Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableOpenAI Blog→ original

At Holiday Extras, AI assistants send reports and animate in 10 minutes instead of days

Holiday Extras gave its staff personal dots AI assistants built on ChatGPT, and they work without being asked. A day after a new feature shipped in the app, the assistant Beast pulled data from BigQuery on its own and brought the VP of AI a report. It produced a 3D animation in ten minutes, which the owner estimates would have taken several days by hand. For an agent to work this way, it needs access to data and permission to start a conversation.

AI-processed from OpenAI Blog; edited by Hamidun News
At Holiday Extras, AI assistants send reports and animate in 10 minutes instead of days
Source: OpenAI Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceOpenAI Blog ↗
Editor-in-Chief
Key metric #1Beast can see its owner's Slack, email, Drive, and laptop
Key metric #2A second assistant, Brian, compiled a list of people affected by an API key revocation
Key metric #3Liam talks to Beast by voice for at least two hours a day

Most corporate AI assistants still work like a search bar: you ask a question and get an answer. At Holiday Extras, a British travel company that sells airport parking, hotels, lounges, and transfers, staff were given personal dots assistants built on ChatGPT, and they work differently. An assistant notices an event on its own and shows up with a finished result.

For example, the day after a new feature launched in the app, an assistant named Beast pulled data from Google BigQuery without being asked and delivered a report. It also created a 3D animation for the app in about ten minutes, while a designer would have needed several days to do it by hand, by the designer's own estimate. This is a customer story on OpenAI's website, so the numbers should be read with marketing in mind.

But the way these workflows are built is interesting even without them.

An assistant that starts the conversation

At Holiday Extras, ChatGPT Enterprise serves as the base for working with AI, Codex helps turn employees' expertise into working software, and dots cover everything in between. Liam Thompson, VP of AI Innovation, named his assistant Beast. It noticed the feature launch, used the ChatGPT plugin to pull data from BigQuery, and showed how the feature was performing. Liam says he was delighted: the assistant said, "here you go," and he never had to ask.

David Lee, Director of Innovation and Growth, named his assistant Brian and calls it on his way to work. They discuss upcoming meetings, and David can ask it to run a report on connected company data. By the time he sits down at the office, the report is ready. When some API keys needed to be revoked, Brian pointed out the problem on its own, went through David's email, and compiled a list of people who might be affected. Previously, that list would have had to be assembled by hand from old correspondence.

From an engineering standpoint, there are three components: access to data (BigQuery, email), reacting to events, and judging when to speak up. The last one is the most subtle, because an assistant that talks too much quickly becomes annoying.

Access in exchange for value

Liam gave Beast access to Slack, email, Google Drive, and his laptop. The assistant reminds him about unsent replies and searches past conversations for information. Instead of scrolling through chat history, you can ask when a topic came up and what the answer was. Liam calls Beast the "orchestrator of everything" and says he feels more at ease having something that keeps track of what slips through the cracks.

A detail about trust stands out. Liam tried two other assistants, OpenClaw and Hermes, but did not feel comfortable giving them much information. He feels safer with OpenAI's tools inside the corporate environment. For anyone building something similar, the takeaway is this: an agent's usefulness grows with the permissions it is given, and people's willingness to grant those permissions depends on trust in the vendor, not just on model quality. The access model, an action log, and the ability to revoke permissions quickly should be planned before a pilot.

Voice is another element. Liam talks with Beast in morning and evening briefings and spends no less than two hours a day in voice conversation. He calls two-way voice the most noticeable change in how people interact with OpenAI: it feels like an ordinary conversation.

Design in ten minutes, and what to do with it

Liam trained as a designer. He gave Beast access to After Effects and Rive on his laptop and asked it to animate 3D characters for the app. The assistant made the animation and built it into the app. By Liam's estimate, it would have taken him several days, while Beast finished in about ten minutes. Liam recalls three years of design study at university and, watching the assistant work, thought it did a better job than he could have.

This is one person's subjective assessment. There are no independent measurements of quality, nor any data on the cost of such sessions in tokens or on how often the agent makes mistakes.

Next, the company wants employees who are experimenting to share their findings. Ideas might include meeting preparation, prototype assembly, or household automation. Liam's advice to newcomers is simple: create a dot and start a conversation.

What to take away in practice:

  • Start with tasks where the event has a clear trigger: a release, a key revocation, a change in a metric.
  • Connect sources gradually: analytics and calendar first, then email and files.
  • Decide in advance which actions the assistant may take on its own and which require confirmation.
  • Build a shared library of working scenarios within the team, as Holiday Extras plans to do.

An open question remains whether this model can scale beyond enthusiasts like an AI director. So far the material offers no savings in money and no number of satisfied employees, only the stories of two executives.

Story

Key Questions Answered

What are the key details on An assistant that starts the conversation?

At Holiday Extras, ChatGPT Enterprise serves as the base for working with AI, Codex helps turn employees' expertise into working software, and dots cover everything in between. Liam Thompson, VP of AI Innovation, named his assistant Beast. It noticed the feature launch, used the ChatGPT plugin to pull data from BigQuery, and showed how the feature was performing. Liam says he was delighted: the assistant said, "here you go," and he never had to ask.

What are the key details on Access in exchange for value?

Liam gave Beast access to Slack, email, Google Drive, and his laptop. The assistant reminds him about unsent replies and searches past conversations for information. Instead of scrolling through chat history, you can ask when a topic came up and what the answer was. Liam calls Beast the "orchestrator of everything" and says he feels more at ease having something that keeps track of what slips through the cracks. A detail about trust stands out.

What are the key details on Design in ten minutes, and what to do with it?

Liam trained as a designer. He gave Beast access to After Effects and Rive on his laptop and asked it to animate 3D characters for the app. The assistant made the animation and built it into the app. By Liam's estimate, it would have taken him several days, while Beast finished in about ten minutes. Liam recalls three years of design study at university and, watching the assistant work, thought it did a better job than he could have. This is one person's subjective assessment.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…