A Reddit post from user sharkymcstevenson2 demonstrates what “long horizon thinking” looks like in practice for AI-assisted game development. A single base prompt, one extended reasoning pass on the tesana engine, and a playable two-player boss-rush platformer emerged — without a single pre-existing asset or visual reference used as input.

The actual prompt — shared in comments after prodding — is notable for its specificity and ambition:

“Build a 2D platformer-inspired boss-rush shooter utilizing state-of-the-art procedural graphics… featuring 1930s rubber-hose style procedurally generated animations, multi-stage boss phases, frame-perfect hitboxes, fluid parry mechanics, procedural ink-and-paint textures, and authentic vintage jazz audio integration.”

Key details from the thread:

  • One-shot output — The AI iterated on its own build internally through “LOOP prompt mode” (a tesana feature extending the reasoning horizon), revising code and assets across multiple passes from a single initial instruction
  • ~$30 cost — API compute for the full generation, on a closed-source engine that allows code export
  • 2-player coop, three multi-phase bosses, ~2 hours of playtime — the output was forkable at tesana.ai for others to build on
  • No post-generation changes yet — the author hadn’t tried modifying anything after the initial oneshot, leaving questions about iterative editability unanswered

The reception was deeply skeptical. Commenters accused the post of being a veiled marketing push for tesana, since the author initially refused to explain the technique beyond vague terms and only shared the prompt after repeated demands. The technique itself — giving the model a long internal iteration budget instead of demanding a single-pass answer — is a genuinely interesting workflow boundary. Whether it generalizes beyond a single demo or produces maintainable code is unproven.

The takeaway: long-horizon prompting is a legitimate tactic for complex AI generation, and $30 for a coherent playable game with multiplayer is a striking cost floor. But the credibility gap here (closed-source platform, refused to explain, accusations of bait-and-switch) means the technique needs independent reproduction to be taken seriously as a repeatable workflow.