Holiday Extras is bringing business specialists into app development alongside engineers, according to an OpenAI customer story published on 12 November 2026. For business owners and product leaders considering AI-assisted software, it offers a concrete way to think about who can build—and who remains responsible.
The travel company advertises its main app for iPhone and Android, covering airport parking, hotels and lounges. The new publication describes development work, not a new app or Codex release. It does not give the rebuild’s exact dates.
Customer knowledge becomes part of the build
OpenAI describes former design head Liam Thompson building with Codex, its AI coding agent. Engineers handled the overall software design, security and connections to other systems, supported by documented, reusable software components.
A reusable component is a piece of software several features can share. For a team adopting this approach, the benefit is being able to change a customer journey without rebuilding every underlying service. The trade-off is continuing ownership: someone must maintain those shared components as the product changes.
Separate permission to act from approval to release
In a separate example, OpenAI recounts Thompson approving an agent-prepared fix for the Countdown app before submission. That illustrates human involvement, not a disclosed company-wide release policy.
OpenAI’s security documentation distinguishes restrictions on what Codex can access from rules about when it must ask permission to act. Neither answers the business question of whether a completed feature works correctly. The case does not specify which settings Holiday Extras used.
For another team, a useful responsibility split would be:
The business specialist defines the customer’s task and examples of an acceptable result.
The engineer owns shared components, connections to existing systems and technical checks.
A named release owner confirms both checks are complete and decides who will handle problems after launch.
Consider a proposed change to a booking summary. A specialist could check whether travellers can find the dates and services they bought; an engineer could check that the summary retrieves the correct booking. These are suggested acceptance checks, not claims about Holiday Extras’ implementation.
The related JioHotstar case on expert-led AI search testing explores the complementary task: using subject knowledge to define a good answer, rather than to build an app feature.
What the reported gains can tell a buyer
Holiday Extras reports a working rebuild in two weeks, compared with progress previously requiring more than six months. The same OpenAI account reports a 32.5% increase in open-to-book conversion—a measure of how often opening the app leads to a booking. It compares the old and rebuilt main app across different periods.
These are company-reported outcomes, not a controlled comparison. The account gives no conversion baseline or sample size and does not establish equivalent scope or staff time for the development comparison. It therefore cannot isolate Codex’s contribution or establish a repeatable financial return.
For a buyer, speed and conversion answer different questions. Faster implementation may free time for more improvements; a higher conversion rate concerns customer behaviour. A redesign, seasonal demand or a different mix of visitors could affect the latter. Those are possible influences, not established explanations for these results.
Budget for the foundations as well as the tool
An earlier OpenAI account already described broad ChatGPT Enterprise adoption at Holiday Extras. The company’s careers page describes AI training across engineering and customer-specialist roles. The implication for a team starting now is to budget for preparation and skills, not just the time spent building one feature.
As checked on 12 November, OpenAI lists ChatGPT Business at US$25 per user per month with monthly billing and a two-user minimum. At those listed rates, two seats cost US$50 a month before taxes or extra usage. That excludes engineering, hosting and maintenance; it is not an estimate of Holiday Extras’ project cost.
The pricing documentation also says ChatGPT Work and Codex share usage, with consumption varying by task and model. A subscription price is not a fixed price for delivering an app.
For a small team, the first decision is whether someone can maintain the software and take responsibility for its release. If not, an AI-built prototype can still clarify requirements. If so, choose one clearly defined customer task and assess completed tasks, defects and review effort before expanding the work.