AP+ uses Codex to build behaving payment prototypes, not just clickable screens

From click-through screens to flows that behave

The most concrete frontend detail in OpenAI's writeup is the shift AP+ describes in its product development process. Early-stage testing previously relied on click-through screens: static mockups that could show what an experience might look like, but not how it would actually behave under real conditions.

With ChatGPT Enterprise and Codex, AP+ says its teams now simulate payment journeys, mobile interactions, authentication flows, and checkout experiences in environments that behave closer to real systems. The distinction matters. A screen mockup can validate layout and copy; a behaving simulation can surface how a flow reacts to timing, an authentication prompt, or a device interaction — the variables that actually move customer behavior in payments.

AP+ puts a number on the speedup: it builds working simulations in one day, down from what could previously take days to weeks. That compression is the point of interest, because it changes when feedback arrives — before significant engineering investment, rather than after.

Why behavior-level prototypes matter in payments specifically

AP+ operates payments and identity infrastructure across Australia, so its frontend concerns are not cosmetic. The company notes that customer behavior can change depending on timing, authentication prompts, device interactions, or transaction flows. A checkout that looks fine as a static screen can fail as an interaction sequence — a two-tap authentication step, a redirect, a timeout.

A prototype that only renders screens cannot expose those failure modes. One that behaves can. That is the practical argument for pulling functional simulation earlier in the cycle: it lets AP+ gather better feedback and, in its own framing, validate or invalidate thinking before committing engineering time.

Our job is to reduce risk and make better payment experience easier to achieve across the ecosystem. With AI, our teams can explore more ideas and validate or invalidate thinking faster—which means we deliver on that faster. —Jason Backhouse, Chief Operations and Delivery Officer, Australian Payments PlusMontana Labs

Adoption breadth, and the review that gates it

The prototyping story sits inside a wider rollout. AP+ reports employees have created more than 300 custom GPTs and more than 1,000 Projects, and that 77% of surveyed employees save two-plus hours a week with ChatGPT. On the frontend and technical side, Codex traced a timestamp inconsistency across system logs and reconciliation data, cutting a reconciliation investigation from four hours to about 30 minutes.

What's notable is that AP+ frames every one of these as gated by human review. ChatGPT surfaces the right eftpos specifications faster, but expert review still applies before a customer answer goes out. Simulations reduce innovation risk, but the goal is earlier feedback, not autonomous delivery. The company repeatedly describes the tools as helping specialists move faster while people stay accountable for the decision.

The implication: faster prototyping only helps if the fidelity is real

For teams building regulated, high-stakes frontends, the AP+ example is worth reading precisely — not as proof that AI compresses design timelines, but as a claim about fidelity. The value it describes comes from prototypes that behave like the system, not from prototypes produced faster.

A one-day simulation that still only clicks through screens would save time and catch nothing new. A one-day simulation that exercises authentication and timing catches problems a mockup never could. The lesson for applied teams is to judge AI-assisted prototyping by whether it raises fidelity early, not just by how quickly a first version appears.