Design systems and design orgs are the same argument.

The shift I'm designing for

Every B2B product built in the last decade assumes the same thing: the user knows what they want to do, and the interface is a menu of ways to do it. AI breaks that assumption. The product should know the state of your business and act on it. My work is figuring out what that looks like when the stakes are an audit you can fail.

Five principles I hold teams to

1. AI as a system, not a feature. Always present, not a button in the corner that opens a chat box.
2. Explanation over configuration. Show why a decision was made. People trust systems they can audit, not systems they can tune.
3. Human in the loop by default. Autonomy is earned step by step and gated on metrics. People authorise, approve and override.
4. Interfaces that adapt to confidence. The UI changes shape with risk and model confidence, instead of a fixed layout pretending to be certain.
5. A readable record. A running account of what the system did and why, readable by a person and by an auditor.

How I run a design org

I treat the design function like a product: it has users, a roadmap, a quality bar and metrics it's accountable for.

• Design tied to business metrics. Team goals are written against NPS movement, support-ticket reduction, retention and feature adoption, not screens delivered.
• Structure instead of my calendar. Six pods, each with an owner and a measurable charter, so design scales without routing everything through me.
• A team operating system. Hiring, onboarding, retros, digests, weekly wins, initiative tracking and an AI prompt library, all written down.
• Design systems as the repeatable bet. Four built from scratch across four companies: 6+ months to 8 weeks at Chargebee, 3 months to 4 weeks at Yubi, and 25% faster with 92% coverage at Sprinto.
• Research as a standing loop. Monthly NPS analysis, five conversations a month with low-scoring customers, quarterly ticket themes, and at least ten user tests per flow before a revamp ships.
• Governance for AI work. Human in the loop by default, autonomy gated on metrics, and a monthly design governance review. Over-automation is a managed risk, not a hoped-away one.
• Opinionated over configurable. When a customer asks for a setting, it usually means we haven't decided something. I push the team to make the call.
• Writing as the main tool. Strategy docs, pattern glossaries and roadmap narratives are how alignment actually happens. Much of my best design work is prose.