SceneBoard — side-by-side arena for coding agents
Dispatch the same task to multiple coding agents and compare results and cost side by side
- Problem
- Developers juggle Codex, Claude Code, GLM, Grok, and more, but struggle to know which agent to trust for a given task. Decisions rest on scattered gut feel and ad-hoc comparisons instead of systematic evaluation.
- Solution
- A local comparison workbench: define a coding task with a repo snapshot or minimal repro, dispatch it to multiple agents in parallel, collect outputs, diffs, latency, and token cost, and generate a side-by-side comparison with scores so developers can pick models and plans based on evidence.
- Users
- Developers subscribed to multiple AI coding plans, teams evaluating downgrades, and technical writers producing model comparisons.
- MVP
- Configure multiple agent CLIs or APIs; run the same task in parallel in isolated directories; capture diffs, latency, and token usage; generate a side-by-side report; save and export results locally.
- Revenue
- Free tier with limits on parallel agents; paid unlock for more seats and report export; affiliate revenue from model subscriptions.
- Why now
- Multi-model workflows are the norm, but head-to-head comparison is still manual and messy, leaving a clear opening for a lightweight arena.