Artificial intelligence can now produce a playable video game within hours. That achievement is impressive, but it does not prove the gaming industry is doomed. In my view, it proves something else: writing code was never the hardest part of making a great game.
The experiment behind The Librarian 2 shows why this matters. AI generated thousands of lines of code, created music and graphics, tested its work, and fixed several bugs. Yet the finished game remained far from a product people would pay to play.
A Fast Prototype Is Not a Finished Game
The creator began with a simple concept. Players control a librarian, collect fallen books, stop disruptive children, and prevent a chaos meter from reaching 100 percent.
AI turned that idea into a 3D game. It added random levels, permanent upgrades, bosses, earthquakes, tornadoes, and other events. The first build took about 90 minutes and contained 10,400 lines of code across 33 modules.
That speed deserves attention. The system even tested the game and adjusted problems it detected. It changed boss behavior and corrected situations that could trap players.
Still, basic faults appeared immediately:
- The movement controls were reversed.
- Some books could not be collected.
- The camera passed through walls and ceilings.
- A boss moved too quickly to catch.
- Several locations looked nearly identical.
- Disasters created inconsistent damage and pressure.
These issues were fixable, but every fix required human judgment. Someone had to notice the problem, decide how the game should feel, and tell the model what to change.
Fun Depends on Thousands of Decisions
The strongest lesson came during a play session with producer Dave. The game looked much better than its earlier version, but its balance fell apart. Chaos rose too quickly, a bully became almost impossible to catch, and a tornado barely moved the books.
“To make this a really enjoyable game with lots of great features, great balancing, clearly defined progression and a replayable game loop, it’ll take a long time.”
I think that observation cuts through much of the hype. A game is not merely code attached to images. It is a long series of choices about pace, challenge, rewards, movement, sound, story, and player expectations.
The creator also cited a statement from Take-Two leadership: coding was never the main bottleneck. The harder work involves choosing what belongs in a game and what should be removed.
AI can propose those choices, but it cannot reliably judge their combined effect. A feature may work alone yet damage the wider experience. An upgrade may feel rewarding early and ruin later difficulty. A joke may be amusing once but irritating after repeated play.
Personal Games Will Not Kill Professional Games
Some AI advocates claim players will soon generate their own games instead of buying them. I find that argument unconvincing. People can cook meals, record songs, or shoot films, yet they still pay skilled creators for better experiences.
Making entertainment and enjoying entertainment are different activities. Most players do not want to spend hours defining rules, testing difficulty, fixing cameras, and revising menus. They want to enter a well-made experience and be surprised.
“I have zero desire to stop buying games and just vibe code them instead.”
The creator admitted that even his own games received little play after their demonstration videos. The Librarian 2 was fun and polished for a rapid prototype, but he considered it “nowhere near sellable.” He had no plans to release it on Steam.
AI Should Accelerate Craft, Not Dismiss It
AI will help developers work faster. Small teams may test ideas sooner, produce early builds cheaply, and spend less time on routine programming. That is useful progress.
But speed does not replace taste, direction, testing, or restraint. Great developers remain necessary because they understand players and shape countless details into a satisfying whole.
We should judge AI games by the same standard as any other game: Are they fun after the novelty fades? Developers should use these tools, while studios should invest the saved time in design and quality. Players, meanwhile, should reject claims that a flashy demo equals a finished product. Code can build the game, but human judgment makes it worth playing.
Frequently Asked Questions
Q: Can AI create a complete video game?
AI can create playable prototypes with graphics, sound, rules, and progression. Commercial quality still requires extensive direction, testing, and revision.
Q: How quickly was The Librarian 2 produced?
Its first major build took about 90 minutes. Later rounds of testing and correction added much more time.
Q: What was the largest weakness?
Balance was the main problem. Difficulty, boss speed, disasters, upgrades, and level identity needed repeated human review.
Q: Will gamers stop buying professional titles?
That seems unlikely. Most players prefer a polished experience created and tested by skilled teams.
Q: How can game studios use AI responsibly?
Studios can use it for prototypes, routine code, early testing, and experiments while keeping people responsible for creative and quality decisions.
