Capcom's REX Overhaul Puts AI in the Development Pipeline

Capcom wants the next phase of its RE Engine to help AI write programs and run test plays. Programmer Satoshi Ishida outlined that goal at Capcom Open Conference RE: 2026 on October 2 as part of REX, an incremental overhaul of the company's in-house engine. His description, first reported by GameBiz and detailed by Video Games Chronicle, was future-facing: a more common programming language would make the engine easier for both developers and AI to understand, with automatic program creation and bug-checking test plays among the intended uses.

Ishida called the destination an “AI-generation game engine.” That phrase invites an obvious question from players: will AI make the art and other assets they see in Capcom games? The proposed coding and testing functions concern a different stage of production. They also raise a question for workers that a screenshot cannot answer: when a tool produces code or runs a test, who decides whether its output is good enough, and who does the follow-up work?

An Engine Overhaul with Several Jobs to Do

RE Engine entered development in 2014 and debuted with Resident Evil 7 in 2017. It has since powered nearly 30 Capcom titles. Ishida identified several pressures behind the overhaul: larger games and data volumes slow builds and iteration, while more than 2,000 developers now use the engine, including international staff and people coming from other engines. The REX systems he described are meant to address that scale as well as his ambition to integrate AI into the pipeline.

An engine used across that many projects has to work for people who did not build its first tools. If a change takes longer to build and check, the delay recurs across an iteration cycle; if formats differ across tools, someone has to make them cooperate. Those are practical reasons to modernize a shared engine before asking AI to operate inside it. They also explain why the REX announcement is about more than the word “AI” in Ishida's description.

REX's named components show how much of the work concerns ordinary, difficult infrastructure. RE:Dox is intended to standardize data processing across formats. RE:UI separates a development tool's appearance from its state to speed up toolmaking. RE:Log replaces text-based logging with an ID-based system. Ishida also described RE:Flows, which converts visual scripting into standard code, and RE:Runtime, a revised runtime core designed for batch processing of large numbers of characters. Some improvements concern data formatting and memory management, apart from the AI ambitions.

Capcom developed RE Engine during the development of Resident Evil 7; it now uses that foundation across many titles and a development organization Ishida says numbers more than 2,000 engine users. REX proposes to improve that shared foundation incrementally rather than replace it. Standardized data and more workable logs could help developers trace problems across projects even if AI-generated programs never become a routine part of production. Conversely, an AI system that can write code would be of limited use if the surrounding tools and formats remained hard for people to navigate. That is an inference from the problems Ishida listed, rather than a result Capcom has demonstrated.

What the Asset Boundary Covers

Capcom drew a boundary earlier this year. In a March 2026 shareholder-meeting response quoted by IGN, it said, “We will not implement assets generated by AI into our games.” The company also said it was exploring AI as a way to improve development efficiency and productivity in graphics, sound and programming. A tool might assist someone working on an asset without the company shipping an AI-generated asset. The wording leaves a practical question: where, in graphics or sound work, would Capcom draw the line between assistance and an asset generated by AI?

Capcom subsequently told investors that generative AI could improve routine operations so developers could spend more time on essential value creation, while saying that humans should handle creativity at the core of games. Ishida's account supplies a concrete candidate for saved time: builds and iteration are slowing as games and their data grow. Future automated testing might also take over repeatable checks. Those are plausible benefits for people waiting on tools or test results, rather than measured savings from the proposed AI functions.

Code generation and test plays need separate scrutiny. Generated code, if adopted, would enter a system that developers must understand, review and maintain. A program that appears to work in one test can still require a person to judge whether it does the intended job and fits the rest of a game's code. Automated test plays, if adopted, could repeatedly check behavior as builds change. A flagged problem still needs someone to decide what happened, reproduce it where necessary, fix it and check the fix. The amount of work could shift rather than simply disappear. Ishida's reported account describes capabilities for the future; it does not give a basis for treating AI test plays as a current replacement for human QA.

The division of responsibility is also a production decision. If automated runs produce more findings, QA and developers need a way to sort useful failures from noise. If generated code speeds up an initial task but takes longer to review or maintain, a faster first draft may leave a slower handoff. These are possible outcomes, not claims about how Capcom's teams currently work. The useful measure would be the whole cycle from suggestion or test result through human review to a stable build, rather than the number of tasks a tool can perform on its own.

Level-5 offers a different view of the same boundary problem. After the company used AI in a presentation, CEO Akihiro Hino apologized and addressed that use. He said Level-5 made its scenarios, character designs and basic world-building by hand, while aiming to use AI to convert human-made assets into digital assets such as polygon models, Anime News Network reported. Hino said the company was testing and refining an approach intended to shorten a five-year development cycle for major titles to two years. That is a target, not an achieved production result.

Level-5 discussed converting assets; Capcom's REX presentation emphasized an engine that could generate code and run tests. Their tools and workplace effects cannot be assumed to match. Hino's apology does, however, expose a limit in a promise focused on game content: a company can say its central creative elements are human-made and still use AI elsewhere, including in material players see outside the game. Capcom's stated rule against AI-generated assets answers one question about released games. It does not, by itself, tell the programmer reviewing generated code or the QA worker examining an automated test result how much time that review will receive.

REX will be easier to judge when its proposed functions move from a conference goal to identifiable production work. A tool that makes builds faster, findings clearer or code easier to maintain could give developers time back. Whether those gains reach the people doing the review depends on how Capcom assigns the work that automation leaves behind.