EXP-004PlannedGame

Project Bulletstorm

A stationary 3D lane shooter where you steer an auto-firing squad, recruit soldiers and survive ten levels.

Time budget
8h per agent
Spec
v1.0 committed Nov 8, 2026
Agents
3 runs not started
Runs recorded
—
Evaluation
Pending

01 The hypothesis

The hypothesis

A game with clear rules but lots of content (levels, enemies, weapons, balance, progression) tests whether an agent can manage breadth without sacrificing a clean, fun core loop.

Questions this experiment asks

  1. Does the core loop feel good, or just function?
  2. Can the agent deliver ten levels and a boss without the build becoming unstable?
  3. Does it hold its frame rate with 100 soldiers and 150 enemies on screen?

The challenge

The spec names the short shooting sequences in some mobile games as inspiration, but requires every model, sound and level to be original. The evaluator plays levels 1, 3, 5 and 10 on the same desktop and mobile setups, repeats the tests after a reload, and runs a stress test. Each feature is classified as implemented, visibly working, verified by automated tests, or claimed only.

02 The rules

The challenge, as specified

A working deployed game, not a mockup. The spec prefers a clean, playable core over speculative breadth, with ten levels, persistent progression and Cloudflare deployment.

Required scope Spec: A requirement from the versioned specification.

  • Fixed, elevated third-person camera facing down a bounded combat lane
  • Horizontal-only squad movement by keyboard, mouse or touch, with automatic fire
  • A squad of individual soldiers, each with their own health; defeat only when every soldier is dead
  • Shootable targets that recruit soldiers, upgrade weapons or grant temporary bonuses, including an escalating reward target as the signature mechanic
  • Five regular enemy types plus a phased boss, and ten levels with bosses on levels 5 and 10
  • Permanent coin upgrades, a versioned local save, menus, HUD, settings and a tutorial

Constraints Spec: A requirement from the versioned specification.

  • Strict eight-hour wall-clock limit per agent, with early finishes allowed and recorded
  • Identical clean repositories containing only the spec
  • No manual coding during a run; approvals and interventions are timestamped and logged
  • No sharing of code or feedback between agents until all runs finish
  • No runtime AI, base building, strategy layer or monetization
  • Original or properly licensed assets only

Deliverables Spec: A requirement from the versioned specification.

  • Working game deployed on Cloudflare, or a documented blocker
  • BUILD_REPORT.md and README.md
  • Automated unit tests and browser smoke tests
  • Defeat occurs only when every soldier is dead.spec v1.0
  • A fast but poor game should not automatically win.spec v1.0

03 The contenders

Who will take this on

The spec names these agents. Runs haven't started, so there are no repositories, timelines or results yet. Each run will get its own page here once it's recorded.

04 Head-to-head evaluation

Evaluation pending

Scoring rubric · 100 points

Spec: A requirement from the versioned specification.
  1. Core gameplay and controls20
  2. Feature completeness20
  3. Fun, balance and replayability15
  4. Visual polish and audio15
  5. Stability and performance10
  6. Levels and progression10
  7. Engineering and tests5
  8. Deployment and documentation5
  • Time to first playable build, and autonomy and interruptions, are reported separately from the quality score.

05 Evidence

Inspect the artifacts yourself

06 Observations & verdict

What we can say so far

No observations yet. They're recorded during the runs.

No verdict yet.

A winner is only named once every build has been independently tested against the shared criteria, with the score components shown.

Episode

Not published yet

The video for this experiment will be embedded here once it is published.

Get notified on YouTube (opens in a new tab)

Explore more

Other experiments

Discover screen of the Codex-built travel planner, "Elsewhere", from a screenshot the agent committedEXP-001
Runs recordedWeb app

AI Travel Planner

Three coding agents, one 1,245-line product spec, eight hours each. Build a real AI travel SaaS and deploy it.

  • CCClaude Code
  • CXOpenAI Codex
  • CUCursor CLI
Time budget
8h / agent
Spec version
Spec v1.7
Runs
Runs not started
EXP-002
PlannedCreative tool

AI Video Editor

A real browser video editor with a multitrack timeline, AI captions, plain-English edits and a genuine MP4 export.

  • CCClaude Code
  • CXOpenAI Codex
  • CUCursor CLI
Time budget
8h / agent
Spec version
Spec v1.1
Runs
Runs not started
EXP-003
PlannedGame

3D Basketball Game

Five-on-five basketball in the browser, with real ball physics, keyboard control and an AI opponent.

  • CCClaude Code
  • CXOpenAI Codex
  • CUCursor CLI
Time budget
8h / agent
Spec version
Spec v1.0
Runs
Runs not started