@buildbrainsystems: A bank cut its AI bill by 90%. Its work got faster. 🕹️ Revolut, the digital bank, used to give its engineers top AI coding tools with strict budget limits. The impact, in the words of its AI operations lead David Abitbol: "not so much, at the time." In April 2026 it switched to an open model, GLM, self-hosted on dedicated chips (about 56 Nvidia B200s for 2,000 engineers) with a flat bill. The bill is now one-tenth of what it spent on Anthropic and OpenAI. Across a sample of 2,000 engineers: a 60% improvement in coding time and a 70% improvement in engineering lead time (picking up a feature to ready to deploy). Why the bigger jump? "The bottleneck for engineering is not necessarily coding." Nobody had to save tokens for the hardest task, so people used AI on the inbox, Jira and Confluence too. The catch: these are Revolut's own before-and-after numbers, not a controlled test. And it's a big bank's engineering team, not a small firm. Our view: don't make your people ration AI. Hand over the busywork first, and keep the top models for the hard jobs. Source: David Abitbol, Revolut, AI Engineer NYC, Oct 2026. Voiceover is AI-generated. Not affiliated with Revolut. #aiforsmallbusiness #smallbusinessowner #airoi #businessautomation #aitraining

buildbrainsystems

Open In TikTok:

Region: US

Tuesday 03 November 2026 16:57:42 GMT