Director of Agentic Delivery | IRIS

23 years in technology: 9 as an engineer and 17 leading delivery, with 3 of those years doing both. From managing engineering teams at Sotheby's, through ML infrastructure at ASOS (Kubernetes on Azure for the in-house LLM team) and engineering pipeline work at Sonovate, to cross-enterprise AI programme delivery at Schroders, and now agentic delivery at IRIS. Back producing code - with agents.

The missing piece in most AI-assisted engineering is not the code generation, it is the structure and guardrails around it. Process that worked at human coding pace breaks when output accelerates ten-fold. That is the problem my engineering and delivery experience is built to solve.

At Schroders I delivered AI programmes across EPP - shaping architecture decisions with 8+ stakeholder groups, governing 80+ Architecture Decision Records, and building production systems: a live portfolio dashboard on a 30-minute pipeline, a meeting intelligence system that converts Teams transcripts into structured actions, and a multi-agent automation layer with a 69-test stability scaffold. Also: Zero Trust AI audits and ADR governance schema design. I also led delivery for the cloud and infrastructure teams - mature Azure estate running in regulated financial services.

On the side I build the tools I wish existed:

• Registrum - production SaaS API for UK company data, built AI-native with Claude Code. Includes @registrum/mcp, an MCP server that exposes company data as agent tools for Claude Desktop, Cursor, and any MCP client.
• Workflow dispatcher - multi-agent orchestration across 12 projects, running every 3 hours. Voice dictation on phone -> Microsoft To-Do -> Node.js dispatcher -> one Claude Code session per task, with a definition-of-done gate, locks, alerting and a weekly retro on how the agents performed. Around 770 tests.
• BigCollage - photo-collage gifts that run entirely in the browser until payment, with face-aware layouts. Idea to live card payments in under two weeks; A3 prints through a UK print partner in testing.
• Cupel - AI trust and credentials governance framework. Original thinking on what EU AI Act enforcement (August 2026) means for professional provenance, with a W3C VC-aligned open spec.

Skills in practice: agentic delivery, AI programme delivery, agentic systems design, ADR governance, Microsoft Graph API, MCP server development, Claude/LLM integration, multi-agent orchestration, EU AI Act.

How I work: delegate deliberately, describe the full context, check what comes back, and own what ships.

Contact: LinkedIn · eugene.vandermerwe@gmail.com