According to Greg Brockman on X, Ethan Mollick tested Excel agents from Claude, OpenAI ChatGPT, and Microsoft Copilot to create a working strategy game inside Excel with basic graphics, and only ChatGPT delivered a fully playable game with a formula‑driven "smart" enemy, while Claude acted as game master with a board and Copilot produced a board without full gameplay (source: Greg Brockman citing Ethan Mollick’s post). According to Ethan Mollick’s original X post, the ChatGPT output leveraged complex spreadsheet formulas to implement turn logic and enemy decision rules, demonstrating that LLMs can operationalize game AI heuristics directly in cells without macros, which lowers deployment friction for enterprise environments that restrict VBA. As reported by the shared posts, this highlights a practical business opportunity: using LLMs to auto‑generate domain‑specific simulation tools and lightweight serious games in Excel for training, planning, and what‑if analysis, with rapid iteration and low IT overhead. According to the posts, the comparative results suggest product differentiation among AI assistants for structured tool creation tasks, positioning ChatGPT as stronger at end‑to‑end Excel logic synthesis, Claude as a collaborative facilitator, and Copilot as UI‑first; this has go‑to‑market implications for vendors targeting finance, operations, and education workflows where spreadsheet‑native AI agents can deliver immediate value.