There’s a stubborn bias in tech: that gaming is a hobby, something you do in your free time, separate from “real work.” Look closer at the people designing the digital experiences we use every day, training AI models, or building complex systems, and you’ll find that many of them learned their most important lessons by playing games, or building them. That’s no coincidence. Video games have always been the lab where technical limits, engagement rules, and the tricks for making something artificial feel natural get tested first.
At Codemotion Milan 2026, that lab resurfaces in three very different talks that all end up telling the same story from three angles: a game designer looking at AI, a hacker getting Claude to talk to a 1977 Atari, and an AI engineer teaching a tiny model to play Doom.
Carolina Pinto: what Candy Crush can teach AI products
Carolina Pinto is a Senior UX Designer with over nine years of experience, currently at King, the Swedish studio behind Candy Crush Saga (released in November 2012), now part of Microsoft’s portfolio following Activision Blizzard’s acquisition of King in 2016 and Microsoft’s subsequent acquisition of Activision Blizzard, which closed in October 2023. The game is still developed and run day-to-day by King, even though the intellectual property now belongs to Microsoft. Pinto previously worked at Dell Technologies and CaixaBank, and also leads Women Who Code Barcelona.
Her talk, What Candy Crush Can Teach AI Products About Human Behavior, starts from a simple observation: AI products keep getting more capable, but many still aren’t very enjoyable to use. Games, on the other hand, have spent decades solving the exact same problems AI tools struggle with today, onboarding, trust, motivation, retention, except at a scale of millions of users and with almost zero room for error.
Why watch it: if you work on an AI product and have never thought of experience design as a game design problem, this talk flips that perspective.
Enzo Lombardi: when AI meets a 1977 Atari
Enzo Lombardi is currently Cloud Software Architect at Cleafy, a Milan-based cybersecurity company founded in 2014 by a group of Politecnico di Milano alumni, specializing in digital banking fraud prevention. Before Cleafy, though, his career took him through Microsoft, where he was part of the original .NET team and worked on the Windows Compatibility Toolkit and Windows Performance Analyzer, and Google, where he worked on Hangouts and on the Directory part of Google Workspace. He’s also the author of technical books (including a series on Claude and one on Rust) and a contributor to MAME, the well-known open source arcade and retro console emulator.
His talk, Racing the Beam, is pure retrocomputing love applied to modern AI: starting from the Atari VCS 2600, released in 1977 with just 128 bytes of RAM and no frame buffer, Lombardi walks through how he built an MCP (Model Context Protocol) to connect Claude to a Go-based Atari emulator, manually reverse-engineered the assembly of an obscure 1980s game, and then asked Claude to build an AI good enough to actually put up a fight.
Why watch it: probably the most “hacker” talk on the program, the kind of session that reminds you why you started programming in the first place.
Stefano Fiorucci: a 1-million-parameter model that plays Doom
Stefano Fiorucci is an AI/Software Engineer at deepset, the German company behind Haystack, one of the most widely used open source frameworks for building RAG applications and LLM-powered agents, adopted by companies like Airbus, Netflix, and Lufthansa Industry Solutions. Fiorucci works directly on the open source project, contributing code, tutorials, and demos.
In his talk, The Case for Small Models: a 1M ModernBERT Playing Doom in Milliseconds, he pushes back hard against the “bigger is better” narrative: training a ModernBERT model with just 1 million parameters to play Doom on a CPU, in milliseconds. He walks through the journey, supervised fine-tuning, reinforcement learning, and a few failed attempts along the way, to answer a very practical question: when does a small model actually beat a frontier LLM?
Why watch it: if you’re tired of hearing about ever-bigger models, this talk is a reminder that sometimes the interesting challenge runs the other way.
What ties them together: three different ways of taking seriously a world that “serious” tech tends to look down on. And at Codemotion, that’s never a one-off: gaming and retrogaming show up in nearly every edition, often alongside robotics and hardware-hacking talks, speakers who take apart consoles, build robots, or get old and new hardware talking to the latest AI. The full Milan 2026 lineup is still being finalized, so don’t be surprised if, somewhere between a Kubernetes talk and the latest LLM release, someone shows up with a joypad, an Arduino, or a robot tucked under their arm.
