Merlin is not a game development engine feature, and he is not an auto game builder. He is a studio operations and production support assistant that helps our teams move faster and stay aligned. Trained on our internal knowledge bases and procedures, Merlin helps staff find the right answers quickly, follow the correct workflows, and reduce repeated questions across departments. He also comes with role specific prompt templates so anyone can use him, from production and QA to art and engineering, without needing to be an AI specialist. p
What does Merlin, the AI Assistant, do day-to-day?
Day to day, Merlin acts like a shared brain for the studio. He helps people find the right information fast, turns scattered documentation into usable answers, and keeps teams consistent when projects span multiple disciplines and time zones. Instead of digging through folders, chat threads, and old tickets, staff can ask Merlin and get a clear next step, including the right process to follow. The result is fewer bottlenecks, fewer repeat questions, and more time spent actually building the game.
Regardless of your familiarity with LLMs and machine learning tools, our development and tech team ensured he was built with these masks and pre-prepared prompts to assist any member of staff in the pursuit of their work. From marketing and HR to developers and security engineers, Merlin is truly a magical individual with a knowledge base to rival any other mage (or LLM!)
How was Merlin created?
Going into this process, our team knew that LLMs are new tech. There’s plenty around already, being used around the world by millions if not billions of people. Given how saturated the market is for LLMs that can handle a variety of different tasks, we saw no need to reinvent the wheel. So, what was out there that we could leverage? What could we use to make Merlin truly magical?
As a base, we used the open-source framework of LangChain. This interacts with LLMs and allows us to build tools on top of their existing features and capabilities. It has a strong capacity for testing to see how effective features are and for benchmarking those potential features. With anything new like this, and with anything as massive as an LLM, you can never know what to expect! This helped us a lot with debugging and improving Merlin overall.
We wrote a lot of code ourselves on top of this framework and integrated more than a few tools. For image generation we have DALL-E and Midjourney. Flux, a new open-source image generator is also being incorporated! For text, we have GPT 3.5 all the way to 4.0 turbo and Gemini including all of its versions up to the latest of Gemini 1.5. Finally, given our security engineers and plethora of coding experts on staff, Claude and Claude 100k was also built into our system.
Merlin is an ongoing build, not a plug and play tool. From the first version to today, we’ve continued expanding what he can do by refining his actions and connecting him with additional LLMs when it makes sense. Just as important, we’ve been grounding him in how Magic Media actually runs, so he understands internal workflows and can connect to systems like Confluence to reduce friction and streamline day to day project coordination.
Merlin is an ongoing build, not a plug and play tool. From the first version to today, we’ve continued expanding what he can do by refining his actions and connecting him with additional LLMs when it makes sense. Just as important, we’ve been grounding him in how Magic Media actually runs, so he understands internal workflows and can connect to systems like Confluence to reduce friction and streamline day to day project coordination.
Building Merlin for an entire company
Building a full LLM from scratch is rarely the smart move, so we focused on making Merlin useful, reliable, and easy for everyone to adopt. Not every team member lives in AI tools, so Merlin had to work for people with any level of familiarity, whether they’ve used Chat GPT, Gemini, or nothing at all. The goal was simple, remove the learning curve, keep the experience consistent, and make sure the assistant supports the studio instead of adding another tool people have to manage.
That’s where the “roles” or “masks” come in. The roles or masks it can take are something that our tech team are quite proud of. Depending on what you need, there’s a role to help you out. Sometimes it can be tricky to generate a prompt to get you the answers you need. In many cases, you can spend a significant amount of time generating and curating the exact prompt required to get it to answer as you need it to. The roles we’ve created cut that down, giving you a long list of pre-prepared prompts that easily get you started. Everything from marketer, coder, copywriter, and more, they allow our team to generate results faster and more efficiently!
Lastly, given the free version on which many models use their prompts as further training data, Merlin allows our staff to keep their information and prompts private from all eyes!
If you’re looking for the bigger picture before diving into studio workflows, our AI game development guide breaks down the most important tools, use cases, and practical considerations for teams planning production today.
If you are exploring AI game development to streamline your pipeline or to push new gameplay ideas, 2026 is a strong time to do it. Just be clear on the goal, choose the right place in the pipeline, and measure results like iteration speed, bug escape rate, and content throughput.
Bring AI into your game without disrupting production.
Magic Media helps studios scope, implement, and integrate practical AI features within existing pipelines, with clear delivery and full creative control.
Explore AI Development Support with Magic Media
Contact Magic Media to discuss your goals and the right support model.
