Case study · Game development
An indie game studio was losing whole sprints to repetitive manual work. Over one weekend we rebuilt their core workflows around AI, and the team returned on Monday to an entirely different way of working.
The studio is kept anonymous — we don't name clients without their consent.
Designers were weighed down by tedious manual labour like asset referencing, work the team reported took at least two weeks each cycle. Developers, meanwhile, lost their days to hundreds of lines of boilerplate before they could even reach their original idea.
In two days we integrated frontier AI models from Anthropic directly into their systems and workflows with custom-built tooling. AI now handles the manual labour and writes the code, all easily directed by staff. We can do the same in any industry.
Tedious manual labour now runs end to end by AI in around ten minutes. Designers spend their time designing, and developers focus on software architecture. No valuable time wasted.
Small studios live and die by what they ship in a cycle, which makes it particularly painful when a meaningful fraction of every cycle goes on work that requires almost no creative judgement. That was the position this team was in.
On the design side, the bottleneck was asset referencing and tagging. Every new batch of work meant hours of locating, cross-referencing, naming and filing, and the designers reported it consuming at least two weeks of each cycle. It was unavoidable, it was necessary, and it was nobody's idea of design work.
On the engineering side, the shape of the waste was different but the cause was the same. Developers were writing hundreds of lines of boilerplate before they could get anywhere near the problem they had actually sat down to solve. The interesting part of the job kept receding behind the scaffolding required to reach it.
We began the way we always do, by sitting with the people doing the work and watching how it really gets done. Process documentation describes the intended path; the actual path contains the workarounds, and the workarounds are where the time goes.
That audit turned up a clear division. Some tasks were pure mechanical execution of rules the team could articulate out loud, and those were the automation candidates. Others involved genuine aesthetic or architectural judgement, and those stayed firmly with the humans. The distinction mattered: automating the second category would have produced confident, plausible, subtly wrong output that somebody would then have to find and undo.
Over a single weekend we integrated frontier AI models from Anthropic into the studio's existing systems using custom-built tooling. Not a separate product to log into, not a platform migration, but automation living inside the tools the team already had open.
The referencing and tagging pipeline was rebuilt so the AI performs it end to end, with the designers directing it rather than executing it. On the engineering side, the boilerplate is now generated rather than typed, leaving the developers to specify what they want and to review what came back.
Both workflows kept a human at the point where being wrong would be expensive. Direction and review sit with the team; execution sits with the AI.
Work that had reliably consumed at least two weeks of a cycle now runs end to end in around ten minutes. The designers' time went back into design. The developers' time went into software architecture, which is what they were hired for and what a small studio most needs from them.
“Referencing and tagging would take me two weeks, now all that gets done whilst I focus on actual design.”UI Designer at the studio
Game development is an unusual business, but the waste in it was entirely ordinary. Locating and cross-referencing things. Naming and filing them. Producing the scaffolding that stands between an idea and the ability to start on it. Those tasks exist in a law firm, a letting agency, a design studio and a manufacturer, and they have the same profile everywhere: repetitive, rule-shaped, time-hungry, low in judgement.
The method transfers along with them. Audit how the work is really done, separate mechanical execution from genuine judgement, automate the first, protect the second, and put a person at every handover that matters.
If your team spends a meaningful part of every week on work that nobody would miss doing, it is worth thirty minutes to find out. Our AI consulting engagement starts with the same audit described here, and workflow automation is what happens once the audit tells us where to aim. If the answer is that your processes should be left alone, we will tell you that too.
Book a free 30-minute consultation. We'll show you where AI fits in your business, with no commitment required.