Capcom has detailed REX, a multi-stage project for extending its proprietary RE Engine with faster production systems and a foundation intended to be easier for artificial intelligence to read and operate. Programmer Satoshi Ishida presented the work at Capcom Open Conference RE:2026 on October 2. The plan builds on the existing engine rather than replacing it, adding individually named components for data, interfaces, logs, visual scripting, and runtime execution.
The company identifies three pressures behind the project: larger games have made ordinary iteration slower, more than 2,000 developers with varied backgrounds use the engine, and AI must be integrated usefully into production workflows. Its long-term description of an “AI-generation game engine” is broad, but the concrete systems announced so far primarily target engineering work: processing project data, building responsive tools, collecting diagnostics, creating gameplay logic, running automated tests, and eventually helping AI understand code and perform playtests.
REX begins with ordinary production bottlenecks
RE:Dox standardizes how different forms of project data are represented and processed. Capcom says this common layer improves loading and editing performance. RE:UI is a lighter framework for development-tool interfaces; it separates visual presentation from state, reduces unnecessary screen updates, and can be introduced gradually alongside older interface components. Because it is not tied to a visible display, Capcom also says tools built with it are easier to test automatically.
RE:Log replaces repeated log text with smaller identifiers, reconstructing readable messages when they are displayed. Logs can be compressed, collected centrally, searched, translated, and connected to the same underlying communication rules. These changes matter independently of AI. A faster editor and structured diagnostics shorten the human feedback loop, while consistent records give automated systems cleaner material to analyze. Capcom’s claims come from its conference presentation; the sources do not provide independent performance tests across shipped projects.
Visual scripting and runtime changes create a shared language
RE:Flows is Capcom’s new visual scripting system. Designers can connect graphical blocks, while the system converts the result into a commonly used programming language behind the interface. The stated goal is to preserve an accessible authoring view without the difficult merging and debugging associated with opaque graph data. RE:Runtime similarly reworks how many game objects are loaded and managed, grouping them into blocks rather than treating every instance separately. A translation layer converts developer-facing code into Capcom’s RE:C++ execution format.
Together, those systems are intended to make more of the engine follow consistent, machine-readable rules. Capcom’s future aim is for AI to understand and generate programs, conduct playtests, and check for bugs. That is a direction, not a demonstrated autonomous workflow. No public benchmark in the inspected reports measures generated-code accuracy, test coverage, defect detection, or the review time required from developers. GamesIndustry.biz also notes Capcom’s earlier statement that it would not put generative-AI assets into its games, distinguishing production assistance from shipping generated art or audio.
The useful indie lesson is architectural, not a tool download
RE Engine is an internal Capcom platform, so most small studios cannot adopt REX as they would a commercial engine plugin. Capcom says some technology, including parts associated with RE:Dox and RE:Log, is being open-sourced in part to support AI training, but the conference coverage does not provide a complete package list, licenses, setup guide, or compatibility matrix. Developers should verify any released repository and its terms before treating it as production-ready software.
The transferable idea is to prepare a project for reliable automation before adding an agent. An indie team can begin by making asset formats consistent, exposing build and test commands, producing structured logs, separating interface state from presentation, and defining deterministic test scenes. A bounded evaluation might ask an assistant to diagnose one known failure from logs or run one scripted gameplay check, then compare its result with a human baseline. Record false positives, missed defects, setup time, review time, and whether another developer can reproduce the result.
REX is therefore more specific than a promise to make games with AI, but less complete than a publicly validated AI production system. Capcom has confirmed the engineering components and its intended path toward code assistance and automated playtesting. Whether that path improves finished-game quality will depend on implementation, measurable test coverage, and human review—not the ambition attached to the engine alone.