Put your dot on call

# Put Your dot on Call: Engineering Workflow Practices with OpenAI Engineer Ali **TL;DR: OpenAI engineer Ali demonstrates how to connect "dot" — a persistent cloud agent with its own cloud computer, local machine access, and Slack identity — to Slack to automate workflows like alert triage and unattended incident investigation at 3 a.m.** ## What Is dot Hi everyone, I'm Ali, an engineer at OpenAI. I'm excited to show you how I use dots in my engineering workflows. We built dots to be a **persistent cloud agent** with three key capabilities: - It has its own cloud computer - It can access your local machine - It has its own identity in the services you already use, such as Slack In other words, dot isn't just a throwaway chat window — it's a long-lived agent that spans both cloud and local environments, and has its own presence in your team collaboration tools. ## Connecting dot to Slack To show what dots can do, I open the Codex app, click on my dot, and connect it to Slack. Once it's added to Slack, my dot has its own identity there. I can then @-mention it in any conversation thread to hand off work I don't want to do, or anything else it can handle — it shows up in the conversation just like any other team member. ## Setting Up On-Call Automation with a Single Prompt With the connection established, I open the composer and do something really cool: set up a workflow for my on-call shift. My prompt is simple: > Monitor the channel, and whenever a new alert notification comes in, triage the alert and create a thread to investigate. That's it — the automation is deployed. No scripts to write, no pipelines to build. A single natural-language prompt does the job. ## A Real 3 a.m. Incident, Walked Through Here's how it helped me handle a real on-call incident at 3 a.m. The story goes like this: my phone buzzed with an alert notification. I picked up my phone, acknowledged the alert, and then opened the ChatGPT app to see what my dot was up to. The results were impressive: - **The investigation was already done** — by the time I opened the ChatGPT app, my dot had already figured out what the problem was - **A Codex thread was created automatically** — I could tap into it, start guiding the conversation, and send follow-up messages - **Messages sync across devices** — anything I sent from my phone was also synced to my laptop Even better, the Codex thread was running on my laptop, not only helping me understand what happened but also meaning **I didn't have to get out of bed to grab my laptop and start investigating** — because it had already done the investigation for me. In that particular case, I was able to resolve the issue right from my phone. And if something had required deeper investigation, I could have set down my phone, walked over to my laptop, and picked up the **exact same Codex thread** right where I left off. This is what "persistent" really means: one continuous session, one shared context, seamlessly moving between phone and laptop — the human can step in at any time, or step back just as easily. ## More Ways to Try It This is just one small feature of dot. Beyond on-call alerts, you can set up other automations with a simple prompt, such as: - **Auto-triage all feedback received in a specific channel** - **@-mention your dot in Slack to have it create a PR**, then collaborate with it iteratively right in Slack There are many other use cases — I'm excited to see what you all try. ## Key Takeaways 1. **A persistent cloud agent**: dot has its own cloud computer, can access your local machine, and has its own identity in services like Slack. 2. **Automation with a single prompt**: no scripting needed — describe the workflow in natural language (e.g., "monitor the channel, triage alerts, create an investigation thread") and dot gets to work. 3. **Seamless cross-device continuity**: dot's investigation results and Codex threads sync between phone and laptop — engineers can acknowledge alerts from bed, then go deeper from their desk. 4. **From doing work for you to working with you**: on-call investigation, feedback triage, creating PRs and iterating — all of it can be delegated to dot as if it were a member of your team. Source: Put your dot on call — OpenAI (https://www.youtube.com/watch?v=GBOZIRK1ujw)