Grid Bot Arena

GridBot Arena is an AI strategy lab where learners build autonomous TypeScript agents, test them in deterministic competitive scenarios, inspect replays, and improve through structured reflection.

Comment

Inspiration

Learning to build autonomous agents is often abstract: you write code, get an outcome, and have little visibility into why it happened. GridBot Arena makes that learning loop concrete, competitive, and replayable.

What it does

GridBot Arena lets learners create TypeScript bots, run them against advanced opponents, and watch deterministic grid matches unfold. Players can configure matches, inspect live/replay state, follow events and scores, and learn from every decision through verifiable replays.

How we built it

We built on a deterministic TypeScript game engine with a Fastify backend, React/Vite frontend, Zod contracts, JSON persistence, and isolated child-process bot execution. During Build Week, we used GPT-5.6 through Codex to inspect the system, plan focused changes, improve the responsive replay experience, validate compatibility, and produce the demo/evidence package. GPT is not required at runtime.

Challenges we ran into

The biggest challenge was improving branding and the replay UI without breaking deterministic game rules, bot isolation, persistence, archives, or replay compatibility. We also had to make a dense 16×16 four-player match understandable in a 1080p demo.

Accomplishments that we're proud of

  • A polished GridBot Arena identity focused on education and strategy.
  • Deterministic four-player matches with advanced built-in opponents.
  • A responsive replay command deck that keeps the board, events, and score context visible.
  • Secure bot validation and isolated execution boundaries.
  • A narrated demo showing a 16×16, 600-turn match at 5× replay speed and its final scoreboard.

What we learned

Good AI-agent learning tools need more than agents—they need reproducibility, visibility, and safe experimentation. We also learned that Codex is most valuable when paired with clear specifications, incremental validation, and human review of architecture and product decisions.

What's next for Grid Bot Arena

We plan to expand learning scenarios, bot templates, strategy analytics, and classroom-friendly workflows. A post-match Replay Coach is a future improvement: it would explain verified public replay outcomes without accessing bot source, hidden state, credentials, or private logs.

Built With

Share this project:

Updates