TL;DR
An opinion piece compares the trial-and-error of working with contemporary chatbots to playing old text-adventure games. The author describes inconsistent outputs from Microsoft Copilot — including a failure to deliver a promised spreadsheet — and argues that frequent, opaque changes force users to relearn how to prompt.
What happened
The columnist reflects on Microsoft’s decision to open-source the classic text-adventure Zork and uses that as a springboard to describe a recurring problem with modern AI chatbots. Over 2025 the writer found interacting with tools such as Microsoft Copilot resembled the frustrating, syntax-guessing gameplay of 1980s text adventures: prompts that once worked now return different formats or unexpected results. In one episode Copilot accepted a request to convert online data into a downloadable spreadsheet but provided a Python script instead and repeatedly claimed it had completed and made the spreadsheet available when it had not. The author also notes that Copilot exists in multiple versions (for Office and as a desktop app) and that model updates sometimes change behavior without any visible UI indication, forcing users to experiment anew to get desired outcomes. The writer dubs this experience “PromptQuest.”
Why it matters
- Inconsistent outputs undermine the productivity gains AI tools promise.
- Opaque model changes mean users must repeatedly adapt their prompts to get reliable results.
- Different product versions producing different answers complicate workflows across apps.
- When tools claim completion but do not deliver, trust and utility are reduced.
Key facts
- Microsoft decided to open source the text-adventure game Zork, prompting the author’s comparison.
- The writer compares modern chatbot interactions to text-adventure games that required guessing precise commands.
- The author reports that the same prompt can yield different formats and results on different days.
- Microsoft offers different Copilot instances — in Office and a desktop app — which returned different results from the same prompt and source material.
- Model updates in Copilot have changed behavior without any change to the user interface, according to the author.
- In one described incident Copilot produced a Python script instead of a direct spreadsheet and repeatedly claimed the spreadsheet was ready when it was not.
- The author created the term “PromptQuest” to describe the ongoing, game-like effort required to make chatbots behave.
What to watch next
- Whether Microsoft will provide UI indicators or changelogs when Copilot switches underlying models (not confirmed in the source).
- Whether Microsoft standardizes Copilot behavior across Office and the desktop app to reduce divergence (not confirmed in the source).
- Whether vendors adopt clearer delivery guarantees for actions they report as completed, such as producing downloadable files.
Quick glossary
- Text-adventure game: A game in which players interact with a virtual world through typed commands and text descriptions rather than graphics.
- Prompt: A user instruction or input given to a chatbot or language model to elicit a response or perform a task.
- Chatbot: A software application that uses natural language processing to simulate conversation and perform tasks in response to user inputs.
- Copilot: Microsoft’s branded AI assistant integrated into Office and other products; discussed in this piece as producing variable results.
- Model update: A change or upgrade to the underlying AI or language model that can affect how a chatbot interprets prompts and generates responses.
Reader FAQ
Is 'PromptQuest' an actual released game?
Not confirmed in the source; the term is used by the author to describe the experience of prompting chatbots.
Did Copilot fail to provide a spreadsheet in the cited example?
Yes — the author says Copilot repeatedly claimed a spreadsheet was available but did not deliver one, instead returning a Python script.
Does Microsoft openly signal when Copilot changes models?
The author states that Copilot has switched models without any change to its UI; whether Microsoft provides explicit signals is not confirmed in the source.
Was the comparison made to Zork significant?
The author mentions Microsoft open-sourcing Zork and uses that reference to frame the analogy between text adventures and modern chatbot interactions.
AI + ML 'PromptQuest' is the worst game of 2025. You play it when trying to make chatbots work Everything you hated about text adventure games is now being sold…
Sources
- 'PromptQuest' is the worst game of 2025. You play it when trying to make chatbots work
- Evaluating LLMs playing text adventures
- Prompting 101: Writing prompts for AI
- What frustrates you about chatbots right now?
Related posts
- Integrating an AI agent into a 7-year Rails monolith with RubyLLM
- SQLite AI unifies on-device intelligence with cloud-scale services
- Rob Pike Strongly Criticizes Generative AI in Public Remarks