September 2026 Summaries

Qonto deployed Edgee as an agent gateway to give engineers access to multiple AI models through their existing coding tools while centralizing identity, governance, usage visibility, routing, and spending controls. After migration began in late July, adoption rose from 10 to 350 engineers during a five-week opt-in period, and 500 engineers were using the platform by 10 September before it became the default through Jamf-managed configuration. Qonto chose to buy rather than build because maintaining compatibility across rapidly changing agent harnesses, desktop tools, identity systems, prompt caching, and optimization features would have required substantial continuing effort. The company assessed Edgee’s EU data residency, GDPR compliance, DORA classification, data-handling practices, and fallback capabilities, while retaining OpenTelemetry instrumentation in agent harnesses as its primary audit source. Compression reduced median costs by 17%, while compression combined with model rerouting produced a reported 37.7% cost reduction, though the company distinguishes these as separate measures. Qonto’s remaining challenge is evaluating which models best fit particular tasks at scale, as routing decisions currently rely on experiments, engineer feedback, and delivery outcomes rather than a formal evaluation platform.