Omio is embedding OpenAI models across software development and customer-facing travel planning. The company connects more than 3,000 transport providers in 47 countries, making both engineering speed and data accuracy central to its AI strategy.
Its CTO argues against attaching technology to unchanged processes. Omio first gave employees broad ChatGPT access, then placed Codex across the development lifecycle and redesigned how products move from idea to production.
Codex becomes part of the complete engineering workflow
Engineers use Codex for research, architecture, coding, tests, review and maintenance. Internal connectors provide access to proprietary context inside development environments so teams can proceed from retrieval to execution.
Omio estimates that some products now require around 20% of their previous technical effort. Work once involving several developers for a quarter may be completed by one engineer in roughly a month.
Those figures can enable more experimentation: a team can prototype demand before committing a large budget. They also need quality context. Smaller staffing is not a benefit if hidden security or maintenance work appears later.
Productivity measures should include defects, review time, reliability and customer outcome alongside initial delivery. Every generated change needs tests and an accountable engineer who understands how it affects booking systems.
Live transport data keeps conversation bookable
Omio launched conversational travel search in 2023 and now offers a dedicated ChatGPT experience. Users can ask natural questions comparing trains, buses, ferries and flights across complex routes.
The model is grounded in current price and availability from Omio’s transport network. That distinction prevents it from inventing an itinerary based on outdated training data and allows results to lead directly to booking.
Travel contains difficult constraints: connections, station changes, baggage, accessibility and disruption. The interface should show which details are confirmed and when a provider’s live rules need direct verification.
A conversational layer can combine modes more naturally than separate search forms. It should still expose trade-offs rather than hide them behind one recommended route. Travellers may value reliability, carbon, price or comfort differently.
When a booking fails or a service changes, customers need continuity into support. The agent’s context should transfer without requiring the traveller to reconstruct the itinerary.
AI-native operation still leaves people responsible
Omio explicitly keeps human accountability for deployed code and business outcomes. Models accelerate analysis and execution but do not own the consequences of a broken route or unstable release.
That principle requires practical gates: protected production access, code review, monitoring and rollback. High velocity makes these controls more important because errors can reach customers sooner.
Expanding Codex beyond technical teams may uncover new workflows, but each function needs data boundaries and appropriate evaluation. A tool successful in code generation is not automatically suited to legal, pricing or customer decisions.
Vendor dependence also deserves planning. Internal connectors, prompts and evaluation data should remain portable enough that Omio can adopt different models if cost, quality or availability changes.
Engineers need space to learn despite compressed delivery. If every project becomes a one-person sprint, peer understanding and mentoring can decline. Rotating ownership and documenting AI-assisted decisions help the organisation retain knowledge behind rapidly produced features.
The company’s advantage lies in combining general models with a difficult proprietary network. OpenAI can provide language and coding capability; Omio supplies real transport inventory and responsibility for the journey.
That combination illustrates a mature AI product strategy. Faster development has meaning because it reaches a grounded customer experience. The conversation is useful because it connects to a ticket that can actually be bought. And the system remains credible because a human organisation stands behind both the software and the trip.
