How to Make AI Write Better Solana Code

AI coding tools can bring the wrong assumptions to Solana. Josh Siegel from Jump Crypto shows how to give Claude and Codex the context they need to write better Solana code.

Josh walks through his core prompt, Solana security checklists, test-driven development, and compute unit benchmarks. He then breaks down an oracle update optimization that reduced compute usage from 34,329 to 293 CU for a single instruction.

Follow six steps: give AI Solana-specific context, prototype the logic and prepare test data, test correctness and coverage, benchmark compute units, protect your tests, and review code with another AI agent. The case study then shows how Josh applies these ideas to a real optimization problem.