At Holiday Extras, dots get to work before anyone asks
- ID
- 47199
- Status
- summarized
- Published
- 16 Nov 2026, 8:00 AM
- Fetched
- 17 Nov 2026, 5:57 AM
- Provider
- OpenAI News
- Category
- ai-labs
- Original URL
- https://openai.com/business/customer-stories/holiday-extras-dots
- Source URL
- https://openai.com/news/rss.xml
Summary
- Score
- 2.5
- Created
- 17 Nov 2026, 5:58 AM
- Tags
- Audience
- ai_agent_userssaas_foundersdevelopers
What happened
OpenAI's customer story describes how two Holiday Extras staff — Liam Thompson (VP of AI Innovation) and David Lee (Director of Innovation & Growth) — use personalised "dots" built on ChatGPT Enterprise, Codex and a BigQuery plugin. Thompson's dot pulled launch-performance data unprompted the day after an app feature shipped, and Lee's dot flagged API keys due for retirement by scanning his email for affected people. Thompson says he talks to his dot about two hours a day. The piece contains no pricing, headcount, adoption metrics or measured time/cost savings.
Why it matters
Treat this as a description of a product pattern, not evidence: proactive agents that read BigQuery and your mailbox unprompted are exactly the ones that need scoped credentials and an audit trail before you let them act, and this story gives you no data on cost, error rate or how the API-key list was verified. If you're evaluating ChatGPT Enterprise or Codex for a team, ask the vendor for the numbers this page omits — seats, spend, and what the dot did wrong — rather than using a customer quote as a business case.
Discussion angle
The dot read BigQuery and old email threads without being asked — what permission scope, logging and human-approval step would you require before an agent does that in your own stack, and who gets blamed when its 'people who might be affected' list is wrong?