The conversation around generative AI in game development has been inescapable for the past two years, but until recently it produced far more heat than light. Advocates painted visions of radical productivity breakthroughs; skeptics warned of ethical disaster and creative collapse. Neither side, however, had much in the way of hard data from genuine experience. That has begun to change. As studios have integrated AI tools into real workflows, a clearer picture has emerged — one that is far more measured than the breathless hype suggested. Productivity gains are real but inconsistent, highly task-specific, and heavily dependent on skilled human oversight. And now, another factor is forcing a recalibration: the cost of AI itself is climbing sharply, as subsidised introductory rates give way to token-based billing models that make heavy use cases — like generating complex game assets or working with massive codebases — significantly more expensive.
The Promise vs. The Reality: What AI Actually Delivers in Game Development
Nearly every company in a tech-adjacent sector has at least experimented with AI tools. The resulting body of experience is now large enough to support meaningful conclusions. The consensus, drawn from actual usage rather than vendor presentations, is that AI can boost productivity — but only in narrow, well-defined domains, and only when supervised by experienced professionals. Code completion tools in IDEs, for example, eliminate a great deal of repetitive “donkey-work” for programmers. Automated code reviews catch certain classes of bugs with reasonable reliability. Some artists find that deep-learning-based image editors speed up tedious parts of their workflow. AI transcription and summarisation tools handle managerial slog like meeting notes. These are solid, incremental improvements — the kind that free up skilled staff to focus on more complex and interesting tasks.
Yet these gains are a long way from the transformative efficiency executives were sold. Developers who have been pushed to use more advanced AI tools — particularly agentic systems tasked with autonomously working on game codebases — report that they hit hard limits very quickly. Game codebases are large, complex, and specialised. Any code produced by an AI agent must be extremely carefully vetted by senior developers, a dull and time-consuming process that often negates much of the time saved. In art workflows, generative AI struggles profoundly with maintaining visual consistency across assets. This is a critical shortcoming for game development, where a coherent visual style is essential. And the unresolved legal question of copyright — most interpretations still lean toward “no” for AI-generated content — creates enormous risk for any studio that might want to use those assets commercially.
Productivity gains from using AI tools are real — but they’re inconsistent, highly task specific, and extremely reliant on human supervision
blockquoteblockquoteblockquote
Rising Costs Force a Business Case Reckoning
Perhaps the most significant change in the AI conversation over the past few months has been the shift in pricing. For much of 2023 and early 2024, corporate clients enjoyed heavily subsidised introductory rates, funded by tens of billions of dollars in private investment. Those subsidies are now ending. The industry is moving toward token-based billing, and the real costs of AI are being passed through to customers. Microsoft’s Copilot services were among the first major dominoes to fall, but the trend is nearly universal. Executives who, a year ago, wanted AI integrated into every facet of their business are now belatedly scrutinising the long-term costs.
Those costs hit hardest for use cases that burn large numbers of tokens — exactly the use cases that are most relevant to game development. Generating complex 2D and 3D assets, working with enormous codebases, and running “thinking” models (which essentially chain multiple LLMs together, consuming an order of magnitude more tokens per response) all become rapidly more expensive as realistic pricing takes hold. When the bills arrive, AI needs a business case based on more than hype and fear of missing out. High costs must be matched by measurable productivity gains and cost savings on the other side of the equation.
This shift has taken much of the puff out of the irrational exuberance that characterised boardroom AI discussions. Developers who voiced concerns about AI’s actual utility often struggled to get a fair hearing from senior management, whose views had been shaped by evangelising salespeople and ready-made dismissive counterarguments. Today, those same boardrooms are far more receptive, because the AI bills — both metaphorical and literal — have started to come due.
The Enthusiasm Gap: A Familiar Warning Sign
What emerges most clearly from talking to people with actual experience of these tools is a sharp enthusiasm gap between developers and executive-level decision makers. Historically, that gap has been a red flag for any new technology. Successful, meaningful technology adoption almost always follows a bottom-up pattern. Artists and engineers in the games industry are quintessential early adopters; they enthusiastically lobby leadership to invest in tools that genuinely improve their work. When the opposite pattern emerges — executives pushing a buzzword-heavy new technology at reticent development staff — it signals trouble. The last time the industry saw that exact pattern was with NFTs, an ill-fated technology whose six months of undeserved limelight now feel like a fever dream.
Generative AI is like having a very fast but extremely unreliable junior staff member on your team
blockquoteblockquoteblockquote
The experience of developers who have actually worked with generative AI tools bears this out. The tools are fast but unreliable. Senior artists and programmers report having to devote large amounts of time to supervising and checking output. One developer likened it to having a very fast but extremely unreliable junior staff member — whose salary demands keep soaring from month to month. Rather than eliminating work, AI in its current form often redistributes it, shifting the burden onto the most experienced (and most expensive) team members.
What the Near Future Holds for AI in Games
AI optimists will argue that the tools will improve over time. The core technology is still young, and ongoing research will undoubtedly address some current limitations. But the escalating cost trajectory and the approaching horizon for token price subsidies may make that improvement somewhat moot. Almost every attempt to meaningfully improve these tools relies on adding more token processing. “Thinking” models, for example, use multiple LLMs working in tandem, each step consuming vastly more tokens than the final output suggests. Even if the underlying cost of processing tokens and training new models falls — and the billions in capital expenditure being announced by AI companies suggests those costs are far from falling — those savings are immediately consumed by new layers of abstraction designed to mitigate the non-deterministic, error-prone nature of AI systems. More AIs checking each other’s work, bigger context windows, more computationally expensive architectures — each “improvement” adds cost.
Meanwhile, there is a growing, if sometimes grudging, recognition that a significant portion of consumers actively dislike AI-generated art and music. In a creative industry where audience perception matters enormously, that cultural resistance is a nail firmly hammered into the coffin of many possible use cases. If AI had turned out to be a true revolution in productivity and development cost, some executives might have been willing to take on that cultural battle. With little evidence of any such revolution emerging, nobody wants to fight it.
The games industry has always embraced new technological trends, and it is unlikely that early ethical concerns will prevent genuinely useful AI tools from seeing widespread adoption. But the radical transformation that some expected is not materialising. The bills are coming due, the productivity gains are incremental, and the enthusiasm gap between developers and executives remains a glaring warning. As the dust settles, it looks increasingly likely that AI will find a lasting place in game development — but in a vastly more limited, far less impactful role than the last two years of hype suggested.
