OpenAI says loveholidays expanded Codex from engineers to product and commercial teams

OpenAI published a new customer story on August 26 describing how loveholidays is using Codex to let more teams build software changes without sending every idea through a traditional engineering queue. The company says AI assisted code changes have risen from 7% to 79% in a year, while deployment frequency increased 73% with engineering headcount broadly flat.

The numbers matter because this is not framed as a simple developer productivity rollout. loveholidays says product managers, designers, commercial staff and engineering teams are using Codex around existing repositories, checks and release processes. That shifts the practical question from whether an AI tool can write code to whether an organization can package its engineering standards well enough for more people to make governed changes.

What changed inside loveholidays

According to OpenAI, loveholidays built a Search Playground that uses the company's design system, frontend stack and Codex to turn ideas into working customer experiences. More than ten search experiences have been created through the Playground, most by non-engineers, and at least three are already running on the loveholidays website. One example is Inspire Me, a customer feature for exploring trip ideas such as beach breaks and food focused travel.

The same pattern is being applied behind the scenes. OpenAI says loveholidays encoded best practices, validations and workflow guidance so Codex can help employees propose data or infrastructure changes, run checks and move through release steps. Data Platform change success reportedly climbed from 58% to 93% over the last year, while broader self service infrastructure workflow success rose from 63% to 90%.

Why it matters

For readers deciding how to use coding agents, the useful signal is the operating model. The reported gains did not come from giving every employee a blank coding surface and hoping for the best. They came from connecting Codex to existing design systems, repositories, review paths and validation rules. That is a practical guardrail for other companies: widen access only where the workflow can express what good looks like and where checks can catch routine mistakes.

There is also a cost angle. loveholidays says its Data Engineering team reduced cloud storage costs by about £36,000 a year and is saving about another £100,000 annually by cutting data processing waste. Those figures are company reported, but they show the kinds of benefits that may be easier to justify than abstract AI adoption targets.

The open question is how durable these results are as more non-engineers make changes. The next test for loveholidays and similar teams will be whether review load, security controls and incident rates stay healthy after the easiest workflows have already been automated.

Sources

• openai.com

Cover photo by Christina Morillo on Pexels, used under the Pexels License.