Summary of Key Points
AI has indeed made the execution processes in game development—such as art and coding—faster and cheaper. However, the root cause of game failures is never “slow execution,” but rather confused decision-making: constant changes in direction, extensive rework, and low efficiency in making decisions. Whether it’s a large company with hundreds of employees or a small independent team, they all face the problem of having a “broken steering wheel” (unclear decision-making) while “stepping on the accelerator” (using AI to speed up execution). Large companies are hindered by their organizational structures and politics, while independent developers struggle with their own judgment and business barriers. Tools can help with executing tasks quickly, but they cannot determine what needs to be done in the first place.
1. No matter how fast AI is, it can’t save a game with a confused direction—rework is the real budget killer
The real cost drain in the gaming industry isn’t the slow creation of art assets; it’s the extensive rework caused by repeated decision changes. Take Ubisoft’s *Sea of Thieves* as an example: The game changed three creative directors over eight years, and its core direction was altered four times. The team grew from 100 to 400 people, costing nearly $200 million. No one complained about the slow art production; the real issue was that it was never clear what the game’s purpose was. Every time the direction changed, the animation, levels, and effects had to be redone, resulting in all previous efforts being wasted.
Senior developer Sandberg said, “400 people working on constantly changing requirement documents is like a collective waste of resources with high salaries. Does AI make them work faster? It just means they end up throwing more away.” In 2023, the Swedish company Embracer Group canceled 29 unannounced projects, each one involving a pile of overturned decisions and wasted effort. These cases show that fast execution is useless; having the right direction is what matters.
2. AI prototypes are a double-edged sword: they look like finished products, so people are reluctant to discard them
AI has an advantage in the prototype stage—it can create six prototypes instead of just two, allowing teams to test different ideas. The problem is that no one actually discards these prototypes and starts over.
For example, AI-generated prototypes may have beautiful textures and working code, making it seem like the project is almost complete. But no one dares to tell the publisher, “We want to throw this away and start from scratch.” As a result, a temporary prototype created in six weeks becomes the foundation for a four-year project, and all subsequent systems have to accommodate the rough decisions made to speed things up (such as messy code logic). Even worse, AI-generated code often lacks a clear “creator”; three years later, no one can explain why it was designed that way, turning it into an unmanageable “technical debt.”
Sandberg compares this to human-made prototypes, which have documentation as a reference, while AI-generated ones lack any clear source of information.
3. The impact of AI on large companies and independent developers
AI affects different teams in different ways, but the “steering wheel” issue is common for both:
- Large companies: The steering wheel is in the hands of higher management, with long decision-making chains and multiple stakeholders. For instance, Sandberg experienced a single-player game being forced to add a multiplayer mode just because it looked better in competitor reports; a multiplayer game’s core design had to be compromised to adapt to different platforms. A single decision from management can cause teams to waste two quarters of work, and using AI to speed up execution only exacerbates this.
- Independent developers: Although they have more control over the direction, the freedom can lead to difficulties in making choices. AI allows a single person to develop a game (as seen with a Turkish developer who created a game in two months), but “creating something” doesn’t necessarily mean it will be successful. AI-generated games on Steam receive 53% fewer reviews and are more negatively rated because many developers use it to quickly produce low-quality, uninspired content. Independent developers also make the mistake of being reluctant to discard prototypes, lacking someone who would suggest starting over.
4. Three indicators reveal the truth: the problem lies with decision-making, not the tools
Sandberg suggests three key indicators to highlight issues in the decision-making process:
1. Rework rate: How much of the completed work is discarded or requires redoing? (e.g., art assets, level design) This number shows whether AI is saving money or causing more waste.
2. Decision delay: How many days pass from identifying a problem to making a final decision? ( Meetings don’t count; only actions taken by the team matter.) The longer the delay, the more rework is necessary.
3. Decision durability: How long does a decision remain valid before being changed? If creative directions change frequently, even the best tools won’t help.
These indicators are easy to collect, but why aren’t they used? Because they measure the ability of management. In large companies, there are metrics for art production and engineering quality, but there’s no evaluation of the decision-making process. Sandberg says, “The steering wheel is still broken; we’re just using a more powerful engine.”
Conclusion: AI is the engine, but decision-making is the key
AI reduces the cost of physical development, but the cost of thinking clearly about what to do remains high, and this cannot be outsourced. Regardless of team size, before stepping on the accelerator, you need to check the steering wheel: large companies need to address confusion in decision-making, while independent developers must improve their judgment and ability to make choices. No matter how powerful the tools are, if the direction is wrong, the project will still head towards failure.