The author previously identified November 2025 as a turning point for coding agents — when Opus 4.5 and GPT-5.1 reached a level that made their coding agents reliable everyday tools for software engineers. In October 2026 the author sees another inflection point, this time for vibe coding: creating small, enjoyable browser games with minimal hands-on coding.
What changed
Claude Opus 5.5 in particular demonstrated meaningful improvements across areas important for making a compelling game:
- 3D modeling
- Music and audio generation
- Game design
Game design is the critical piece: models have been able to produce things that look like games for some time, but producing a genuinely fun, well-balanced gameplay loop is harder. The author reports that a simple directive — "Make it a LOT more fun" — produced substantial improvements in playability when applied to Opus 5.5.
Four browser games built mostly by prompting
The author built four HTML/JavaScript games that run entirely in the browser, each created largely by chat-based prompts. Brief descriptions follow.
Tree Appreciator
The smallest and most relaxing example is Tree Appreciator. Using the Claude chat interface (no special coding mode), the author asked for a three.js game showing two algorithmically generated trees where the mechanic is appreciating differences. Claude produced a mobile-friendly spot-the-difference style game in a single shot, with pleasing colors and a small "herbarium" collection mechanic that provides a cozy sense of progress.
Emberwick
A more ambitious project, Emberwick began as a three.js top-down (angled) RPG prototype: a fantasy village with walkable exteriors, transparent walls when entering buildings, NPCs, furniture and multi-story buildings. The author iteratively asked the model to add terrain variation, basements connected by tunnels, a rocky mountain with a cave and a sleeping troll, then to turn the prototype into a full game: expand the playable area, introduce fun mechanics, add a non-cliché plot with twists and a magic system.
This title required multiple rounds of playtesting and prompt corrections, mainly to resolve gameplay bugs such as characters being in the wrong places for quest completion. The final build is a story-driven experience of roughly 15 minutes and includes an experimental level editor for adding houses, furniture and landscape features.
Battle Drones
The author’s most successful project is Battle Drones, a Command & Conquer-style real-time strategy (RTS) where robot armies build units, harvest scrap and fight over territory. Early versions from July used Claude Fable 5; Opus 5.5 transformed the game when prompted to make it much more fun and later to port it to three.js.
Today Battle Drones supports 2–4 player multiplayer (via a vibe coded WebSockets relay server), medals, replays, seven maps, a random map generator and a set of mutators that change rules. It also supports a mode where different language models play each other by advancing the game five seconds and then pausing to let models interpret the new state and queue commands.
Battle Drones also showcases Opus 5.5’s 3D model generation: the game’s assets can be browsed and were created by the model. The author released Battle Drones under the Apache 2.0 License; shortly after, Gabriel Turpin published Panache, a fork that reimagined the game as steampunk Belle Époque France with new models, music and an animated intro.
Pumpkin Farmer (Quarter Sun Cove)
Inspired by living in Half Moon Bay, home of the Half Moon Bay Art and Pumpkin Festival and its giant pumpkin weigh-off, the author built Quarter Sun Cove: a cozy farming simulator about growing giant pumpkins and managing the October rush. All characters are vegetarian animals (the mayor is a pelican) and the game includes many small delights and Easter eggs: mischievous raccoons, an elephant seal running the weigh-off, harbor seals rolling pumpkins back to the sea, a wandering goat, and superstitions that sometimes come true.
This project is still being refined; the mechanics are nontrivial (for instance, deciding planting times so pumpkins ripen on busy October weekends). The author notes that Opus 5.5 (Ultracode) consumed many tokens and nearly two days of automated processing to build the current version.
Pushback and possibilities
If someone objects to AI-generated art, music, writing or code, vibe coded games combine all of those at once. The broader games community already shows resistance to AI-generated assets, and this is a live debate — some examples have provoked complaints about quality and sourcing of assets.
At the same time, the author argues that this lowers the barrier to making games. What used to be one of the most complex software projects — requiring diverse skills and months or years of work — can now be prototyped quickly on a phone. The author expects a split to emerge: some players and communities will reject AI-built games, while others will embrace the expanded creative access. The author leans toward embracing it but says they won’t submit their vibe coded games to traditional jams because they don’t feel they’ve earned that right.
What if someone takes it seriously?
The author is excited — and a bit embarrassed — that interesting results came from very little creative input. None of the projects took more than four hours of the author’s manual work (though the models did a lot of processing). They suggest imagining what could be achieved by people who invest substantial creative time, better taste in design, and rigorous playtesting and iteration.
How to get started
All of the author’s games so far are HTML and JavaScript hosted on static hosting (GitHub Pages, Cloudflare Pages). Modern browsers — including mobile — are capable enough, and frontier models already know libraries like three.js and Phaser. The author recommends starting with a prompt such as "Build a mobile-friendly three.js game where..." on chat platforms (chatgpt.com or claude.ai). Once a proof-of-concept exists, switch to Codex or Claude Code for further development. And if the game isn't fun enough, try: "Make it a LOT more fun."
Conclusion
In October 2026 the author concludes that AI models have reached a point where they can produce playable, charming browser games with very little human coding. This creates new creative opportunities and raises questions about acceptance, ownership and standards for AI-generated games, while also suggesting that more polished work from human designers building on these tools could produce novel games that wouldn't have existed before.
