Monta runs the software behind EV charging infrastructure for the world, powering charging for Toyota, Stellantis, and Copenhagen Airport, plus a majority of Nordic energy providers. I joined as the first UX leader at Director/VP level, hired to formalize design practice and scale it without being able to hire our way out. The answer was infrastructure.
Design was a structural bottleneck
Engineers outnumbered designers 10:1. PMs built without design input. Research was ad hoc. Every team had its own tools, processes, and quality bar, with no shared infrastructure to tie them together.
Monta competes with car companies and energy companies daily. Product velocity is existential. The CEO is a three-times-exited founder with one playbook: enter a niche technical space, dominate it with software. Headcount wasn't the answer. Infrastructure was.
The mandate: Build a capability layer that multiplies the people we already have, making every designer, PM, and engineer a better contributor to the product process. Not just a design system. Not just AI tooling. A full stack of interlocking systems.
Don't hire more designers. Make everyone a better one.
The world was already moving toward PMs designing, engineers designing, and agents designing. The question wasn't how to protect design. It was how to raise the floor for everyone doing design work, while freeing senior designers for the work only they could do.
This became a technical architecture: one governed AI backbone, a growing library of shared capabilities, a pipeline that removed the translation layer between design and engineering, and a monitoring system that caught quality drift before it became a problem.
Four systems. One backbone.
Each system compounds on the others. Together they form a complete capability layer, from individual contributors to the automated product lifecycle.
The AI backbone of the organisation
An API gateway giving every employee governed, air-gapped access to Claude, connected to every internal data source. We run the Danish power grid. Data cannot leave. Bridge gave us full Claude intelligence against live internal data while keeping everything within our VPN perimeter: GDPR compliance by architecture, not by policy.
250+ shared capabilities across every department
Pre-built Claude capabilities scoped to a role or task. Any employee could build and publish one. The generative UI skills read the live design system, pulled real Storybook components, and generated production-ready code in the actual stack. Output: a merge-ready PR that could go to QA that day. Sales reps generated custom demos without touching product.
We eliminated the handoff. Entirely.
The traditional design → spec → eng → QA → prod cycle replaced with a model where designers and engineers operate as peers in the same codebase. We run one of the largest Flutter apps outside Google, writing ~30% of its open-source code. PMs define Fibonacci 1–3 point scope. No mediation between design and engineering. Honest trade-off: governance had to catch up with velocity.
You can't govern what you can't see
Token usage dashboard by role, region, and seniority. Deeper telemetry than a standard deployment, surfacing the most common request types clustered by team, and similar-but-different prompts to unify as shared skills. Weekly: top 50 usage patterns reviewed. Cherry-picked patterns promoted to shared skills, diverging teams flagged before quality became a problem.
30 agents. 90% of the product lifecycle automated.
Inspired by Booking.com's experimentation culture and built for an AI-native context. Signal detection to production deployment to performance verification: a closed loop where every measurement feeds the next round of signals.
Signal sources included social media comments, App Store reviews in all markets and languages, SDR/AE/CSM touchpoints, public roadmap submissions (roadmap.monta.com), PostHog analytics, and support ticket clusters. Everything connected. Nothing manually entered. The biggest challenge stopped being "how do we get people to use AI" and became "where should we not use AI." That's when you know infrastructure worked.
Four AI agents shipped from internal infrastructure.
None would have shipped without Bridge, the Skills Library, and the monitoring system as the foundation. Each agent was a product in its own right, built on the same capability layer the internal team used daily.
Driver Support Agent
Autonomous customer support covering all 11 supported languages, resolving the majority of driver issues without human escalation.
NOC Agent
Network Operations Center automation: autonomous infrastructure monitoring and response, reducing manual intervention in grid-connected operations.
Data Agent
Natural language to SQL: enabling any team member to query operational data without engineering dependency or SQL knowledge.
Knowledge Agent
Institutional memory at scale: surfacing relevant historical decisions, patterns, and precedents across the codebase, design system, and documentation.
Five education programs. Two leaders. One direction.
88% weekly AI usage didn't happen by mandate. It happened through five structured programs, none of which were required.
AI & Automation Huddles
Company-wide sessions led by Brian Rountree (VP Engineering, AI & Automation). Strategic direction, tooling decisions, infrastructure updates, cross-department use cases.
Design×Claude Weekly Training
Not lectures: actual building. Live prompt engineering, skill creation, running agents, querying real data via Bridge. Every participant left with something working they'd built themselves.
AI Show & Tell
Open mic format. Volunteers signed up via open agenda. Anyone showed what they'd built, what worked, what failed. No hierarchy. Best ideas spread because people saw peers doing real things.
AMAs with AI Experts
External experts for Ask Me Anything sessions on advanced prompt engineering, LLM fundamentals, AI ethics, and governance. Addressed the "mountain of unknown unknowns" causing paralysis in less experienced roles.
AI Ambassadors
One per function. Bridged central infrastructure decisions and day-to-day team needs. Fed insights back into roundtables. Helped less experienced members overcome initial paralysis in a context they trusted.
External Trainers
Organised the Product offsite with Booking.com, ABSmartly, and PostHog trainers on experimentation methodology and AI data analysis, giving the Product team rigour that fed directly into the ADLC system.
New infrastructure for how designers grow.
I joined as the first UX leader at Director/VP level, so I built the org from scratch. Hiring across design, research, and content with clear levels, career frameworks, and ways of working that didn't exist before. Every hire was deliberate: we needed range, not just seniority.
AI fluency changed what every role meant. I led the rewrite of every role definition, level framework, and JD across design, research, and content, defining what good looked like in an AI-native team, what player-coaches owned, and how to evaluate people in a world where everyone ships code.
What I'd do differently.
The hardest part of first-in-role leadership isn't building the thing. It's building trust while building the thing. You're being evaluated constantly while operating without precedent. The answer is to lead by example early, delegate fast, and let the team's output speak for the org's value.
On the infrastructure side: governance ships with capability. Always. The AI monitoring system should have launched with the pipeline. The PR quality gates should have been defined before the first designer opened a repo. The Ambassador programme should have preceded the tool rollout, not followed it. Speed created real problems: the designer-to-code pipeline scaled velocity faster than oversight could follow. Shipping capability and governance in parallel isn't a nice-to-have; it's the architecture.
"The biggest challenge stopped being how do we get people to use AI, and became where should we not use AI."Kevin Hawkins · VP Product Experience, Monta