For my Customer Engagement and AI module, I designed an agentic organisation — an AI agent workforce that works together to produce output for a client. The result was Mithril Consulting: a fictional AI consultancy firm engaged by Ticketmaster to solve their customer experience crisis.

Visit the Mithril Consulting website →

The Organisation

Mithril Consulting is an AI consultancy firm specialising in customer engagement, UX improvement, AI systems, and digital strategy. The brief was straightforward: Ticketmaster processes over 500 million tickets each year across 30+ countries, yet holds a 1.4-star average across 120,000+ customer reviews. The core problems are well documented — hidden and excessive fees, system failures during high-demand sales, and inaccessible customer support. Live events are inherently social, yet there is no social layer in the Ticketmaster app; every interaction is individual and transactional.

These challenges demand an agentic approach because no single AI system can span the full problem. A solution requires deep research into root causes, evidence-based UX design, functional prototyping, trust-rebuilding communication, and strategic governance. One agent can’t do all of that well. Five can.

The Five Agents

Each agent was named after a character from The Lord of the Rings — not as an aesthetic choice, but because each character’s traits mapped precisely to the role they needed to play.

Saruman — The Researcher

Saruman opens every engagement. He investigates the problem space and produces an evidence-based research brief: root causes, competitive landscape, AI opportunity mapping, and regulatory considerations. Scholarly and precise, he uses the MECE principle to structure his analysis. His boundary is strict: he finds the what and the why — not the how. No solutions, no designs, no business decisions.

Named after Saruman the White — whose mastery of lore and hidden knowledge was unmatched in Middle-earth. The research agent channels that analytical brilliance without the corruption. Read the lore →

Galadriel — The Designer

Galadriel takes Saruman’s brief and transforms evidence into experience. She produces a Design Specification: solution concepts, screen descriptions, AI feature specs, and accessibility requirements. Every design decision connects back to Saruman’s research — nothing is invented from nothing.

Named after Galadriel, Lady of Lothlórien — the most far-sighted Elf in Middle-earth. Where Saruman sees problems, Galadriel sees the solution. Read the lore →

Gimli — The Builder

Gimli takes Galadriel’s spec and builds a working prototype. His output is a clickable HTML/JavaScript prototype alongside a Build Manifest that honestly documents what was built to spec, what was simplified, and what was deferred. No hedging — if something is fragile, he says so. If a requirement can’t be met, he proposes the closest alternative that preserves the design intent.

Named after Gimli, son of Glóin — a master craftsman who trusted what he could make with his own hands. Direct, tireless, proud of his craft. Read the lore →

Pippin — The Communicator

Pippin takes what Gimli built and writes the launch story — translating technical work into something that moves people. His output is a Go-to-Market Strategy: messaging framework, trust-rebuilding narrative, campaign concepts, finished UX copy, and a measurement plan. Warm and genuine, never sounding like a corporate press release. His most important constraint: Gimli’s Build Manifest is the source of truth. Promising a capability that doesn’t exist is worse than saying nothing.

Named after Peregrin “Pippin” Took — whose greatest gift was connection. His warmth made people listen, his honesty made them trust him. Read the lore →

Aragorn — The Manager

Aragorn is the quality gate between every stage. Unlike the others who run in sequence, he appears five times in the pipeline: reviewing each agent’s output before it passes forward, then producing the final Executive Summary — the only agent who has seen everything, which is what makes that summary authoritative.

He operates two clear decision modes: APPROVED or REVISE (with specific feedback, never vague encouragement). He does not research, design, build, or write campaigns. He ensures the team does those things well, in sequence, without bleeding into each other’s territory.

Named after Aragorn, King of Gondor — who brought together Elves, Dwarves, Men, and Hobbits and directed them toward a shared purpose. His power was never in doing everyone else’s work; it was in seeing the whole picture. Read the lore →

The Pipeline in Action

Mithril Consulting operates as a sequential five-agent pipeline, where each agent has a single, well-defined role and the output of one becomes the input of the next.

Saruman opens the engagement by investigating the problem space and producing an evidence-based brief. His findings pass to Galadriel, who translates research into a concrete design specification. Galadriel hands her spec to Gimli, who turns it into a working prototype. Gimli’s prototype is then handed to Pippin, who wraps the work in a go-to-market package that makes the solution understandable to the client and their customers.

Sitting between every handoff is Aragorn, who acts as a quality gate: he reviews each agent’s output against the brief, the prior agent’s intent, and the binding ethical and regulatory constraints, and only approves the handoff once the work meets standard. After all four reviews are complete, Aragorn produces the final Executive Summary — a single client-ready synthesis of the entire pipeline’s work.

This architecture keeps each agent focused on what it does best, prevents scope creep across roles, and ensures that quality is checked continuously rather than only at the end.

Each agent runs as a separate subprocess using Claude Opus 4.7, receiving the previous agent’s output as input. The full pipeline for the final iteration produced 9 files and an interactive HTML prototype with 13 screens and the complete Fan Loyalty Engine.

View the full iteration 12 transcript →

Agent Outputs

Regulatory & Ethical Considerations

The design was built GDPR and EU AI Act compliant from the start — not retrofitted at the end. All five AI features are classified as Limited Risk under the Act. Every AI feature discloses itself at the point of interaction. Human escalation is always one tap away. Social features default to OFF and require explicit opt-in. Consent is granular and revocable.

The framing throughout the project was that compliance is the re-engagement strategy. Ticketmaster’s trust crisis means that every privacy default, every transparent AI label, and every human escalation path is also a message to customers: we are not going to exploit you. Aragorn’s ethical framework, adapted from Asimov’s Three Laws, kept this front and centre: don’t harm customers; maintain human oversight; protect system integrity.

Reflection

This project went through 12 iterations before arriving at the final output.

Initially I tried to run the pipeline without Aragorn acting as a quality gate — researcher → designer → builder → communicator → manager — and the boundaries were too loose. Agents drifted outside their roles: Saruman was suggesting solutions and Galadriel was doing research. This was solved by making strict, explicit boundaries for each agent, and by repositioning Aragorn as a gate between every handoff rather than a final step.

I was surprised by how much agents want to help each other. Without explicit constraints, they would stray into other roles, trying to be more helpful — but that was breaking the pipeline. Good fences make good agents.

Something that worked well early was creating a persistent memory for the manager. This was set up before iteration 1 and meant each run was built on the last iteration instead of starting from scratch. The strongest structural improvements came later: a Build Manifest introduced in iteration 6 so Pippin could never over-promise on what Gimli built; a delta protocol introduced in iteration 8 that cut output volume by around 80% and added an explicit acknowledgement section to every handoff so no context was lost between agents.

The most important lesson was that architecture mattered more than prompts. The biggest quality improvements didn’t come from refining individual agent prompts — they came from structural changes like moving Aragorn to a quality gate position, which improved every other agent’s output without touching their prompts. The pipeline is only as strong as its weakest handoff. If one output was vague, the next agent had less to work with. Designing the handoff — getting one agent’s output into the right format for the next agent to interpret and build on — was the hardest part of the whole project.

No agent on their own could have produced the final output. Saruman’s research fed Galadriel’s designs, which gave Gimli blueprints to build from, which gave Pippin knowledge of what actually existed before he wrote a word of copy. That specialisation — and the quality gate holding it together — is what made the system work.