How Dark and Darker's Devs Used Aura to Build a Major Feature in Half the Time
Dark and Darker developers IRONMACE explain how Aura helped a non-technical designer work directly with the game's codebase, prototype tutorial systems, communicate with engineers, and help complete in just a couple of weeks what the studio estimates would normally have taken five to six weeks.
There is a wide variety of use cases for AI in game development that don't involve generative artwork, and some studios are already experimenting with new technologies beneath the surface. For Dark and Darker developer IRONMACE, that recently meant seeing whether the AI coding assistant, Aura, could help a designer with limited programming experience contribute directly to the implementation of a major new feature.
The experiment centered on something Dark and Darker had conspicuously lacked: a proper tutorial. IRONMACE had historically leaned on the extraction dungeon crawler's "hardcore" identity to justify its absence, but the real obstacle was development resources. Building one meant pulling developers away from immediate problems to invest in onboarding new players. Working with Aura, the team instead developed a workflow in which its designers could build tutorial content, understand unfamiliar systems, prototype functionality, and even implement some code themselves while experienced developers remained part of the production process.
According to IRONMACE, the results dramatically accelerated development. We interviewed IRONMACE President Terence Park to learn how he used the technology.
Dark and Darker has a very strong identity as a hardcore extraction dungeon crawler, so any new feature has to fit into an existing balance of tension, risk, and readability. Can you start by explaining what feature you used Aura to help ship, and what problem you were trying to solve?
Terence Park, IRONMACE CEO and Co-Founder: Yes, Dark and Darker has an identity as a hardcore extraction game, but it shouldn't be hardcore in the "how do I do this very basic action" sense. Up until recently, we used that “Hardcore” label as a crutch to put off working on a tutorial. Given our limited dev resources, the problem has always been: Do we put out the current fires or invest time and effort into the future?
Aura made it possible for us to do both, allowing our development team to finally build the critical tutorial system, which I expect to directly increase our player base.
Aura and IRONMACE set up a collaboration where our designers worked directly with the Aura team. The goal? See how quickly and efficiently a core feature could be built when developers fully use Aura
The interesting angle here is that your background isn't really very technical at all, is that right? Before using Aura, what would the normal path have looked like for turning this idea into something playable inside the game?
Terence Park: Yes, that is true. My background is in game design and as a producer. I can do some scripting and am familiar with data structures, but programming is not my strong suit.
The normal path would have been to assemble a full task team to cover each specific role. This usually leads to lots of meetings and managing an unwieldy and complex pipeline.
Since the process for this task was experimental, we took a flexible approach. We got the main skeleton of the idea out in an initial meetup. After that, I would make broad decisions and then rely on using Aura to fill in the detailed context gaps. For instance, once we decided to go with a checkpoint-style tutorial, I worked with Aura on building each granular tutorial checkpoint.
I mean it worked really well considering we had an engineer that was completely new to the codebase come in and, using Aura, start contributing directly to our project in a couple days and complete the main task in two weeks.
How did you translate a design idea into instructions Aura could understand? Did the process feel more like writing a design brief, talking through logic, prototyping with an engineer, or something else?
Terence Park: I approached it like trying to solve a complex problem with a really smart partner. I’d start by asking Aura simple, predictable questions on design tasks that I was very confident in to get us calibrated on the direction and goal. For example, I had it fill in multiple multilayered data assets that I was very familiar with to see that it understood my intent. Once that direction was established, then it was pretty easy to have Aura follow instructions and give feedback for things I may have overlooked.
Can you walk us through the workflow step by step, from the first prompt or design note to the first working prototype, internal testing, iteration, and final implementation?
Terence Park: After our initial meeting with the Aura team, I created a feature outline for the tutorial, and an example storyboard that was shared with the team. We then created a collaboration chat and instead of relying on traditional documentation, whenever a new feature for the tutorial was developed by the engineer, he would simply create an example asset and push that to our repository. Since members of this collaborative team were in different time zones, I’d come in later that day, use Aura to figure out the new stuff and simply get to building content. I’d leave feedback for new features I’d want created in our group chat, and usually by the next day, I’d have another new feature to play around with.
We actually had a couple of complicated tutorial triggers that we weren’t able to complete in time, but by then I had gotten comfortable enough with Aura that I was able to have it directly code the extra functions for me. So now when you get to the stage of the tutorial where you learn to grab the bandage from the dead skeleton, you’ll know that was built by Aura.
Where did Aura prove most useful: generating code, explaining technical concepts, helping structure the feature, debugging, documenting the logic, or helping you communicate more clearly with the engineering team?
Terence Park: It was super helpful in all of those tasks. For the coding area, I was fortunate that one of our internal technical leads had already set up an AI flow. Aura was able to quickly integrate that process, knowing when to reference those conventions and rules, to keep it within boundaries.
Aura also did an excellent job of explaining the changes it would make to the code in terms that I could understand. It really felt like having a custom real-time wiki for our project. That clarity in laying out the logic just made everything easier.
What parts of the feature still required experienced developers to review, rewrite, optimize, or safely integrate into the production build?
Terence Park: I still needed to do the verification and testing within the game editor, although I think that was a feature the Aura team was looking to directly address. Aura did a commendable job in creating functional UI widgets and hooking up logic, but we still had our artist do the final pass.
Game developers are understandably cautious about AI, especially when the conversation turns to art generation or replacement. How do you think this kind of AI-assisted workflow differs from the more controversial uses of generative AI in game development?
Terence Park: I think one of the big differences is the perception of emotion that goes into the work. Since code and many of the structural components are “under the hood” and feel more logical, I think we look at them as problems that have solutions for the AI to solve.
Whereas for more interpretive areas such as art and music we give much more weight to the process. I think this might be one of the reasons we are more accepting of the problem-solving AI assistance that Aura provides.
Did using Aura change how you communicate with technical team members? For example, did it help you ask better questions, understand constraints earlier, or bring a clearer prototype to engineers?
Terence Park: Yes, although a lot of my technical back and forth was done directly with Aura. It probably saved a lot of headaches for our engineers by cutting down a ton of trivial questions that would have been directed at them instead.
It also helped me bring a more complete concept when I needed to present things to the engineers. Instead of explaining abstract concepts with a document, it was simply easier to build a mockup using Aura. The engineers were then able to better understand the intent and transfer that into a production-ready feature.
Were there any moments where Aura gave you something that looked right but was not production-ready, unsafe, inefficient, or incompatible with how Dark and Darker is built? How did the team catch and correct those issues?
Terence Park: I might not be the right person to ask since I’m not a programmer by trade and I am suspicious of AI to begin with. Due to those reasons, I tried to give the tasks in bite-sized amounts, so they were pretty easy to verify and contain. Maybe that’s a sign that I should have been more aggressive, but for my personal use case, things worked out well.
After going through this process, where do you think AI assistants can realistically help non-technical founders, designers, or creative leads contribute to feature development without bypassing the expertise of programmers?
Terence Park: I’m a believer now. It helped me develop an entire tutorial from scratch in a few weeks with a lot of the base system created by an engineer who previously had little familiarity with our codebase. What would have taken us 5-6 weeks to complete was completed in half the time because of Aura.
I do want to say that am extremely fortunate to have super compassionate and competent tech leads, so I had an unbelievable backup plan if things went haywire. That let me to really explore using Aura and it was an awesome sense of freedom to be able to mess around and learn more deeply about our project without having to be tech wiz.
