The Age of Magic studio at Playkot has taken a practical route to AI adoption. Rather than building its strategy around one headline-grabbing tool, the team has threaded AI through the repetitive and time-consuming stages of everyday development. The result is a series of measurable accelerations across a long-running live-service game.
Sergey Sergey, AI Lead at the Age of Magic studio, works on Brain – the studio’s internal AI system – and helps integrate AI into coding, documentation, game-design configurations and operational workflows. The development studio behind a similar game, Age of Magic.
“We are trying to introduce AI everywhere it can create value,” Sergey says. “My job is to help set up those processes, see how they perform and decide what the next step should be.”
Building a Brain for a living game
The Brain (own name for custom Playkot’s AI Tool) was created a few months ago. It connects Age of Magic’s analytical database in ClickHouse, internal project documentation and market intelligence with Claude. It can search the studio’s knowledge, query product data and help game designers examine competitors and player behaviour.
Brain is available directly to game designers. They can ask how a feature performed, examine different player cohorts, review an offer or advertising campaign, or retrieve the context needed to prepare a new hypothesis.
Before Brain, many of those questions went through the analytics team. With two analysts supporting several game designers, even routine requests could wait between two days and a week. Today, for standard questions, that delay has effectively disappeared. “A game designer can write a question directly into Brain and receive an answer immediately,” Sergey explains. “The delay went from days, and sometimes a week, to almost zero.”
This does not make analysts unnecessary. It changes what they spend their time doing. Instead of assembling short-term operational reports, they can focus on forecasting, predictive models, deeper investigations and data quality.
The Brain also helps with documentation. It can prepare templates, fill routine sections and turn loosely structured ideas into readable requirements. A game designer can dictate a stream of thoughts, let the system organise it, and then check whether the feature has been described accurately.
The person remains the architect of the idea. AI handles part of the scaffolding around it.
From 24–30 hours to 8
The clearest engineering result comes from creating new heroes, a recurring task in a character-driven RPG.
Previously, implementing the standard gameplay logic for a hero required around 24 to 30 hours of programming. With the current AI-assisted workflow, that initial coding stage has fallen to approximately eight hours. The figure excludes later bug fixing and polishing, but still represents a three-to-fourfold reduction in a repeatable part of production.
For Playkot’s team, focusing only on the hero pipeline misses the larger change. Code is written faster across other features. Bugs are fixed sooner. Game-design documents take less time to prepare. Product data becomes available without waiting for a queue. Together, those gains shorten the distance between an idea and a production release. Sergey estimates that overall productivity across the studio may have improved by roughly 20% to 30%, although he presents that as a directional assessment rather than a single audited metric.
“AI is not a magic button that creates one enormous jump,” he says. “You become faster at many different points, and the combined effect increases the overall pace.”
That compounding effect matters especially for an evergreen game. Age of Magic has an established audience, years of code, interconnected mechanics and a continuous need for new characters, eventsand improvements. A saved hour is not saved once; the efficiency can repeat across releases for years.
The developer becomes an architect
The studio’s view of AI is not built around replacing specialists. It is built around moving their attention upward.
One Age of Magic developer described the shift simply: “I am becoming less of a programmer and more of an architect.”
Coding models can already produce useful code when they have access to the project’s conventions and context. The developer’s value increasingly lies in deciding how a system should work, how it should interact with existing architecture, where it may fail and what trade-offs are acceptable.
The same principle applies to game design. AI can structure documentation and retrieve analytical context, but it cannot decide what Age of Magic should become. The game designer still defines the player problem, creates the feature concept and judges whether the result belongs in the game.
“We are freeing people from routine so they can take on more challenging tasks,” Sergey says. “Developers can think more about resilient systems and more interesting features. Game designers can focus on the feature itself.”
For players, the goal is not simply more output. It is giving the team room to attempt features that would previously have been too expensive, slow or demanding for the available capacity.
Trust, but review
Speed becomes dangerous when it outruns quality. The studio therefore treats AI output as production material that must pass through professional controls. AI-assisted code is reviewed by multiple developers. They check whether it follows the project’s code style, uses the correct methods and classes, includes the necessary comments and fits the surrounding architecture. The feature then goes through QA and testing.
Brain follows a similar risk-based model. Sergey says it performs reliably on many small statistical requests, with an observed error rate of up to roughly 5% in the team’s testing. But sensitive analysis is not accepted blindly. When an important product decision depends on the result, Brain provides the initial summary and the analytics team verifies it. Early versions of Brain could hallucinate or return incorrect data. Earlier coding models sometimes ignored the studio’s conventions. The response was not to abandon the tools, but to improve prompts, instructions, context files and review mechanisms.
There was no single spectacular AI project that collapsed. Instead, the studio went through a continuous loop of imperfect output, diagnosis and refinement – a more realistic picture of enterprise AI adoption than simply installing a finished product.
AI in art – carefully and selectively
Characters, icons, splash screens and 3D objects remain artist-created. AI is used more selectively for references or as a starting point for backgrounds and supporting 2D elements, which artists then refine. Human specialists retain responsibility for coherence and the final quality bar.
Why evergreen games are a special AI challenge
AI can accelerate new and mature games, but the value appears differently. A new team can use it to prototype faster and test a market hypothesis sooner. A long-running game presents the opposite challenge: not a lack of history, but an abundance of it.
Age of Magic contains years of code, design decisions, dependencies and accumulated knowledge from different teams. AI can help specialists navigate that complexity inside a system too large for any one person to hold completely in memory. For a new project, AI primarily shortens the road to validation. A team can build a prototype, put it in front of an audience and learn whether its central idea deserves further investment. That reduces the risk of spending years developing a product before discovering that the market does not want it.
For an evergreen game, the product has already proved its relevance. The challenge is to keep evolving it without losing the qualities that made players stay. AI helps the studio move through accumulated technical and product complexity more quickly and deliver new content to an existing community. The stakes are also higher. Age of Magic already has a global audience. A faster release is valuable only when it respects the product players know and the trust they have built with the studio.
This makes AI less of a shortcut and more of a navigation system. It can surface information, reveal patterns and accelerate production, but it cannot independently determine which direction is right for the game.
The next two years
The next stage is deeper integration. QA is one of the areas where the studio is only beginning to apply AI, and Sergey expects these tools to become embedded more naturally across production pipelines.
His broader prediction is that Age of Magic features will become larger, more ambitious and better informed. Faster analysis should help teams choose hypotheses that match real player behaviour. Lower production costs should make previously expensive ideas feasible. Engineers and game designers should spend more time designing systems and less time carrying information between tools.
The shift could also change the scale of what a relatively focused team can attempt. When improvements appear across analysis, documentation, coding and testing, the benefit is not limited to saving money. The same capacity can be redirected toward richer systems, more ambitious features and a faster response to what players are doing inside the game.
Sergey and his team expects AI to become less visible as a separate initiative and more deeply rooted in normal production. Instead of asking where AI is being used, teams may eventually see it as part of the infrastructure behind almost every discipline.
“Two years ago, we could not write code this quickly, understand our audience this deeply or analyse results with the same speed,” Sergey says. “Now we can make more precise conclusions and use them to build better features for our players.”
The Age of Magic studio is not betting on AI as a replacement for game development. It is betting on AI as a force multiplier for the people who already understand the game.
In an evergreen product, that distinction is everything. Technology can accelerate the journey, but people still decide where the game should go.
