AI Solves Software, Game Development

AI is making significant strides in writing code, a development Matthew Berman emphasizes will be highly disruptive across industries.

This capability allows for the creation of game mashups, such as integrating Super Mario into Elden Ring or Spider-Man into Batman, by rapidly reverse-engineering machine code into human-readable source code, a task that previously took years now takes weeks.

Compilers strip away comments and simplify mathematical operations, rendering machine code notoriously difficult for humans to reverse-engineer. Decompiling, as one expert stated, means taking that machine code and working backwards, essentially knowing the intended output.

Compilers optimize human-written source code into efficient machine code by removing elements like comments and simplifying operations, making the compiled output difficult to reverse-engineer and often requiring years of manual effort to understand.

AI is uniquely suited for decompilation due to its ability to perform continuous, tireless guessing-and-checking loops, verifying generated code against the original output.

This allows multiple AI agents to work simultaneously on different parts of a game, drastically accelerating the production process until an exact functional clone of the game is achieved, even if the resulting code is not line-for-line identical to the original.

This dramatic reduction in time demonstrates AI's efficiency, prompting the author to use AI to recreate 'Super Mario World' for the Mod Retro device.

OpenAI and Anthropic have introduced efficient, cost-effective models like GPT-6.1 Soul, which drastically lower token consumption for complex tasks. It is now so easy to recreate these types of games, according to a recent assessment.

These advanced models have drastically reduced token consumption for intricate AI tasks, making the replication of software and game assets increasingly accessible and less resource-intensive.

Analysts expect these techniques to increase in sophistication.

These methods include Asset Swapping, where character visuals are replaced while mechanics remain; Pass-through Mods, which synchronize two games with an interface; and Behavioral Integration, involving importing specific game mechanics into different engines.

The most advanced method is Full Reconstruction, where AI studies and rebuilds game logic from scratch without needing the original source code.

Photocraft serves as a prime example of an open-source, AI-powered recreation of Adobe Photoshop that focuses on output behavior rather than source code. This trend suggests that software is solved, a development that compels companies to shift their value proposition toward enterprise support and maintenance.

This recreation focuses on observing and replicating software output and behavior, rather than direct source code, indicating that software itself may become commoditized.

Consequently, software companies can no longer rely solely on their product as a competitive advantage, needing to shift value towards training, enterprise support, and ongoing maintenance.

Development tools like Claude Code, Codebase, and Cursor utilize slash-goal or slash-loop functions, enabling AI to run iterative development cycles without constant human intervention.

Data visualization reveals a shift from human-led mathematical discovery to AI-driven processes, a change that leads one user to conclude, "So we can see the days of human solving math alone are gone."

OpenAI's recent publications of significant research and proofs in mathematics, alongside data categorized by Twitter user Will Depu, illustrate a clear shift towards AI-driven mathematical discovery over the past three years.

Matthew Berman asserts that AI can effectively solve any problem where the correctness of the answer is verifiable, even if the method to arrive at that answer is initially unknown.

This capability extends AI's reach to diverse fields, including scientific discovery, material innovation, and medical advancements such as finding cures for diseases like cancer.

Each successful self-improvement amplifies the AI's ability to make further discoveries, creating an accelerating cycle.

This phenomenon is a key reason why frontier labs are discussing the need to pace and potentially slow down AI progress, as evidenced by Google's Alpha Evolve project, which optimized its own infrastructure, resulting in billions in savings.

The value of human creativity lies in non-verifiable areas such as artistic meaning and storytelling, providing emotional resonance that mass-generated content often lacks.

While AI can generate a vast volume of content, Matthew Berman emphasizes that quantity does not equate to emotional resonance or meaning, leaving human perspective as essential for curation.

The unique human ability to verify emotional impact means authentic human taste will become increasingly valuable as AI lowers the barrier to content production.

While AI can integrate existing characters like Mario into games such as Elden Ring, it cannot originate entirely new, compelling intellectual properties.

The creation of future iconic franchises like Mario or Elden Ring still fundamentally requires human creativity to devise engaging stories and enjoyable gameplay experiences, as humans must discern what concepts are truly fun amidst AI-generated variations.

Concerns about copying and derivative works are not new, as the entire history of art and music, particularly with remixing and sampling, demonstrates that most creations are derivatives of existing forms.

AI accelerates this rate of derivation, making it increasingly challenging to find genuine 'signal' amidst the 'noise' of mass-produced content, thereby increasing the importance of original creative input.