At Revolut, one engineer built a code review service with Codex, and seven years of design docs got AI-ready in a day
Revolut showed how small teams with Codex, OpenAI's AI assistant for programmers, can take on projects that would normally need a whole department. One engineer built an internal code review service on their own, a tool for checking others' code before release. Designers turned seven years of documentation into a format AI can understand, in a single day. The takeaway: first put the team's knowledge in order, then hand work to an agent.
AI-processed from YouTube: OpenAI; edited by Hamidun News
Conventional wisdom says that if a project looks like a task for a whole team, you need to assemble a whole team. In a fresh case study published by OpenAI, Revolut, one of Europe's largest fintechs, shows the opposite. Using Codex (OpenAI's AI assistant that writes and edits code on human instructions), one of the company's engineers alone built an internal code review service, a system for checking others' code before release.
Meanwhile, designers turned seven years of accumulated documentation into a format an AI model can read, in a single day. This is marketing material from OpenAI itself, and it contains almost no savings figures. Still, the three scenarios it describes are useful as a model for where to point AI tools inside a company.
A code review service built by one person
In the first story, a Revolut engineer used Codex to build an internal code review service on their own. Normally, a project like this would be a full team task: designing integration with repositories, review rules, a developer-facing interface, and ongoing support. Here one person took all of that on, while Codex worked as a partner that writes and edits code.
Why this matters for practitioners: code review is a typical bottleneck in any growing engineering organization. Pull requests keep piling up, while the number of experienced reviewers stays flat. An internal service tuned to a company's specific rules can take over part of the routine work. However, the description gives no data on how much the service cut review time or how many bugs it catches. The story should therefore be read as proof that something is possible, not as a measured result.
Seven years of design documentation made AI-ready in one day
The second story gets less attention but is perhaps the most transferable. Over seven years, Revolut's designers accumulated a large body of documentation: interface principles, components, usage rules. People find such archives hard to navigate, and AI agents do even worse, because the documents were written for humans, not machines. With Codex's help, the team converted this archive into a format the model can read, and finished in one day.
In everyday terms, it is like clearing out a cluttered storeroom and putting a clear label on every box. Afterward, both a new employee and an AI agent can quickly find what they need. For teams that want to connect agents to their own products, this is a practical lesson: before expecting good results from a model, give it a well-organized knowledge base. The description does not say which specific format Revolut chose, and that is a limitation: the approach cannot yet be reproduced as a ready-made recipe.
Anti-fraud: triaging suspicious payments faster
The third area is fighting financial crime. The team uses OpenAI to investigate payments that the system has flagged as suspicious and to respond to them faster. The stated goal is to make online purchases safer for customers. Here AI does not make the decision in place of a human; it speeds up the review by gathering context on a transaction and preparing material for an analyst. The material does not disclose specific metrics, such as the false positive rate or the time to process a single case.
What to take away
First, the most real gains from coding agents right now come from internal tools. Companies need them, but they rarely make it to the top of a large team's priority list, while one engineer with Codex can bring them to working condition. Second, documentation is becoming infrastructure: if a machine cannot read it, agents work blind. Third, in regulated industries such as finance, it makes more sense to use AI for preparing material for people than for making final decisions.
For founders and product managers, a useful question is this: what projects do you have that are "too small for a team, but too big for an evening"? That is where first experiments should go. OpenAI published the full story with more detail on its website, but to assess the economics of these solutions, you would need figures that the public description does not yet provide.
- Nov 21, 2026Altman Calls Handing Decisions to AI a "Terrible Idea" as OpenAI Pulls 25% of Engineers Onto Security
- Nov 21, 2026Anthropic shares opened at $111.50, valuing the seven co-founders' stakes at $180 billion
- Nov 21, 2026Soros Fund Sold 19.3% of Its Nvidia Stake Before Huang Stepped Down as CEO
- Nov 21, 2026OpenAI Lets Six Apollo Auditors Inside for Four Months
Key Questions Answered
What are the key details on A code review service built by one person?
In the first story, a Revolut engineer used Codex to build an internal code review service on their own. Normally, a project like this would be a full team task: designing integration with repositories, review rules, a developer-facing interface, and ongoing support. Here one person took all of that on, while Codex worked as a partner that writes and edits code. Why this matters for practitioners: code review is a typical bottleneck in any growing engineering organization.
What are the key details on Seven years of design documentation made AI-ready in one day?
The second story gets less attention but is perhaps the most transferable. Over seven years, Revolut's designers accumulated a large body of documentation: interface principles, components, usage rules. People find such archives hard to navigate, and AI agents do even worse, because the documents were written for humans, not machines. With Codex's help, the team converted this archive into a format the model can read, and finished in one day.
What are the key details on Anti-fraud: triaging suspicious payments faster?
The third area is fighting financial crime. The team uses OpenAI to investigate payments that the system has flagged as suspicious and to respond to them faster. The stated goal is to make online purchases safer for customers. Here AI does not make the decision in place of a human; it speeds up the review by gathering context on a transaction and preparing material for an analyst. The material does not disclose specific metrics, such as the false positive rate or the time to process a single case.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.
