One year of LoxiLabs: what held up and what we changed

Published 1 September 2026 · LoxiLabs

What held up

P&L impact as the only scoreboard. Every engagement this year had a metric, an owner, a control, and a kill criterion — and some Prove sprints ended in a fast no, which is exactly what the kill criterion is for. Small senior teams. Production readiness before demos: we haven't had to rebuild anything between prototype and production, because the prototype was already the production codebase at an early stage.

What we changed

The Find phase got tighter: up to four weeks, one ranked value map, one decision. The Agent Ops Framework became a product rather than a set of habits after the assurance review we wrote about in December. The Customer 360 Accelerator lost scope and gained speed.

Where the work went

Financial services, telecommunications, government, and utilities. Trading volume, prepaid market performance, and governed operations automation are the outcomes we can describe; client names are shared under NDA.

The partners

OpenAI, Anthropic, Databricks, Snowflake, and Google Cloud. Depth on each, and the independence to say which one fits — including, more than once this year, none of them for the problem at hand.

Next

More reusable IP, published as it is proven. More writing from delivery. And the same test on every piece of work: did a number the business reports move, and can we show the control?