerts by creating an investigative thread. Shobeiri then recounts using this automation during an on-call incident at 3:00 AM. When an alert arrived, he acknowledged the page on his mobile phone and opened the ChatGPT app to inspect the dot's activity. The agent had already completed the triage, diagnosed what had failed, and generated a dedicated Codex investigation thread.
Within the ChatGPT mobile application, Shobeiri was able to view the agent's findings, enter the Codex thread, and steer the ongoing investigation by sending messages directly from his phone. Because the thread runs in sync with his local laptop, all incoming and outgoing messages appeared on both devices. This capability allowed him to resolve the 3:00 AM incident from bed without needing to open his laptop. Shobeiri notes that if an incident requires more extensive investigation, an engineer can transition from mobile to laptop, where the exact same Codex thread is active and ready for further input.
Shobeiri concludes by highlighting other engineering applications that can be configured with simple prompts. In addition to on-call triage, dots can monitor specific channels to process and triage incoming product or system feedback. Developers can also tag the dot directly in Slack discussions to have it generate pull requests and iterate on revisions within the conversation.
Overview and Identity of Dots
Ali Shobeiri introduces dots as persistent cloud agents created by OpenAI. Each dot operates with its own cloud computing environment while retaining access to the developer's local computer. In addition, dots possess dedicated identities across workplace services such as Slack, allowing developers to tag them in discussions and assign tasks that they prefer not to handle manually.
Connecting to Slack and Automating On-Call Duties
Through the Codex application, a dot can be integrated into Slack and assigned an identity, such as an engineering alias. Shobeiri demonstrates configuring an automation via the composer with a single prompt instructing the dot to monitor an incident channel, triage any incoming pages, and create dedicated threads to investigate issues.
Autonomous Incident Triage and Cross-Device Handoff
Shobeiri recounts an incident where he was paged at 3:00 AM on his phone. Upon acknowledging the alert and opening the ChatGPT mobile application, he found that the dot had already completed the investigation, identified the underlying failure, and generated a Codex thread. Because the thread synchronized with his laptop, he could steer the investigation from his phone without getting out of bed or smoothly switch to his laptop if deeper analysis was required.
Additional Engineering Workflows and Capabilities
In addition to managing on-call duties, dots support a wide range of engineering automations configured through plain text. Shobeiri highlights other practical applications, including triaging user feedback across specific communication channels, opening pull requests upon being tagged in Slack, and collaborating on code changes directly within chat threads.
The Bottom Line
The demonstration establishes that dots can act as persistent, identity-bearing cloud agents capable of autonomous on-call monitoring, triage, and multi-device interaction across Slack, Codex, and ChatGPT. It shows how routine operational duties like responding to pages or creating pull requests can be handed off using simple natural-language prompts. The video lands on the ease of steering these autonomous investigations across both mobile and desktop environments, while leaving unresolved the underlying infrastructure requirements, security guardrails, and broader deployment limitations.
FAQ
What is a dot and what capabilities does it have in engineering workflows?
A dot is a persistent cloud agent created by OpenAI that runs on its own cloud computer, has access to the user's local computer, and maintains its own identity across external services like Slack. It can pick up tasks, monitor channels, triage alerts, and interact across development tools.
How does an engineer connect a dot to workplace tools like Slack using the Codex application?
An engineer opens the Codex application, clicks on their dot, and selects the option to connect it to Slack. Once added, the dot receives its own identity and alias in Slack, allowing team members to mention it in conversation threads.
How can a dot be configured to monitor alerts and triage incidents during an on-call shift?
An engineer enters an instruction into the Codex composer telling the dot to monitor a specified channel and incoming pages. When an alert arrives, the dot autonomously triages the incident and creates a dedicated thread to investigate the issue.
How does an engineer steer an ongoing dot investigation using mobile devices and desktop computers?
An engineer can open the ChatGPT mobile application to view the dot's triage results and send messages into the investigation thread. The thread synchronizes directly with the Codex application on the engineer's laptop, allowing smooth transitions between mobile and desktop.
What additional engineering workflows can a dot automate besides on-call incident response and triage?
Beyond on-call duties, a dot can be configured to monitor specific channels and triage incoming feedback. Developers can also tag the dot in Slack to generate pull requests and iterate on code changes directly in chat threads.
Worth watching for
Software engineers, site reliability engineers, and engineering managers interested in integrating persistent AI agents into operational workflows and on-call rotations.
- openai
- codex
- ai-agents
- slack
- on-call-automation
- software-engineering