AI news ·
Simon Willison builds tree appreciation game with Claude after three-hour AI coding session
Simon Willison built a browser game called Tree Appreciator using a single prompt to Claude Opus 5.5 Max, which took 3 hours and 7 minutes to generate the full experience with music, animation, and achievements.
Developer Simon Willison published a new browser game called Tree Appreciator on October 14, 2026, built entirely through a single prompt to Anthropic's Claude Opus 5.5 Max. The game presents two algorithmically generated trees and asks players to spot the differences between them, a concept Claude developed during a three-hour processing session after being told to make something "both fun and beautiful."
Willison documented the experiment on his weblog, sharing the exact prompt he used: "Build a three js game called 'Tree appreciator' - it shows you two algorithmically generated trees and the game mechanics is about appreciating the differences between them. Think hard about how to make sure it is both fun and beautiful." He ran this through the default Claude iPhone app in chat mode rather than using the more technical Claude Code environment.
The result, after 3 hours and 7 minutes of processing time, was a cozy spot-the-difference game complete with music, animation, and an achievement system. Willison noted that telling the model to "make sure it is fun" works "uncomfortably well," a finding he described as somewhat embarrassing given a recent talk he had delivered.
The prompt behind the game
Willison's approach reflects a growing practice sometimes called vibe-coding - describing desired outcomes in natural language and letting the model handle implementation details. His prompt specified the technology (Three.js), the core mechanic (appreciating differences between two trees), and the aesthetic goal (fun and beautiful). He did not provide specifications for the music, animation system, or achievement mechanics. Claude generated those elements independently.
The three-hour processing window is notable. Most AI interactions complete in seconds or minutes. Extended thinking modes allow models to explore design possibilities, test approaches internally, and refine output before presenting a final result. For creative professionals watching these developments, the length of processing correlates with the complexity and polish of the final product.
Why a spot-the-difference game works
The game's premise is deceptively simple. Two trees appear side by side, both generated algorithmically. Players examine them closely, identifying subtle variations in branch structure, leaf placement, or trunk shape. The act of close observation - of appreciating what makes each tree distinct - becomes the core gameplay loop.
This design choice sidesteps a common problem with generative AI games: novelty that fades quickly. Because the trees are procedurally generated, each round presents a fresh pair. The underlying mechanic of careful comparison remains constant, but the specific content never repeats. Willison described the final product as "quite compelling to play," suggesting the combination of music, animation, and achievement tracking created enough reward structure to sustain engagement beyond novelty.
The broader context of AI-generated games
Willison's experiment arrives amid rapid advances in AI coding capabilities. His weblog links to related posts about Claude Haiku 5.5 and broader discussions about LLMs and generative AI. The post also carries tags for vibe-coding and Three.js, situating Tree Appreciator within a growing catalog of projects where natural language prompts replace traditional programming workflows.
For creatives and writers, tools like Claude Opus 5.5 Max represent a shift in who can build interactive experiences. A writer with no JavaScript knowledge can describe a game concept and receive a working implementation. The barrier moves from technical skill to clarity of description and taste in evaluating results. Courses on prompt engineering and AI code generation have emerged to teach exactly these skills - how to specify intent precisely enough that models produce useful, polished output.
Why this matters for creatives and writers
The Tree Appreciator experiment demonstrates that a single well-crafted prompt can produce a complete, polished creative work - music, visuals, interaction design, and reward systems - without the creator touching a line of code. For writers and other creative professionals, this changes the calculus of what's possible to prototype alone. The skill that matters most is not technical execution but the ability to describe an experience clearly, evaluate what the model produces, and iterate toward a compelling result. Willison's prompt worked because it named a specific technology, defined a clear mechanic, and set an aesthetic direction. Those three elements - platform, mechanic, and feel - form a template worth studying.