ABOUT
US
// 01_ORIGINS
We saw the future before the hype.
Barnacle Labs was founded by Duncan and JD. The idea started when we got early access to GPT-3 (pre ChatGPT) — before most people had even heard of generative AI. We saw what was possible before the hype machine started, and we've been building production AI systems ever since.
Duncan spent years as European CTO for IBM Watson, leading AI implementations across enterprise. JD built open source projects in Duncan's team that scaled to production systems used by enterprise clients.
When generative AI went mainstream, we knew how to build with it — not just talk about it.
// 02_FIRST_MISSION
500 papers a day. Zero way to keep up.
Our first major project was with the National Cancer Institute. Researchers were drowning in literature. Traditional search wasn't working. They needed something smarter.
We built NanCI — an AI system that personalises research recommendations and facilitates collaboration. Built in partnership with Google Cloud, it's now used by thousands of scientists daily.
That set our pattern: find real problems, build real systems, ship to production.
// 03_THE_NAME
Patient work enables breakthroughs.
While working with NCI, Duncan visited Darwin's house — Down House in Kent. Darwin spent 8 years meticulously studying 10,000 barnacles. That detailed, unglamorous work provided crucial evidence for his theory of evolution.
That's what we do. We build the meticulous systems — the infrastructure, the architecture, the production code — that enable scientific and commercial breakthroughs. It's not always glamorous, but it's how real progress happens.
MEET
OUR FOUNDERS
FAQ
Common questions about working with us.
Are you a traditional consultancy?
We build AI-powered systems. Discovery involves strategic work to figure out what to build, but we're always focused on getting to the point where you know what that is. Our deliverables are mostly working systems, not PowerPoint decks.
How long does a typical AI project take?
Discovery takes 2 weeks. Full builds vary - simple AI chat implementations might take 6-8 weeks, while complex agentic systems or custom ML models can take 3-6 months. We've worked on ambitious AI systems that took over 2 years to deliver. We work in sprints and get you working software early, so you see progress throughout.
Do you work with companies outside the UK?
Yes. We work with clients globally. Our team is mostly based in the UK but we've delivered projects for organisations in the US and Europe.
What if we want to build it ourselves?
That's what our Advise service is for. Our forward-deployed engineers work alongside your team at critical decision points - helping you choose architectures, pick models, and avoid the dead ends we've already seen. It's hands-on engineering guidance when you need it most.
Can you train our team on AI?
Yes. We run training for leadership teams (how to manage AI initiatives), developers (hands-on bootcamps where they build real solutions), and business teams (practical sessions on using ChatGPT, Claude, and Copilot in their actual work).
What if I want to own the IP?
That's standard. Most clients own the IP for what we build. Your AI systems are strategic assets - you should control them.
Are Alexandria and Minerva software products?
No. They're frameworks from our research programme that let us start your build further ahead. Think reusable components, not SAAS products you buy.
What are AI agents and why do they matter?
AI agents are systems that can reason about a goal, plan actions, use tools, and adapt based on results — without step-by-step human instruction. Unlike chatbots that respond to individual messages, agents pursue objectives across multiple steps. We've built production agent systems for BT, the National Cancer Institute, and iQ Student Accommodation. Our guide covers the four levels of agent autonomy and when to use each.
Why do AI agents need memory systems?
Without memory, every agent task starts from zero — no learning from past experience, no access to institutional knowledge. Vector databases handle similarity search but can't reason about relationships, time, or contradictions. Production memory needs layered architecture: working memory for the current task, long-term memory for experience, and structured access to existing knowledge. We built this for our Minerva platform.
What is sovereign AI and why should enterprises care?
Sovereign AI means running AI on your own infrastructure, under your control, with your data staying yours. Most cloud AI providers process your data on shared infrastructure in jurisdictions you don't control. For regulated industries — financial services, healthcare, government — this is a compliance risk. Sovereign AI eliminates it: your models, your hardware, your data never leaves your environment.