loveholidays saves 2,000 hours a week by making everyone a builder with Codex

From isolated experiments to concrete savings

While most companies are still experimenting with AI, UK online travel agency loveholidays has already made it the center of its operations. By utilizing OpenAI Codex and other AI models, the company saves an average of 2,000 hours of work per week. This isn't just about developers writing code slightly faster. According to Dmitri L., Head of Engineering, Codex has enabled the realization of projects that wouldn't have been viable previously.

This is well illustrated by the data team, where AI helped identify optimizations saving £36,000 annually. Before Codex deployment, this opportunity was shelved because it required too much manual work with low priority. With an AI assistant capable of quickly processing data, an initially unattractive task became a profitable venture.

Who sets priorities when execution becomes cheaper

For software teams, this marks a fundamental shift in how they evaluate what is worth building. When execution costs drop, the backlog of projects there was "no time for" suddenly opens up. The engineer no longer serves merely as a person translating requirements into code, but rather as an orchestrator of agents and reviewer of their work.

Loveholidays didn't reach this state overnight. They created a dedicated AI taskforce and systematically upskilled employees across the entire company, not just a narrow group of specialists. They taught them how to properly formulate problems for models and how to verify outputs. This turned every team member into a potential builder of new internal solutions.

Results are delivered, security questions remain

A productivity increase of 2,000 hours a week and savings in the tens of thousands of pounds sound great, but rapid AI adoption also carries risks. Codex and other tools rely on inputs and can make mistakes. Companies must have strong control mechanisms to catch hallucinations or faulty code before it reaches production.

At the same time, the question of vendor lock-in arises. Relying on a single ecosystem (e.g., OpenAI) might yield quick results, but it makes the company vulnerable to pricing changes or service outages. The success of loveholidays doesn't mean the path is without obstacles.

The cleanliness of the review queue will show if teams are serious

The proof that integrating AI agents into workflows truly works in the long run won't just be the volume of generated code. It will be how effectively humans can review and manage this code.

If review queues become bottlenecked with automatically generated code that no one wants or has time to verify, velocity will ultimately slow down at the human oversight chokepoint. Successful companies will be those that not only know how to generate solutions but also have processes for their rapid deployment to production.

Lilith's verdict

The loveholidays case study demonstrates what happens to a company's priorities when the cost of executing previously shelved ideas drops to a fraction. All the old equations about what isn't worth touching have to be rewritten.