A senior developer shipped a fully playable web game at downtime.partridge.works in roughly 48 wall-clock hours last month—roughly 12 to 14 hours of actual focused work. No team. No pre-existing engine. Just Claude Opus 4.7 and a tight prompt-review-iterate loop that treated the AI as a fast pair programmer rather than an autocomplete button. The result wasn't a barebones prototype; it was something with taste, juice, and replayability.

The Context Window Unlock

Here's what changed in early 2026: the bottleneck shifted from "can I write this code" to "can I describe what I want clearly enough." Opus 4.7's 1M-token context window means the model can hold the entire design doc, current codebase, and running decision log across multiple days of sessions without losing thread. That's not just convenient—it's the actual unlock that makes weekend game development feel qualitatively different than it did even a year ago.

Day One: Foundation in Focused Chunks

Day one was almost entirely planning and core loop work. The developer wrote a one-page design brief—what genre, what happens in the first ten seconds, what hooks replay—and dropped it into Claude as the seed. From there, each code generation targeted a single concern at a time: input handling, then collision detection, then enemy spawning logic, then the scoring loop. Each chunk got tested in a real browser before the next prompt fired. By end of day one, the prototype was playable but ugly. The foundation mattered more than the polish.

Day Two: When Iteration Became Cheap

Day two was polish—particle effects, screen shake, sound design, title screen, difficulty curve tuning. This is where the workflow paid off most dramatically. The cost of "try a different juice effect" dropped to about two minutes of round-trip prompt time, so the developer tried dozens of variations they would never have hand-coded because each iteration used to be too expensive. AI-assisted development didn't eliminate the need to step back and think—it just made the feedback loop brutally fast.

The Three-Move Workflow

The pattern this developer used has three moves you can copy directly. First: plan before code, every time. Ask Claude to write the implementation plan—what files exist, what functions go where, what data structures hold state—before touching any generation. Read it, push back on overengineered parts, then approve and execute. This sounds slower than generating directly. It's not. Second: constrain the diff surface ruthlessly. "Add a particle effect when player gets coin" is a bad prompt because Claude doesn't know your codebase's conventions. "Touch only src/effects/particles.ts and the coin pickup handler in src/entities/coin.ts. Match existing emitter pattern" is a good one. Specificity about which files can change is the single highest-leverage habit in this workflow.

The DECISIONS.md Discipline

One less obvious but critical habit: keeping a DECISIONS.md file that gets pasted into every new Claude session. It captures the why behind early choices—"we picked Phaser over plain canvas because X" or "enemies spawn from screen edges only to preserve tight hitbox feel." With a 1M context window you can paste entire repos, but a clean decision log is worth more than raw code because it tells the model what not to relitigate on day two. No quiet backpedaling on settled architectural choices.

What This Signals for Solo Devs

The story isn't "AI replaces developers." Senior judgment does the heavy lifting at every step: deciding what to build, recognizing generated approaches that are overengineered, knowing what "juicy" feels like without being told. The model is a fast collaborator who has read every game dev tutorial ever written and has zero ego about redoing work. What's actually new in 2026 is the ratio—a weekend used to mean a barebones prototype; with this workflow it produces something that has taste.

Key Takeaways

  • Opus 4.7's 1M-token context window enables multi-day sessions without losing design thread
  • Generate code one concern at a time; test each chunk before the next prompt fires
  • Constrain which files can change in every prompt—specificity is leverage
  • Keep a DECISIONS.md file to prevent Claude from relitigating settled choices
  • The bottleneck shifted from "can I write it" to "can I describe it clearly enough"

The Bottom Line

If you've been waiting for AI tools to be "good enough" for serious solo work, they crossed that line. The remaining question isn't whether the stack is ready—it's whether you can articulate what you want with the precision of an art director rather than the vagueness of a customer. That skill is the new competitive moat.