What happened
Reco.ai developed gnata, a pure-Go implementation of JSONata 2.x, using AI to port the test suite and build the evaluator. This seven-hour project, costing $400 in tokens, eliminated an expensive JavaScript RPC dependency within Reco.ai's data pipeline. The new gnata library achieved a 1,000x speedup on common expressions and initially cut compute costs by $300K annually, later expanding to $500K/year through further optimisations to the rule engine.
Why it matters
Platform engineers and CTOs face substantial operational costs from cross-language communication overheads. Reco.ai's gnata project demonstrates how AI-assisted refactoring can directly address these, reducing compute spend by $500K annually and improving performance significantly. This follows Cloudflare's similar AI-driven Next.js rewrite, establishing a pattern where robust test suites support rapid, high-impact code modernisation with minimal human effort and token expenditure. Existing test coverage accelerates targeted rewrites, delivering substantial cost reductions and performance gains.
