Levelling Up: How AI is Transforming Video-Game Development

Few industries illustrate the co-evolution of artificial intelligence and commercial technology as clearly as video games. Now, generative AI is poised to alter the economics of game production itself, with consequences that reach into workforce planning, creative ownership, and the development prospects of countries still building their digital economies.

Policy discussions about artificial intelligence (AI) have largely ignored the video-game industry. This omission is untenable given that the global gaming market generated in excess of US$180 billion in annual revenue in 2025, surpassing the global film and music industries combined. It is also growing fast in regions like the Asia-Pacific, the Middle East, and Latin America that feature prominently in development agendas. Game studios were among the earliest commercial adopters of pathfinding algorithms, Goal-Oriented Action Planning (GOAP), behaviour trees, and procedural generation, often deploying AI techniques at scale years before they gained traction in other sectors. AI researchers, for their part, have long used game environments as testbeds for reinforcement learning and agent design. The two fields are not adjacent but rather have been structurally entangled for decades.

The connection between gaming and AI is architectural, built into the hardware that both industries depend on.

The connection between gaming and AI is architectural, built into the hardware that both industries depend on. NVIDIA, founded in 1993, developed the graphics processing unit (GPU) to meet the rendering demands of video games. For over a decade, the GPU market was driven almost entirely by gamers who wanted faster frame rates and richer, smoother visual environments. The parallel-processing architecture of GPUs, however, turned out to be well-suited to the operations underlying machine learning. By the early 2010s, researchers had begun repurposing gaming GPUs for neural network training, and NVIDIA pivoted to serve both markets. The dynamic between gaming and AI is not of parallel industries occasionally intersecting, but one of co-evolution. The recent introduction of generative AI techniques in game development has jump-started another evolutionary dynamic that can have implications for labour markets, creative economies, and the global distribution of technological capacity.

A Long Alliance

AI and video games have been entangled since the industry’s inception, though the nature of that entanglement has changed with each technological generation. In the 1970s and 1980s, games like Pong and Pac-Man relied on simple rule-based logic to govern opponent behaviour. These systems were rudimentary, but they helped to establish the principle that software could simulate decision-making in real time, a principle that would scale dramatically in subsequent decades. By the 1990s, developers had adopted finite state machines (FSM) to give in-game enemies and allies more believable movement and reactions. Titles like GoldenEye 007 (1997) demonstrated how augmenting NPC (Non-Player Character) intelligence through relatively simple FSMs could transform a player’s experience of a game world, imbuing it with a density and unpredictability that linear and scripted encounters could not replicate.

The 2000s brought behaviour trees and GOAPs. Games like F.E.A.R. (2005) famously used these architectures to enable layered NPC conduct. F.E.A.R. became particularly notable for its enemy AI that evoked tactical dynamism rather than mere reactive movements. Procedural content generation matured in parallel with titles like Minecraft (2011) and No Man’s Sky (2016), using algorithms to generate worlds rather than replicating or hand-crafting every asset to produce near-infinite variations from finite rule sets. The through-line across these developments was the gaming industry functioning as an overlooked but consequential laboratory for AI techniques, stress-testing them at commercial scale in conditions that academic research environments could not provide.

Generative AI as a New Toolkit

Traditional game AI could execute behaviour, but it could not generate content. Generative AI introduces a qualitatively different capability that cannot be surmised as a single phenomenon. In asset creation, tools powered by diffusion models and large language models can generate concept art, texture variations, and 3D model drafts much faster than humans. These are tasks that previously required hours or days of specialist work. For narrative design, large language models enable dynamic NPC dialogue that responds to player input contextually rather than cycling through fixed dialogue trees. The result is the possibility of emergent storytelling that adapts to individual playthroughs, a prospect that has been discussed theoretically in game design for years but has only recently started becoming technically viable.

As the requirement for human capital in game development is shrinking due to AI, questions about market access and industrial geography that go well beyond the games sector are becoming more critical.

World-building benefits from similar advances. Generative systems can produce terrain, architecture, and environmental detail procedurally with a consistency and aesthetic quality that established algorithmic approaches struggled to achieve. Rapid prototyping may be the most transformative near-term application. Small teams, or even individual developers, can now produce functional game prototypes in much shorter time frames.

The distinction that matters for policy is not between AI-assisted and non-AI development since that boundary is becoming outdated. More importantly, as the requirement for human capital in game development is shrinking due to AI, questions about market access and industrial geography that go well beyond the games sector are becoming more critical. If the barriers to entry in game development are falling, the question becomes who is positioned to walk through the door, and that is a question with geo-economic dimensions.

Labour and Industry Dynamics

The productivity gains that generative AI offers carry a corollary that is uncomfortable but unavoidable: some of the work currently performed by human specialists will be automated or restructured. Concept artists, texture designers, quality assurance testers, and junior writers occupy roles that are most immediately exposed to displacement. Generative tools can now approximate, if not yet match, the output of these positions. The gaming industry lost an estimated 14,600 jobs in 2024, surpassing the 10,500 recorded in 2023. While the causes of these layoffs cannot be singularly ascribed to AI adoption in lieu of other factors such as post-pandemic overcorrection and rising production costs, the coincidence with accelerating AI adoption is difficult to ignore.

The skills gap is real, but AI workflows are still new for most sectors, which means that latecomers are not as far behind as they might assume.

New categories of work are emerging alongside these losses. Prompt engineering, AI pipeline integration, training-data curation, and the oversight of AI-generated content all require human judgment and technical fluency that the tools themselves cannot provide.

Studios that once hired artists who could paint digitally now seek artists who can direct an AI model and refine its output to prevent generic and low-quality output, in other words, AI slop. This is a hybrid competency, part aesthetic judgement and part technical orchestration, that training and education systems will need to adjust for. The transition presents a genuine opening for emerging markets hoping to build a competitive game-development workforce. The skills gap is real, but AI workflows are still new for most sectors, which means that latecomers are not as far behind as they might assume.

Conclusion

The relationship between AI and the video-game industry is entering a phase that differs from its predecessors not in degree but in kind. For decades, AI was a component of the in-game mechanics. It is now becoming a component of development studios. This shift carries consequences for labour markets, creative industries, and the global distribution of development capacity that policymakers have been slow to reckon with, in part because gaming continues to be treated as a consumer entertainment product rather than as a crucial economic sector.

The countries and companies that act now to build enabling conditions rather than waiting for them to materialise will be best positioned to capture the value that this transformation creates.

Policy responses need to move beyond the binary of unconditional acceptance or instinctive dismissal. Players, developers, and policymakers bring legitimate but competing interests to the table, and durable outcomes will depend on frameworks that take all three seriously rather than privileging one at the expense of the others. For emerging markets in particular, the opportunity is genuine but conditional. It depends on deliberate investment in infrastructure, skills, and legal institutions, not simply on the availability of cheaper tools. The countries and companies that act now to build enabling conditions rather than waiting for them to materialise will be best positioned to capture the value that this transformation creates. Those who treat gaming as peripheral to the AI agenda will find, in time, that they have misunderstood both.

This commentary originally appeared in Observer Research Foundation.