~ stockholm, sweden · ai architect · 20 years shipping software
I'm Martin McCarthy — the architect leading the AI vision for Telenor's software development lifecycle. I design the systems, guardrails, and practices that bring agentic AI into how organisations deliver software: safely, measurably, and at scale — and I keep my architecture honest by building complete multi-agent systems hands-on, end to end.
I'm an Irish software engineer based in Stockholm. My career started in payment systems in Dublin, then ran through lead engineering roles across Swedish fintech, healthtech, and logistics — Qliro, Budbee, Doktor.se — followed by five years as an independent consultant for Nordic enterprises like Nobia and Capacent. I trained in deep learning back in 2017, when it was still a niche pursuit; today AI is the centre of everything I do.
Now I work at the intersection most organisations struggle with: making AI genuinely useful in how software gets built. At Telenor I'm the architect for DevBoost, the team chartered to drive AI uplift across the SDLC — and I lead its AI architectural vision: agentic development workflows, the knowledge systems that make large codebases consumable for AI, and the governance that keeps humans approving every stage. At the heart of it sits what I call the enterprise second brain — knowledge compiled by AI, validated by humans, governed centrally.
And I practise what I preach. As my own applied R&D, I've designed and built a complete multi-tenant agentic platform end to end — three cooperating AI agents for reception, accounts, and marketing — on Google Cloud with Gemini function calling, retrieval-augmented generation, and MCP tooling. Every layer is mine, from the data model to the guardrails — a zero-to-production build that automates real business operations (bookings, invoicing and tax compliance, campaigns) — and I use it to pressure-test every architectural opinion I hold.
That combination is my edge. I can stand in front of an executive team and argue the strategy, and then open a terminal and build the system. And I think in products, not just architectures: technology counts when it moves a business outcome. I don't just advise on AI — I test my ideas against reality before I recommend them.
Joined Telenor as a Solution Architect; since 2025 I'm the architect assigned to DevBoost, the team chartered to drive AI uplift across the software development lifecycle. I lead the AI architectural vision for that mission — agentic development workflows, AI knowledge systems that make large codebases consumable for models, multi-agent delivery pipelines spanning architecture to release, and the governance layer that keeps a human approving every stage. The remit runs from strategy and platform architecture to tool evaluation and hands-on delivery across the engineering organisation.
Five years of senior frontend and full-stack engineering consulting for Nordic enterprises and scale-ups — clients included Doktor.se (digital healthcare), Nobia (Europe's leading kitchen group), and Capacent (Nordic management consultancy). Alongside client delivery, I went deep on applied AI, designing and building a complete multi-tenant agentic platform end to end as personal R&D — the hands-on foundation for the AI architecture work I lead today.
Led frontend development and the agile process at one of Sweden's largest digital healthcare providers, building real-time patient-facing care flows.
Led frontend development at the logistics scale-up, building the real-time systems behind last-mile delivery.
Led a team of six engineers building online payments and checkout for one of the Nordics' largest e-commerce groups.
Opinions and working notes from building agentic systems and rolling AI out across engineering organisations.
Karpathy's LLM wiki became Google's OKF in eight weeks — the knowledge layer is being standardised. My architecture: compiled by AI, validated by domain owners, governed centrally.
read →Identity, guardrails, cost control, knowledge — how the two clouds' managed agent stacks compare on the four pillars that decide production-readiness, and what to build vs rent.
read →A chargeback dispute walked through an automation pipeline, stage by stage: extraction, policy retrieval, calibrated confidence, a decision-class routing table, and the audit trail — with real industry numbers.
read →Agents will draft every stage of delivery. The organisations that win won't be the ones with the most autonomy — they'll be the ones with the best gates.
read →Developers in a rigorous study felt 20% faster and measured 19% slower. How to evaluate AI tooling with evidence instead of vibes.
read →If you're working on agentic development, AI platform architecture, or automating real business processes with AI — the hard, ambiguous, high-trust kind — I'm always glad to compare notes.