adminAI for coding is already part of many developers’ workflows, but its usefulness depends on the task, the codebase, and the person reviewing its output. In game development, programmers describe it as a helpful assistant for exploring unfamiliar code, generating prototypes, and automating small tasks—not a reliable replacement for engineering experience.
In a 2025 Stack Overflow survey, 47.7% of developers said they used AI tools daily, while 36.8% viewed the technology favourably. Those figures show adoption, but they do not capture every benefit or frustration developers encounter in practice.
Image credit: Markus Spiske
Where AI for coding can help
Kristinn Þór Sigurbergsson, director of gameplay engineering at CCP, says the studio uses AI tools extensively for code-related work. Their value varies by task, but the tools can help developers find their way around a large, established project. That is particularly relevant to EVE Online, a 23-year-old MMO with a substantial codebase.
Sigurbergsson says tools such as Cursor and Claude Code can summarise code, trace logic across files, and help new team members understand how different parts of a project connect. These capabilities can make investigation faster, though developers still need to check whether an explanation matches the actual implementation.
AI can also assist with debugging, but a suggested fix should be treated as a proposal rather than a diagnosis. Sigurbergsson notes that an AI response may recommend suppressing a log message, even when the underlying issue needs a different solution. Experienced review remains important.
CCP uses AI tools extensively for code-related work. Image credit: Fenris Creations
For CCP, one practical effect is a shift in how developers spend their time: less on initial implementation and more on planning and reviewing. Faster iteration can also make it easier to explore an idea before committing to a full build. The studio uses AI-generated code to prototype EVE Online features and behaviours, but Sigurbergsson says these experiments are rarely intended to be production-ready.
Small, temporary scripts may be an especially useful application. Developers often need quick tools to generate data, investigate a problem, or automate a one-off task. When the result matters more than polished, reusable code, AI may reduce work that would otherwise take much longer to do manually.
Other developers value AI as a learning and reference aid. Positech’s Cliff Harris says he uses Anthropic’s Claude to explore algorithms, consider optimisation ideas, and look for bugs. Facepunch Studios founder Garry Newman describes using ChatGPT or similar tools to explain concepts and assist with repetitive refactoring. These examples point to a useful approach: ask for help understanding or exploring a task, then assess the answer yourself.
Garry Newman founded Facepunch Studios, the developer of Garry’s Mod. Image credit: Valve
Paul Kilduff-Taylor of Mode 7 Games also sees AI as a supporting tool. It can help developers consult documentation, discuss possible optimisations, or get a starting point for debugging. For routine questions, that may be more convenient than searching through multiple references, provided the response is checked against reliable documentation and the project’s requirements.
Where AI-generated code falls short
A common concern is that code can be difficult to maintain when the developer does not understand how it works. Kilduff-Taylor points out that current tools may not have enough context to account for an entire project. This can help explain why impressive demonstrations often involve small prototypes or lightweight frameworks rather than complex, established games.
Chet Faliszek of Stray Bombay makes a related point: producing a working result is not the same as understanding the system behind it. Learning how a language or engine behaves can help developers spot design opportunities and make informed changes later. For that reason, AI-generated code is less useful when it replaces learning instead of supporting it.
Stray Bombay released The Anacrusis in 2023. Image credit: Stray Bombay
Bram Ridder, technical director at Kythera AI and formerly of Rebellion, has used AI for basic boilerplate code but warns against relying on it as a substitute for understanding. Adam Grimley, senior programmer at Huey Games, says he uses AI to brainstorm problems outside his usual knowledge, then checks the answers against research papers, human-written tutorials, or other references.
Accuracy is another limitation. Generative AI can present incorrect information confidently, so code that looks plausible may still contain errors or misunderstand the task. Alex Darby, formerly principal programmer at Bithell Games and Roll7, says the time spent reading, testing, and correcting generated code can outweigh the time saved typing it.
Failbetter Games senior programmer Hannah Rose raises a similar concern about large suggested code blocks. Even when a suggestion saves keystrokes, it can take time to review, adapt, or remove it. If typing is not the main bottleneck, generating more code does not necessarily make development faster.
Matthew Davis of Subset Games describes AI as unreliable for substantial coding tasks beyond autocomplete. He warns that debugging a long generated solution can take longer than writing it, while code that a team does not understand can add to technical debt. Failbetter backend web developer Jem Frisby also criticises generated code for potentially fragile structure, weak architecture, and poor alignment with an existing project.
Image credit: Subset Games
How to use AI coding tools responsibly
John Ogden, CTO at Huey Games, says AI may be useful for functional tasks but is not a complete replacement for programmers, particularly when architectural decisions are involved. He also notes that specialist areas such as console development may be harder for AI tools to support because relevant training information can be limited.
For AI-generated code to work reliably across a team, developers need a process that catches mistakes. Alex Darby argues that automated tests and continuous delivery can help by making software components easier to test and review. In game development, however, tools also need enough context about the project’s architecture and constraints to make useful suggestions.
- Use AI for bounded tasks: Try it for summaries, small scripts, prototypes, or explanations before assigning it a large feature.
- Review and test every change: Check generated code against project conventions, documentation, and automated tests.
- Keep the design in human hands: Make sure the team understands the code and can maintain it after the initial task is complete.
- Compare time saved with review effort: If verification and correction take longer than writing the code, a different approach may be more efficient.
Conclusion: Is AI for coding useful?
AI for coding can be useful when it supports a developer’s work: finding information in a large codebase, exploring a design, creating a quick prototype, or automating a limited task. Its weaknesses become more significant when a project requires deep context, dependable architecture, or code that a team must maintain over time. The practical value comes from combining AI assistance with human judgement, careful review, and testing—not from accepting generated code without scrutiny.
