Learnings From GDC 2025

For us, GDC 2025 was all about immersing ourselves into the day-to-day lives of game developers and how AI is changing them. Our Bezi team soaked up 50+ sessions, hosted an indie meetup, and chatted with hundreds of developers of every stripe: solo creators, small consultancies, and AAA titans. We wanted to dig into how AI is shaping workflows, where it’s falling short, and the key bottlenecks that keep game devs from shipping faster. After talking to so many devs, one overarching theme emerged: despite wildly varying budgets and team sizes, we’re all fighting the same fires, just with different tools.

Before jumping in, note that we’re focused on developmental AI: the kind that helps you code, debug, search, or streamline dev pipelines. This article won’t delve into generative art tools (those remain a hot topic, but a separate can of worms that we're not interested in). Instead, we’ll highlight the most pressing pain points we heard around GDC and where the industry thinks AI solutions can step up. From the hundreds of chats we had, four core themes emerged:

• Asset and documentation hurdles

• Debugging inefficiency

• Contextual AI’s untapped potential

• Onboarding and knowledge sharing

Asset Search & Documentation Chaos

“I once spent two days looking for a single sound effect buried in an unnamed folder.”

No matter how big or small a studio is, teams consistently struggle to locate assets and relevant documentation. One consultancy lead said, “Our Notion is a graveyard of outdated tutorials.” Meanwhile, at larger companies, multiple teams have built the same internal search and review tools from scratch to solve this without realizing it, causing wasted time and duplicated effort.

Studios are interested in AI that can automatically label or “tag” assets (for instance, “forest-ambiance” or “low-poly-water”), then retrieve them via natural language. These auto-tagging and vector search capabilities makes sorting through a large project with thousands of models, animations, textures, etc. significantly faster and more accurate. As it stands currently, features like Unity search just is not cutting it. People also mentioned wanting the ability to bookmark, save frequently used filters, and index across entire projects so no one ever loses valuable solutions.

In one extreme example, a developer at a mid-sized studio said they had 5-10 hours of pure “Where is that file?” time each week, multiplied across a dozen employees. That kind of search and documentation headache becomes a significant resource drain and robs artists, designers, and developers of time that could be spent doing something more productive.

Debugging: The Necessary Evil

“ChatGPT can’t see how 1,000 scripts interact, it’s like fixing a car blindfolded.”

It came as no surprise that debugging remains a massive time sink. We heard from multiple teams, new indies and AAA veterans alike, that debugging can take up 20-30% of a project’s development cycles. This isn’t something most devs are excited to spend time on. It’s a frustrating process, and, just like searching for assets, takes focus away from the creative work that helps a game standout.

• Complex Interdependencies: Bugs often arise from how different scripts or systems collide. A texture or lighting setting could glitch out physics, a single script could misalign an entire animation chain, and so on. When one of these interactions occur, it’s often difficult to identify the cause and devs will spend hours chasing the source down endless rabbitholes.

• Runtime Awareness: The ideal scenario is an AI that keeps a watchful eye on gameplay as it runs, logging every relevant event in real time. “If it could point to the exact moment a character’s animation breaks and tell me why, that’d be gold,” said a VR indie dev.

• Automated Fixes with Guardrails: Several people told us they would welcome AI-generated code fixes, as long as they come with step-by-step explanations, version history, and an easy undo. One indie dev described wanting “AI-generated fix previews,” so they can avoid breaking legacy code in the process.

We also noticed that many developers see performance optimization as part of that “debugging” bucket. This means, if an AI tool can parse an entire codebase and scene graph to diagnose a slow frame rate or a memory leak, it would have a huge impact on day-to-day workload.

Moving Past Generic AI Responses

“ChatGPT can generate code that technically works, but it rarely actually lines up with our project. I spend more time fixing ChatGPT code to fit than I would just writing the script from scratch.”

While popular language models (such as ChatGPT and Claude) can handle boilerplate code, the biggest complaint we heard at GDC is the lack of deeper project context. Many developers think AI needs to see the full picture of a game’s structure and use that to inform responses, rather than operating on as much copy-and-pasted content as possible and blind assumptions to fill in gaps.

• Project-Wide Context: A truly contextual AI would need to understand your entire codebase, scene hierarchy, and code style, then serve suggestions that work in your existing framework. “I want it to just implement suggestions in-engine,” said one developer, adding that the current lack of context in AI tools makes them feel like a partial solution.

• Automation Beyond Code: Developers also brought up the desire to automate tricky or repetitive “editor tasks”, such as hooking up animation state machines or building scalable UI mockups from Figma designs. That would allow them to focus on creative innovation instead of doing the same chores again and again.

• Workflow Reusability: Some consultancies that prototype a lot mentioned they need a quick “setup wizard” for new projects. For instance, saving “turn-based combat with a twist” as a template, instead of rebuilding it from scratch every time.

Onboarding

“I didn’t know we already had a solution for this collision bug until I asked the right person.”

The final theme relates to onboarding and collaboration. Teams that use internal tools and packages often don't have central repository for documentation. In addition, many studios rely on short-term contractors, who may need weeks to get familiar with a codebase. But when those contractors leave, all that knowledge leaves with them and the learning process restarts with the next contractor.

• Automated Onboarding Guides: Many developers said they crave project-specific tutorials or documentation that summarize the most up-to-date coding guidelines, internal tools, and existing systems. The faster a new hire understands the code, the more value they can bring before their contract is up.

• Institutional Memory Indexing: Some studios want to surface the rationale behind past decisions, such as “Why we switched to a certain physics engine in 2022” or “Which UI pipeline we used for our VR prototype.” Being able to easily navigate previous design decision would prevent multiple teams from rediscovering the same lessons at great cost.

• Best Practice Enforcement: Others dream of an AI that nudges them if they stray from established code review rules or asset organization standards, so everyone remains consistent without an avalanche of Slack reminders.

AI in game development is moving away from being a novelty and becoming a wrench that many developers use, but these GDC conversations showed just how badly that wrench still needs tightening. From digging through asset graveyards and chasing phantom bugs to wrestling with context‑blind language models and rebooting onboarding for every new hire, the friction points were surprisingly similar across teams of all sizes. Developers aren’t asking for silver bullets; they’re asking for tools that truly understand their projects, surface answers at the speed of thought, and allow them to spend their time creating instead of troubleshooting. The next leap forward won’t come from adding more features to engines or “generating” assets - it’ll come from reducing the clutter that keeps great games from being shipped.