The office of mobile game company Bagelcode on Teheran-ro in Gangnam District, Seoul, was warmed by the energy of its employees. The company held back-to-back in-house game development contests in Mar., the GameJam and the AI First Fair, and received results that even surprised itself.
Over the seven days of the contest, 124 mobile games were completed. Compared with nine completed works at the same event three years ago, it was not a simple increase but a leap in productivity.
The preparation period for the contest was reduced from 60 days to seven days, and the number of participating teams increased from 10 to 124. The final winners are not chosen by judges but determined by user return rates after release on the app store.
Bagelcode is a mid-sized game developer in its 14th year. A game that brought casino play to mobile succeeded, pushing 2025 revenue above 100 billion won. We asked Co-CEO Kim Jun-young and employees about the secret behind the explosive productivity.
AI production engine rollout... Individual participants surged
Since ChatGPT first came out in Nov. 2023, I've barely slept. I sensed the changes artificial intelligence (AI) would bring, and I thought the opportunity would pass in a flash.
So said CEO Kim Jun-young. Bagelcode immediately began developing its own AI-based game production engine, GameBakery.ai. At the same time, it encouraged employees to use various AI tools without worrying about token consumption.
The explosive increase in completed works at this contest was thanks to the first rollout of GameBakery, developed over the past two years, and the rise in employees' AI literacy working in tandem.
Data scientist Song Ha-young, who took part in the contest, said, "I made a game where an alchemist creates new weapons by combining various materials, and I was able to build the materials combination databases at a speed incomparable to previous years."
Communications Director Kim Yoo-jung said, "Three years ago, there were no individual participants, but this time 88% of participants were individuals," adding, "It was also meaningful to confirm that an era has opened in which one member can handle tasks that used to be done by a team."
A ‘SaaS-pocalypse’ unfolding in real time
The AI First meetings chaired by CEO Kim Jun-young are held four to five times a week. At first, they were meetings only for AI-related departments, but they have expanded into open meetings that any employee can join online or offline.
At the meeting the reporter observed, the main agenda item was ToDos, a project management tool developed by the AI Lab team.
"The existing collaboration tool Jira (the most widely used collaboration tool for project management and issue tracking worldwide) requires a lot of manual input. With game productivity surging, we couldn't manage game histories the old way, so we built a tool called ToDos ourselves."
"Our team decided to modify part of ToDos' code, developed by the AI Lab team, to suit our situation. Since we are a team of artists, we also made the UI (user interface) more intuitive and colorful."
The "SaaS-pocalypse" emerging on Wall Street in the United States — the outlook that existing software-as-a-service will be replaced by AI — was not an exaggeration. On the ground, the mood was shifting toward building needed tools in-house and using them.
What if big tech builds it all?
The reporter asked, "If big tech like OpenAI or Google updates related functions, won't the painstakingly built tower collapse?" In fact, there is talk that when big tech updates AI service features, dozens of startups disappear.
CEO Kim Jun-young said, "We have actually experienced that countless times," adding, "Back in the ChatGPT 3.5 era, various tools we built ourselves, including the Model Context Protocol (MCP), have now outlived their usefulness."
Even so, Kim stressed that preliminary research should not stop. That is because the level of technical understanding differs between an organization that merely uses a function and an organization that has thought hard about why the function is needed and has tried building it.
Kim added, "It may seem like big tech is building every function, but it still hasn't reached the realm of making 'fun games,'" and "In that sense, 2026 is the last window in which game developers can widen the gap and keep the initiative."
AI is speed; people set the direction
The most important lesson Bagelcode learned while pushing AI transformation over two years is that AI is, after all, an accelerator, and people ultimately decide the direction. AI without direction races out of control like a supercar with no one at the wheel.
The reason Bagelcode has held in-house hackathons like GameJam every year since 2023 is to let every member freely experiment with new technology and apply it to work. Projects discovered at hackathons lead to further development and are reflected in actual work.
Now, Bagelcode's in-house code output has surged from about 10 items a day to 100–200. It has entered a stage where it is difficult to manage code by hand alone.
On top of this, if the AI engine GameBakery, first applied to game development, is expanded companywide to marketing and operations, the pace of code production will only steepen.
Bagelcode has decided to redesign its work environment in the process. In the first half, it plans to provide every employee with a Mac mini to build a "one-person AI agent" work environment in which each person operates one or more agents.
The Mac mini is an ultra-compact desktop that comes as a standalone unit. As the open-source AI agent framework OpenClaw has recently become popular, this device has also gained traction.
Employees have begun to envision delegating tasks such as organizing messages and emails and repetitive data processing to agents so they can focus on more important judgment and planning.
At Bagelcode, AI was no longer a slogan or a catchphrase. It was the language of practical work—built, revised, and woven into operations.
CEO Kim Jun-young said, "A game is a comprehensive art in which the entire process is connected, from planning, art, and development to data, operations, and customer response," adding, "Talent with deep domain knowledge (expertise) in each field augmented by AI and creating synergy with one another—that is the moat we will build."