The AI Design Authority sets the standards, guards the quality, and clears the path for every AI product we build for clients. One bar for good. Applied consistently, from first prototype to production.
AI products fail in predictable ways — unmanaged model risk, architecture that cannot scale, interfaces users do not trust, and delivery that cannot be maintained after handover. The Design Authority exists to make those failures rare.
We are not a gatekeeper that slows teams down. We are the shortest route to a product that is safe, sound, usable, and ready to run — and to the evidence that proves it.
What we own. The standards, the review gates, the reference patterns, and the decision on whether an AI product is fit to ship. That runs both ways across Beacon, our AI portal: we review the internal Claude skills that enter the AI Library, and we clear the client-ready accelerators and products that list on the Asset Exchange.
One thing we will not do. We will not wave through a model into production on accuracy alone — nor onto an unapproved tool or a prohibited use case. A model that is accurate but unmonitored, unexplained, or unowned is a liability we decline to sign off — every time.
We operationalise Capgemini's Group guidance — we do not reinvent it. Every standard and gate traces to source.
Every AI product we build is measured against the same four domains. Each pillar carries mandatory standards and recommended practice.
Ethics, bias, transparency, human oversight, and regulatory alignment — including the EU AI Act.
02Reference architectures, model and tooling choices, LLMOps, security, and evaluation.
03Discovery, human–AI interaction patterns, accessibility, and the design system.
04Lifecycle gates, testing, code quality, documentation, environments, and handover.
The Authority operates through three connected mechanisms. Teams stay in control; the Authority provides the standard, the review, and the sign-off.
A living library of what good looks like across the four pillars. Mandatory where risk demands it, recommended everywhere else. You build to these from day one.
Browse standards →Five lightweight checkpoints across the product lifecycle. Clear entry and exit criteria. Proportionate to the risk class of your product.
See the gates →Office hours, on-demand reviews, and a fast lane for waivers. We meet you where the work is — not the other way round.
Engage us →Not every product needs the same weight of review. We classify every engagement so effort follows exposure.
Self-certification against the standards. Light-touch gate at design and pre-production.
Full gate reviews with the Authority. Mandatory Responsible AI assessment.
Named reviewer, all gates, independent Responsible AI and security sign-off before launch.
The cheapest fix is the one you make before you write the code. Engagement at discovery costs least and saves most.
