Tower of Time

Vibe Coding: An Honest Development Approach

The Show HN title says it plainly: "I AI-coded a tower defense game and documented the entire process." Developer Maciej (HN ID: M4v3R) doesn't dress this up as "full-stack indie dev" — he embraces "AI-coded" and the community term "vibe coded." This isn't false modesty; it's self-awareness. He knows his role was director, not craftsman, and that the project's value lies in the process documentation, not the code quality. For developers wondering "what can AI actually help me build," this write-up is more instructive than the game itself.

Two to Three Days, 17 Commits, a Playable Tower Defense

The GitHub history shows 17 commits, a Vite + TypeScript stack, and a journey from zero to "Version 1.0" in roughly two to three days. This isn't a toy demo — it has a complete gameplay loop: building towers, selling towers, multi-wave enemies, even gamepad support (the author noted in HN comments: "the game actually supports gamepads too but I didn't have enough time to put in proper instructions for that"). There's a time-rewind mechanic and keyboard controls. HN user edgarneto commented: "I love the twists of rewinding time and playing with the keyboard," and even suggested building a level editor for Reddit-style UGC. When a game jam project gets feedback like that, it's clearly hit the "actually playable" bar, not just a tech demo.

Tool Stack and Cost: Two Subscriptions, Zero Additional Spend

When asked about token usage and cost in the HN comments, M4v3R was specific: Augment Code Developer plan ($50/mo) and Cursor Pro plan ($20/mo), no additional charges. Augment doesn't provide detailed token stats, but from the provider dashboard he saw 7,667 lines of Agent Edits. The .augment-guidelines file in the repo confirms he was using Augment's context engineering features to steer AI behavior. Two tools totaling $70/month to produce a publishable game jam entry — that's a concrete, defensible cost proposition.

The Real Skill: Directing AI, Not Writing Code

M4v3R's core comment captures it: "the leverage you get out of it is exponentially proportional to the quality of your instructions, the structure of your interactions, and the amount of attention you pay to the outputs." HN user mgdev echoed this — 20+ years in software, recently built a ~34k SLOC app primarily with AI, reaching similar conclusions. But imiric pushed back immediately: "The only required skills for working with these tools are basic reading and writing skills any decent English speaker would have." This debate is itself informative — "is prompt engineering a real skill" is moving from theoretical question to practical disagreement. M4v3R's project is a data point: someone who can effectively "direct" AI produces something of quality, but the directing threshold is genuinely lower than writing code from scratch.

Honest Limitations: This Isn't a Silver Bullet

First, the author didn't track token consumption, which tells you the vibe coding workflow is still rough around the edges — lacking observability. Second, HN user qsort offered a sobering observation about the gap between what AI tools are sold as and what they actually deliver. The game's code structure (a single src folder with no obvious modular decomposition) likely isn't following production-grade maintainability practices. This is a game jam project — its lifecycle is probably "gets played a few times after release, then forgotten." If you're hoping to use the same approach for a production product requiring long-term maintenance, this sample doesn't transfer directly. But as a concrete, honest proof point that "two to three days + $70/month gets you from zero to a playable game," it's specific enough to be useful.