The dots demo, take two | OpenAI DevDay 2026

**TL;DR:** At OpenAI DevDay 2026, the keynote speaker re-ran the DOT demo that didn't go as planned the day before — using a live phone call to Dot, he demonstrated on stage how Dot could simultaneously handle travel arrangements, analyze user feedback, and draft Slack message replies, all as real work. ## Why He Re-Ran the Demo The launch demo didn't go according to plan yesterday, and the speaker admitted he was "a little disappointed, because I really, really wanted to show you what DOT can do" — so he replayed this segment in full today. The focus of the demo was showing **how to handle real work, real launches, real things**. DOT genuinely participated in the launch, and the speaker genuinely had a lot on his plate, so the demo was presented "raw and unfiltered" — showing how he normally uses his own Dot day to day. The demo kicked off with that "famous phone call to Dot." ## On Stage: A Phone Call to Dot On the call, the speaker told Dot that he had to head to LA on short notice the next day, felt a bit stressed, and wanted to see if Dot could pull all of this together. His specific requests: - He needs to be at an important meeting in LA the next day, arriving in Hollywood by around 3 PM - He estimated he'd need to leave the office by about 10 AM at the latest - Book a flight that **complies with company travel policy and is refundable** - Provide hotel options near East LA - He needs screenshot photos of the hotels to help him pick - He needs screenshots of the hotels' locations on **Google Maps** to evaluate whether he could meet with people nearby ## Dot's Response and Execution ### Initial Flight and Hotel Results Dot immediately got to work and confirmed it would ensure the flight was both policy-compliant and refundable. The summary included: - **Outbound flight**: Southwest, departing at 11:30, arriving at 12:55, refundable fare around **$326** - **Hotel candidates**: Silver Lake Pool & Inn, Innsowan, and a popular choice closer to Hollywood, Cara Dot noted that the hotel total price and policy compliance still needed verification, and pulled up hotel photos and map screenshots, sending them to the chat. ### Travel Time Assessment The speaker followed up: if the flight lands at 1 PM, can he reach East Hollywood before 3? He mentioned he wasn't really sure how LA traffic works. After checking, Dot gave the following estimate: - Assuming an on-the-schedule flight, allow **30 to 45 minutes** to leave LAX - The trip itself could take **45 to 75 minutes** - Conclusion: landing before noon would give him more comfort At this point, Dot found an even better Southwest option, landing at **11:20** — giving a much larger buffer — and was verifying whether a policy-compliant refundable fare was still available. The speaker ultimately decided to aim for "landing as close to midday as possible" and selected **Silver Lake Pool & Inn** as the hotel, with Dot preparing a one-night booking and verifying the total price and cancellation terms. ## Second Task: Analyzing the User Feedback Pipeline While Dot handled the travel logistics, the speaker raised a second request: there was far too much in the user feedback channel, and he wanted a more specific collection process. He most wanted to know three things: 1. What are the **top few themes** across the feedback? 2. Are there already solutions in place for **the most pressing issues**? 3. If not, what **unsolved problems** has the team not noticed? ### Current Feedback Collection Process Dot reported on the current state: - Right now it's **manually processing new channels in batches**: reading the reports and replies in the channel, adding each one to a master list, categorizing and grouping them - Linking actual issues to **verified Linear tickets** - Matching tickets to **PRs**, tracking whether a fix is in progress, released, or has been re-tested - A **dashboard** aggregates all of this work - But the dashboard **does not automatically pick up new messages**, so ticket reconciliation remains incomplete ### Closing the Loop Requirements The speaker confirmed the manual process itself was acceptable, but he emphasized two things: 1. Identify what **major issues currently have no solution** 2. When a **PR is merged** and the team believes it addresses a major theme from user feedback, actually **go into Slack and reply to users**: include a PR link and let them know they can try again Dot summarized this as a "gaps + replies" closed loop, and said it would surface the biggest issues with no credible fix, reply in the relevant threads with the PR link, and ask everyone to retry once the feature ships. The speaker further stressed that **re-testing** is the step that truly confirms the loop has been closed, and that this part hadn't been reliable before. Dot committed to prioritizing themes that currently have no fix. ### Feedback Themes from the Past 24 Hours Dot reported on the most prominent feedback theme right now: **reliability issues** — Dot appears active but doesn't respond, or can't access its tools. Specifically: - Messages go unanswered - Tool access fails - Progress is unclear Dot added that some of these were genuine failures, while others were just **visibility issues** — people aren't sure whether their Dot is actually working. It also noted it was reviewing the full ranking and would confirm further before concluding that this was the highest-volume issue. ## Third Task: Summarizing Slack Replies The speaker mentioned there had been a lot of activity on the marketing side lately, and people were curious about where DOT fits alongside products like ChatGPT, Slack, and Teams. So he asked Dot to: - Summarize all **Slack messages that haven't been replied to yet** - **Draft replies** for him Dot immediately got to work. ## Demo Summary As the demo moved into its later portion, the speaker summarized what Dot was doing simultaneously: - It booked his flight and found hotels, the hotel photos looked good, and he'd already picked one - It understood the feedback pipeline the team was implementing, and would **prioritize** what he needed to deal with next - It was sending the summary of feedback themes — centered mainly on **reliability**, with mobile and notifications being mostly UX polish At the end of the demo, the speaker said: "I'm just going to say, let's re-run the feedback pipeline, and see what comes out." Source: https://www.youtube.com/watch?v=fHEIw5CcN5U