AutoScout24 puts Codex in front of 1,000 builders to speed marketplace delivery

A two-tier rollout, not a single deployment

AutoScout24 Group describes a deliberate split. ChatGPT went to roughly 2,000 employees to set a baseline of AI literacy across functions. Codex went to about 1,000 people the company calls "builder" roles — a coding agent placed directly inside engineering, data, and product workflows.

That distinction matters. The company did not treat a chat assistant and a coding agent as the same intervention. One is horizontal enablement; the other is embedded in the specific tasks where code gets written, reviewed, and shipped. Keeping them separate is what makes the reported outcomes attributable rather than diffuse.

It also chose Codex only after a three-month evaluation across teams, judged on usability, workflow compatibility, and measurable gains in productivity and code quality. That is a longer trial than most tool adoptions get, and the criteria are the ones an engineering org can actually defend.

What a marketplace frontend gets from faster cycles

AutoScout24 connects more than 30 million monthly users with over two million listings across brands including AutoScout24 in Europe and AutoTrader.ca in Canada, serving 45,000 dealer partners. That is a product surface where search, evaluation, and purchase flows are the business — the frontend is where buyers decide and dealers convert.

CTO Frederik Kraus frames the change in delivery terms rather than tooling terms:

AI is changing how we build, but more importantly, it's changing what we can deliver to our users and dealer partners. Faster iterations mean better experiences for buyers and more effective ways for dealers to reach and convert customers.Montana Labs

The headline number — select projects moving from 2–3 weeks to 2–3 days — is what shortens the loop between a product idea and something buyers can use. For a marketplace, iteration speed on the buyer-facing surface is a competitive variable, not an internal metric.

The champions network and the non-engineer prototype

The company built a cross-functional AI Champions network to create a feedback loop between central leadership and individual teams. Its stated aim was organic adoption — embedding AI into existing workflows rather than mandating a new standalone tool. One of the leadership lessons is explicit: prioritize real-world use cases over top-down mandates.

On the engineering side, Codex showed up in concrete, unglamorous work: automated pull request reviews, large-scale refactoring, technical documentation, and post-incident analysis. These are maintenance and quality tasks, not greenfield feature generation — and they are where consistency across a large codebase actually pays off.

The detail worth flagging for frontend teams is that non-technical roles began prototyping ideas and validating concepts independently. In a product organization, that pushes early UX exploration upstream, letting the people closest to buyer and dealer needs test a concept before it consumes engineering time.

The implication: measured tooling adoption beats a mandate

AutoScout24's specific contribution here is a template: evaluate an agent against engineering metrics for three months, separate broad literacy from deep workflow integration, and use a champions network to translate capability into use cases. The 10x figure applies to "select projects," not the whole portfolio — a qualifier the announcement keeps in place.

For teams building customer-facing products, the lesson is not that a coding agent makes everything faster. It is that the gains showed up in review, refactoring, and documentation — the work that keeps a fast-moving frontend from accumulating debt — and in letting non-engineers prototype before committing build capacity.