Capcom AI plans moved from investor Q&A answers to engine architecture on 2 October 2026. At the company’s own technical conference in Tokyo, Capcom engineer Satoshi Ishida said the publisher intends to turn the engine behind Resident Evil, Monster Hunter and Street Fighter into what he called a game engine for the “AI generation”. The line every outlet quoted was the goal at the end: “a future where we create games together with AI.”
The Verge, IGN, Eurogamer, PC Gamer and Polygon all picked the Capcom AI story up within a day, mostly from IGN’s translation of a Japanese report by GameBiz. Read against that original report, and against the first REX component Capcom has published as open source, the plan is more specific than the headlines suggest. It is also less about generative art than many readers feared.
This article covers what Ishida announced, why the Capcom AI push starts with the engine rather than the games, the six technologies inside the REX project, what the open-source RE:Dox code shows, where the coverage got the details wrong, how the plan fits Capcom’s stated AI policy, and what other studios and software teams can take from it.
Table of contents
- The Capcom AI Announcement at RE:2026
- Why the Capcom AI Push Starts With RE Engine
- Inside REX: Six Technologies Behind the Capcom AI Plan
- RE:Dox, the First Open-Source Piece of the Capcom AI Plan
- Capcom AI Coverage Checked Against the Source
- Capcom AI Policy So Far: No Generated Assets, Lots of Automation
- “Create Games Together With AI”: Why the Capcom AI Slogan Worries Players
- What the Capcom AI Strategy Means for Studios and Software Teams
- Capcom AI FAQ
- References and Further Reading
The Capcom AI Announcement at RE:2026
The Capcom AI talk opened CAPCOM Open Conference RE:2026, the company’s first in-house tech conference in three years. Its official site lists Tokyo sessions on 2 and 3 October and Osaka sessions on 17 and 18 October, with this year’s releases among its case studies: Resident Evil Requiem, Pragmata, Onimusha: Way of the Sword and Monster Hunter Stories 3.
The Capcom AI talk and its speaker
IGN translates the session title as “The Outlook and Future of the REX Project, Further Evolving the RE Engine for the Next Generation”. Most English coverage describes Ishida simply as a programmer. The conference site is more precise: it advertises the REX session as an explanation by the RE Engine’s lead architect. That matters for how seriously to take the Capcom AI roadmap, because this was not a side talk by an enthusiast. It was the engine owner setting the direction.
“A future where we create games together with AI”
GameBiz reports that Ishida linked the whole REX effort to one long-term aim: evolving RE Engine into a game engine for the AI generation. Capcom has unified the system under the rules of what he called the world’s most common programming language, so that code is easier for people to use and easier for AI to read. Once that base is unified, he said, AI should eventually be able to understand programs, write them, run test plays and check builds for bugs on its own.
Why the Capcom AI talk landed now
Capcom is having one of its strongest years. Eurogamer reported in June that Resident Evil Requiem became the fastest-selling entry in the series and passed 7 million copies within two months. A publisher on that run does not need to rebuild its engine, which is exactly why the Capcom AI framing surprised people. As PC Gamer’s Rick Lane put it, “If it ain’t broke, don’t fix it.”
Why the Capcom AI Push Starts With RE Engine
RE Engine is the base of almost everything Capcom ships. According to GameBiz, development began in 2014, the engine was first used commercially in Resident Evil 7 in 2017, and every one of the 27 or more Capcom titles released since has been built on it.
Nine years and 27 games
More than 2,000 developers, almost every developer at Capcom, now use RE Engine daily. Ishida said that after more than eight years of continuous use the engine faces problems that adding features can no longer fix. Any Capcom AI tooling that touches production has to go through this engine, so the engine has to be ready for it first.
Three problems, one of them new
Ishida listed three problems. First, games have grown so large that a small tweak can mean a long wait before you see its effect. Second, the people using the engine are more varied, including overseas staff and developers who learned other engines elsewhere. Third, “successfully integrating AI technology into development workflows has become an essential condition for streamlining future development.”
The third item is the new one. Capcom first described REX, short for “RE neXt”, at CAPCOM Open Conference Professional RE:2023 in October 2023. Automaton’s report of that event lists the same first two problems: asset counts and code size had grown more than five times since Resident Evil 7, and outsourcing partners abroad and hires from commercial engines needed multilingual documentation and familiar usability. AI does not appear among the reasons in that report. In three years, the Capcom AI goal has gone from absent to one of the three pillars.
| Reason for REX | RE:2023 (October 2023) | RE:2026 (October 2026) |
|---|---|---|
| Scale | Assets and code more than 5x larger than Resident Evil 7 | Data volume makes simple edit-and-check loops slow |
| People | Overseas outsourcers, hires from commercial engines, multilingual docs | 2,000+ users, overseas staff, multilingual support |
| AI | Not listed | Integrating AI into workflows is “an essential condition” |
| Approach | Add new technology to the existing engine, no rewrite | Same: add “RE:” technologies step by step |
Inside REX: Six Technologies Behind the Capcom AI Plan
REX is not a new engine. GameBiz describes it as an umbrella name for a set of systems, each called “RE:” something, that Capcom adds to the existing RE Engine one at a time. Ishida walked through six of them. None of the six is an AI system in itself. They are plumbing, and the Capcom AI ambition depends on that plumbing.
RE:Dox, one set of rules for every data format
Game data covers character movement, screen layout, saved settings and much more. Until now each type of data had its own processing rules, which made the code complicated and slow. RE:Dox translates data of any format into one common set of rules and processes it quickly. Capcom says loading and editing speeds both improved sharply.
RE:UI, tools that stop freezing
As Capcom’s tools grew, their interfaces became heavy. They froze when screens held too many elements, they paused for garbage collection, they needed a lot of code for small changes, and they ran only on Windows. RE:UI separates how a control looks from its state, so 100 checkboxes share one appearance and track only whether each is ticked.
It also replaces event-driven updates with a one-way, view-driven model. The screen checks the data for itself, redraws only if something changed and checks less often while nothing does. Because RE:UI does not depend on a particular display layer, Capcom says automated tests can now run without a person watching the screen. That last detail is quietly central to the Capcom AI testing goal.
RE:Log, logs written as ID numbers
RE:Log replaces text log lines with small ID codes, looking up the matching message only when someone displays it. That shrinks the data, lets messages be edited without touching code, and allows automatic translation for overseas developers. Logs from individual PCs are compressed and gathered on a central server, where anyone on the team can search them. Communication and logging now share one protocol.
RE:Flows, visual scripting that produces real code
Building game logic used to require serious programming, and earlier node-based visual tools caused data collisions when several people edited the same file. RE:Flows keeps the drag-and-connect interface but converts the result into a standard, widely used programming language behind the scenes. Designers can prototype mechanics themselves, and teams can merge and debug the output like any other code.
RE:Runtime and RE:C++
RE:Runtime rebuilds the core that runs a game. Modern games keep huge numbers of characters and objects alive at once, and managing each one separately strained memory. RE:Runtime groups instances into fixed-size blocks and processes them in batches. It also introduces RE:C++, a Capcom-specific language that code written in a human-friendly language is automatically translated into, so the machine runs it fast.
| Component | What it does | Problem it targets | AI inside it? |
|---|---|---|---|
| RE:Dox | One token model for every data format | Slow, inconsistent data handling | No |
| RE:UI | Lightweight, view-driven tool interfaces | Freezes, GC pauses, Windows-only tools | No |
| RE:Log | ID-based logs, central store, shared protocol | Noisy logs, lost bug reports | No |
| RE:Flows | Visual scripting compiled to standard code | Designers blocked on programmers | No |
| RE:Runtime | Batched, fixed-size memory blocks | Many characters on screen | No |
| RE:C++ | Capcom language for fast execution | Mixed languages in one codebase | No |
RE:Dox, the First Open-Source Piece of the Capcom AI Plan
GameBiz reports that, as part of the Capcom AI groundwork, the company has published some REX technologies, including RE:Dox and RE:Log, as open source, partly so they can help with AI learning. Automaton’s Japanese site confirmed on 2 October that Capcom had announced the RE:Dox release on GitHub to coincide with the talk.
What is in the repository
The code sits under a new GitHub organisation, CAPCOM-TD-OSS, which describes itself as CAPCOM CO.,LTD.’s Technical Research Division and was created on 5 June 2026. The REDox repository was created on 30 September and is written in C# for .NET under the Apache-2.0 licence, which allows commercial use and modification. When we read it on 3 October it had 961 stars and 26 forks.
The README calls RE:Dox “a high-performance, token-based structured data engine for .NET” and “a core component of REX”. It parses JSON, JSON5, CBOR, MessagePack, TOML, XML, HTML, CSV and INI into a compact DOM of fixed 64-bit tokens, then reads, edits and converts from that one structure. The core package, CAPCOM.REDox 1.0.0, went up on NuGet on 1 October, and the README lists eleven more format and compatibility packages, six of them still previews.
There is a small joke for fans in the sample code. The 30-second example serialises a player called “Leon” at level 42 carrying a handgun and a green herb, renames him “Claire”, and adds a knife and a first aid spray.
What the benchmarks claim
Capcom’s benchmarks compare RE:Dox with Microsoft’s System.Text.Json on .NET 10 on an AMD Threadripper PRO 5975WX. The README warns that the ratios “are not universal performance guarantees”, so treat them as Capcom’s own measurements on three standard test files.
RE:Dox speed-up over System.Text.Json, Capcom’s own benchmarks
Memory use is the more striking result. Deserialising the numeric-heavy canada.json file, the README reports RE:Dox allocating 2,619.92 KB against 8,734.66 KB for System.Text.Json, under a third as much. For a game engine that has to move huge volumes of data without stalling, allocation counts at least as much as raw speed.
Memory allocated deserialising canada.json (KB, lower is better)
Why open code helps the Capcom AI goal
Publishing part of a proprietary engine is an unusual step. GameBiz says one aim is to help AI learn. The logic is straightforward: coding assistants handle libraries they have seen before far better than private code they have never seen. Public code, its README and its tests can all end up as training data for future models, so assistants learn the conventions Capcom wants its own engineers to follow. Open code is a cheap way to make the Capcom AI workflow better without sharing a single game asset.
What is not public yet
GameBiz names RE:Log as open source too, but on 3 October the CAPCOM-TD-OSS organisation listed exactly one public repository, REDox. RE:Log may follow, or may be published somewhere we have not found, but for now RE:Dox is the only REX component anyone outside Capcom can download and test.
Capcom AI Coverage Checked Against the Source
Most English reports on the Capcom AI talk were written from IGN’s translation, a day after Ishida spoke. Comparing them with the GameBiz original, the conference site and GitHub turns up a handful of errors, one of them significant.
“Six programs that integrate AI” is wrong
PC Gamer wrote that REX “involves six different programs that will integrate AI and machine-learning technology into RE Engine”. It also described RE:UI as “built to provide AI-assisted optimisation” of Capcom’s tools, “sort-of like adding DLSS support for Blender”. GameBiz describes nothing of the kind. RE:UI’s gains come from sharing one appearance across controls, one-way data flow and redrawing only on change. None of the six components, as reported, contains machine learning.
That distinction matters. The Capcom AI claim is about readiness: a consistent, well-structured codebase that future AI coding and testing systems can work with. It is not a claim that the engine already runs on AI.
Smaller slips in the Capcom AI coverage
DualShockers wrote that AI could eventually “generate parts of a game”. GameBiz’s wording is narrower: AI understanding programs and creating them automatically, plus automated test play. IGN said Capcom is open-sourcing components “for the purpose of training AI”, while GameBiz presents that as one aim among others.
| Claim in coverage | What the sources show | Verdict |
|---|---|---|
| Six programs integrate AI (PC Gamer) | GameBiz describes data, UI, logging, scripting and runtime rework with no AI inside | Wrong |
| RE:UI provides AI-assisted optimisation (PC Gamer) | Speed comes from a view-driven design and shared appearances | Wrong |
| AI could generate parts of a game (DualShockers) | GameBiz: AI could understand and write programs and run test play | Overstated |
| Open-sourced “for the purpose of training AI” (IGN) | GameBiz: helping AI learn is one of the aims | Overstated |
| RE:Dox and RE:Log are open source (GameBiz) | Only REDox was public on GitHub on 3 October | Partly true |
| Ishida is “a programmer” (most outlets) | Conference site bills the REX talk by the RE Engine lead architect | Incomplete |
What nobody has named
GameBiz quotes Ishida on unifying REX under “the world’s most common programming language” without naming it. The only REX component in public is written in C#, while RE:C++ is Capcom’s own execution language. Until Capcom publishes slides or a recording in English, any claim about which language Ishida meant is a guess, and this article does not make one.
Capcom AI Policy So Far: No Generated Assets, Lots of Automation
The REX talk fits a Capcom AI line the company has held, in writing, for most of 2026. It keeps generative AI out of what players see and puts it to work everywhere else.
The March 2026 Capcom AI line
On 23 March, Capcom published the Q&A from its shareholders’ meeting. “We will not implement assets generated by AI into our games,” it said. “However, we plan to proactively use it as a contributing technology to improve the efficiency and productivity of the game development process,” across graphics, sound and programming. In a June investor Q&A it added that “the creativity at the core of such experiences should be handled by humans.”
Capcom AI playtesting: 30,000 hours a month
The clearest evidence of Capcom AI in production comes from Google Cloud. An April 2026 Google Cloud release says Capcom built playtesting agents on Gemini Enterprise Agent Platform. They include visual inspection agents that use computer vision to tell intended horror atmosphere from rendering faults, predictive agents that send test bots to the areas most likely to break, and knowledge agents that answer questions about how veterans solved past problems. Google says the agents run “more than 30,000 hours per month”.
To give that figure a sense of scale, assume a tester works 160 hours a month. On that assumption, 30,000 hours equals roughly 188 full-time testers, more than many entire studios employ. The chart below is our arithmetic on Google’s figure, not a Capcom headcount.
Monthly playtesting hours, Google’s figure against illustrative human teams (160 hours per tester)
In May, Shinichi Inoue, Capcom’s vice president of game development platform and AI solutions, explained the Capcom AI testing design to 4Gamer, as translated by Automaton. “A huge volume of this checking and evaluation work is carried out while humans are asleep,” he said. Another agent screens the bug reports against the game director’s concept before people see them. “Even top-tier AI still cannot match our creators when it comes to sensibility,” he added.
The idea generator that went quiet
There is one awkward entry in the Capcom AI record. In January 2025, technical director Kazuki Abe described a prototype “idea generation” system on Google Cloud that read design documents and proposed concepts for the “hundreds of thousands” of background objects a big game needs. Automaton noted in May that this sits uneasily beside the no-AI-assets promise. Capcom’s later statements suggest the line it draws is final assets, not brainstorming.
| Date | Capcom AI milestone |
|---|---|
| Oct 2023 | REX (“RE neXt”) first described at RE:2023; AI not among the reasons |
| Jan 2025 | Kazuki Abe describes an idea-generation prototype on Google Cloud |
| 23 Mar 2026 | Shareholder Q&A: no AI-generated assets in games, AI for efficiency |
| 22 Apr 2026 | Google Cloud details agent playtesting at 30,000+ hours a month |
| 19 May 2026 | Shinichi Inoue tells 4Gamer AI cannot match creators’ sensibility |
| Jun 2026 | Investor Q&A: core creativity “should be handled by humans” |
| 30 Sep-1 Oct 2026 | REDox repository created, NuGet package 1.0.0 published |
| 2 Oct 2026 | Ishida’s REX talk: “create games together with AI” |
"Create Games Together With AI": Why the Capcom AI Slogan Worries Players
The Capcom AI engineering details are reassuring. The slogan is not, and the reaction shows why.
Pragmata and the DLSS 5 demo
The Verge opened its story by noting that “Capcom’s Pragmata might be all about the horrors of AI”. Polygon’s Tomas Franzese called the comments “red flags” for anyone who does not want AI in game development and said they were “nerve-wracking”. Fans also remember March, when Nvidia used Resident Evil Requiem in its DLSS 5 reveal and the AI-redrawn faces of Grace and Leon drew heavy criticism. GamesRadar later reported that Capcom was glad players “really liked the original design of Grace”.
The open door in the Capcom AI policy
The Verge’s Terrence O’Brien put the tension well: Ishida’s comments seem to line up with Capcom’s policy, “though definitely seem to leave the door open to more extensive use.” Nothing in the REX talk generates art. But “together with AI” has no stated limit, and a codebase built for AI to read and write is also a codebase that could, one day, let AI do more. The Capcom AI line on assets is a policy, not a technical barrier, and policies can change.
For now, the record points one way. Every Capcom AI use made public so far, from testing and logs to code and idea boards, sits behind the screen. The test is whether Capcom keeps stating that line as plainly once REX is in full production.
What the Capcom AI Strategy Means for Studios and Software Teams
You do not need to ship Resident Evil to learn from REX and the wider Capcom AI playbook. The same pressures, including more code, more varied teams and AI assistants that only help when they understand your systems, apply to any large software estate.
Make the codebase legible to machines first
Ishida’s argument is that AI help depends on consistency. One language, one data model and structured logs give both new hires and AI coding assistants a fair chance. Many businesses buy AI assistants before they tidy the code those assistants have to read. The Capcom AI sequence runs the other way, and it is the more reliable order. An AI strategy that starts with data and code hygiene usually beats one that starts with a tool purchase.
Incremental beats rewrite
Capcom chose not to replace an engine that 2,000 people depend on. It is swapping subsystems one at a time and keeping games shipping throughout. For any organisation sitting on a large legacy platform, that is the safer pattern: modernise behind stable interfaces, measure each step, and let a software development team replace parts without stopping the business.
Turn testing into the first AI win
The most proven part of the Capcom AI story is testing, not creation. Agents that play builds overnight and pre-screen bug reports save expensive human time without touching what customers see. RE:UI’s design, which lets tests run without a person watching, shows the groundwork that makes this possible. Testing is a sensible first target for AI agents in any software organisation.
Draw the asset line in public
Capcom has written down where AI may and may not go, and repeats it every time the question comes up. That clarity is part of why most players read the REX news as engineering rather than betrayal. Businesses using generative tools should set the same kind of public boundary, backed by IT governance that records what AI produced and who approved it.
The wider industry is moving the same way. Sony’s QSSR upscaling for the base PS5 puts machine learning into the rendering path, and AI labs are already learning from video game data to build world models. Capcom’s contribution is to show what the groundwork looks like before any of that reaches the player.
Capcom AI FAQ
Is Capcom using generative AI to make its games?
Not for anything players see. According to the company, Capcom AI use covers testing, development tools and efficiency work. Its March 2026 shareholder Q&A says it “will not implement assets generated by AI” into its games.
What is the REX project?
REX, short for “RE neXt”, is the engine programme at the centre of the Capcom AI plan. It modernises RE Engine by adding new subsystems to it rather than writing a new engine. It was first disclosed in October 2023 and detailed at RE:2026 on 2 October 2026.
What does “AI-generation game engine” mean?
In the Capcom AI roadmap, it means an engine built for an era in which AI helps develop games. In practice that means one consistent programming base that AI can read, write code for, test-play and check for bugs. It does not mean an engine that generates games by itself.
Can I use RE:Dox?
Yes. RE:Dox is on GitHub and NuGet under the Apache-2.0 licence, which permits commercial use and modification. It is a .NET library for parsing, editing and converting structured data, not a game engine.
Which Capcom games use RE Engine?
GameBiz says more than 27 Capcom titles have been built on it since Resident Evil 7 in 2017. Recent examples include Resident Evil Requiem, Pragmata, Onimusha: Way of the Sword, Monster Hunter Wilds and Street Fighter 6.
Who is Satoshi Ishida?
He is the Capcom engineer who gave the REX talk. English outlets describe him as a programmer, and the conference site bills the session as an explanation by the RE Engine’s lead architect.
References and Further Reading
Capcom is preparing for a ‘future where we create games together with AI’ (The Verge)
How will RE ENGINE evolve? The future of the new REX project (GameBiz, Japanese)
Capcom Announces Plans to Transform the RE Engine Into an AI-Generation Game Engine (IGN)
Capcom plans to evolve RE Engine into an AI-generation game engine (Eurogamer)
Capcom announces plans to transform its RE Engine into an AI-generation game engine (PC Gamer)
Resident Evil and Onimusha developer plans to embrace AI in game development (Polygon)
A look into the future of Capcom’s game engine (DualShockers)
CAPCOM Open Conference RE:2026 official site (Japanese)
REX foundation technology RE:Dox released as open source (Automaton, Japanese)
Next-generation REX ENGINE in development, RE:2023 (Automaton, Japanese)
How the company plans to use generative AI in game development (IGN)
The publisher clarifies its stance on generative AI (Eurogamer)
Engineers say even the smartest AI can’t match our creators (Automaton West)
How the Resident Evil maker is leading the way for enterprise agentic AI (Google Cloud)
Generative AI still cannot match the devs who make Resident Evil (GamesRadar)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
