Pac-Man, but You're the Ghost: A Small Browser Game That Turns a Classic Power-Up Into the Core Loop

Garrit's Notes has published a small playable experiment called Pac-Man, but you're the ghost. There is no funding announcement, no investor list, and no claim that this changes gaming. The interesting part is narrower: the post takes one of Pac-Man's oldest emotional beats and makes it the whole game.

The premise is simple. You control a ghost. Pac-Man is handled by AI. Your job is to catch him before he clears the maze. If he eats a power pellet, the tables flip for a few seconds, he hunts you, and then you run.

That role reversal matters because Pac-Man is often remembered as a maze-clearing game, but its tension comes from pursuit. The ghosts are not just obstacles. They pressure the player, cut off routes, and make hesitation expensive. The power pellet works because it briefly reverses fear into confidence, then back again. Putting the player in the ghost's role exposes that loop without needing a full remake of the original arcade game.

Creator and problem

Garrit's Notes appears to be a personal technical blog rather than a studio or startup. The surrounding post list includes Home Assistant, n8n, VSCode, and context-window skepticism, which suggests a builder writing in public. That context matters. This is closer to a devlog experiment than a company launch.

The problem being solved is not that Pac-Man lacks clones. It has many. The more specific problem is that most clones copy the surface: maze, dots, four ghosts, power pellets. This version asks what happens when the player controls the side that usually has scripted behavior. It turns the ghost from a background pressure system into the main character.

The market positioning is also clear. It sits in the same neighborhood as browser games, game-jam experiments, and devlog demos. Its advantage is clarity. A player can understand the joke in one sentence, then immediately test whether they can actually survive it.

How the game works

The core loop has three states. First, the ghost hunts Pac-Man. Second, Pac-Man eats a power pellet and becomes dangerous. Third, the temporary fear window ends and the chase resumes. That is enough structure for a small game because the player is always making short, readable decisions.

For developers, the implementation can be surprisingly compact. The maze can be represented as a grid of blocked and open tiles. Pellets and power pellets can be tile flags. Each actor can have a position, direction, speed, and state. A browser version does not need a heavy engine. The Canvas API is enough for drawing the board, and requestAnimationFrame can drive the update loop.

Pac-Man's AI can be built as a scoring function rather than a full planning system. At each intersection, the game can score legal directions based on nearby pellets, distance from the ghost, and whether a power pellet is active. If Pac-Man is normal, the AI should prefer pellet-rich paths. If Pac-Man is powered, the AI should prefer distance from the ghost and may even move toward the player. This can create behavior that feels intentional without pretending to be a research project.

If the maze grows, A* pathfinding becomes more attractive. Red Blob Games' guide is a practical reference because it explains nodes, costs, and heuristics with visuals. For a Pac-Man-sized board, though, a greedy scorer may be enough. The danger is not technical complexity. It is making Pac-Man too perfect. A perfect opponent can be less fun than a readable one.

The state machine is the game

The power pellet is really a state change. In the original Pac-Man setup, pellets change who has agency. In this version, the ghost player should have clear states: hunter, vulnerable, recovering, and maybe returning after being caught. The State pattern is a useful mental model here because each state changes movement rules, collision rules, scoring, and visual feedback.

Timing is the hard part. If the vulnerable window is too short, the player never feels the panic. If it is too long, the normal chase loses tension. If Pac-Man is too fast, the ghost cannot catch him. If Pac-Man is too slow, the power-up reversal feels fake. Small games are useful because they expose these balance problems quickly. There is nowhere for the design to hide.

Movement also needs care. Tile-by-tile movement is easier to reason about. Sub-tile movement feels smoother, but it creates more collision edge cases. A simple version can move actors from tile to tile, while a more polished version can interpolate between tiles and still resolve decisions on the grid. That hybrid approach is common in arcade-style games because it keeps logic clean while improving feel.

Funding and traction

No funding amount, investor list, revenue model, or player count was disclosed. The supplied text also does not include a GitHub repository, source code link, license, or original playable URL. The post says the game can be played from the original article, but the destination link was not included in the text provided.

That means traction is qualitative rather than measurable. The visible signal is publication on a personal blog and a playable concept that can be explained in one sentence. For a solo developer, that is useful public iteration. For investors, there is not much to underwrite yet. The project would need source code, player metrics, retention data, or a clearer commercial path before it becomes a venture story.

The legal side also matters. Pac-Man is a protected brand and style. A commercial version would need original art, names, maze design, and rules, or a license. The safer path is to keep the role-reversal concept and build a new world around it.

What to watch next

The most interesting next step would be a public repository. If Garrit publishes the code, the best places to inspect are the AI scoring function, the power-pellet timer, collision handling, and the maze representation. Those are the systems that decide whether the game feels clever or merely cute.

A stronger version could add multiple ghosts, different ghost personalities, speedrun mode, local multiplayer, or procedural mazes. The opportunity is not a venture-scale market by itself. It is a compact proof that old mechanics can be inverted into fresh play. For a solo developer, that can become a portfolio piece, a game-jam entry, or a small browser title. For a larger team, the same idea would need progression, variants, analytics, retention hooks, and IP clearance.

The project is modest, and that is its strength. It gives developers a clean example of how a single state change can reframe an entire game. If the next version adds source code, clearer AI documentation, or a few more ghost behaviors, it becomes more than a joke. It becomes a compact case study in pursuit, panic, and balance.