When AI Becomes the New Adventure Game: A Reflection on PromptQuest

Recently, Microsoft decided to open source the iconic text adventure game Zork, prompting me to consider revisiting it during the holiday season. However, I soon realized that much of 2025 has been spent navigating similar virtual obstacles—only these are AI chatbots masquerading as companions, often leading to frustration rather than fun.

The Nostalgia of Adventure Games

Back in the 1980s, adventure games like Zork invited players into a world of words—imaginary caves, castles, and mythical creatures. The core mechanic was simple: you typed commands like "Hit Goblin," expecting the game to understand your intent. But games often responded unexpectedly. For example, "Hit Goblin" might cause the game to say, "You punch the Goblin," resulting in the Goblin dodging and stabbing you instead—game over.

Success depended on learning the game’s quirky language, such as using "Stab Goblin" or "Hit Goblin with Sword." Frustration was part of the experience, but it was tolerable because of the game's nostalgic charm and the limited AI capabilities of that time.

Moving from Pixels to Prompts

Fast forward to today, and AI chatbots promise to make our digital interactions seamless. But my recent encounters with Microsoft's Copilot AI have felt more like navigating a labyrinth of "PromptQuest," reminiscent of those old text adventures—without the fun.

For instance, I asked Copilot to parse online data and generate a downloadable spreadsheet. It replied with a Python script claiming it would do so. Yet, when testing it later, I noticed inconsistent results—sometimes the same prompt yields different outputs across days. Even my regular tests for typos varied in format each time.

Microsoft’s decision to roll out different versions of Copilot across Office and desktop applications compounds the problem—each version interprets prompts differently, forcing me to endlessly relearn what works. It's like playing a game where the rules keep changing, and every success feels temporary.

The Frustrating Reality of AI Assistance

This ongoing experiment has become a frustrating game—what I call "PromptQuest." Copilot often promises to complete tasks like creating or downloading spreadsheets but fails to deliver. Despite numerous assurances of progress, I end up with nothing but an empty download or an unchanging progress bar (which, amusingly, resembles a text adventure progress indicator).

Ironically, this progress bar looks strikingly similar to a text adventure game’s output—a visual cue of my ongoing battle with the AI’s unpredictability.

The Moral of the Quest

My point isn’t that chatbots are infallible or that mistakes aren't part of the process. Instead, working with such AI tech often resembles groping through a dark cave—you're told you're making progress, but the outcomes aren't reliable. The sense of advancing is illusory, and the promised productivity gains often feel like participation in a game of "Hit/Kill/Stab Goblin"—a repetitive, frustrating cycle that wastes time and patience.

Conclusion

As AI continues to integrate into our daily routines, I can’t help but feel that we’re stuck in an endless, unwinnable game. The challenge is less about mastering the system and more about managing the inevitable failures—an ongoing PromptQuest where the only real victory is recognizing when to step back and reconsider the game itself.