At Merluno, a Belgian outreach company, call reports were written by hand after every call. I designed Merlin, an AI agent that writes them instead, cutting the time the team spent on reporting.
Scope
AI agent
Internal tool
B2B outreach
My role
AI service design
Autumation
Tools
Airtable
n8n
Figma
Soniox
Timeline
Q1&Q2 2025
01- Context
As the service designer on this project, I...
Designed the flow
I figured out how a call should turn into a report, and what the agent needed to know to write it well.
Built the agent
Started with n8n. Later added RAG and a multi-agent setup to make the reports more reliable.
Tested it with the real user
Sat with the person using it every day, watched what confused the agent, and fixed prompts based on that.
02- The problem
After every call, someone had to write a report by hand. This took time away from actual outreach work.
03- Solution overview
An AI agent that saves the team time
Built on n8n, Merlin turns each call into a report right away. That gave the team more time for outreach, not writing.
04- Building Merlin
I believe that
when AI is part of the build, design and testing stop being separate steps. They start happening together, in one loop: build a little, test it, learn, then build again. That loop moves fast, and you go through it far more times than in a normal product process.
Round 1
First prototype
A one-week prototype showed us exactly what the system needed to do.
We built a basic reporting agent in one week, using n8n and a simple interface.
We tested it with one person, using a small set of real call data.
Security wasn't strong enough, and it couldn't scale. But it showed us exactly what to fix next.
Round 2
Making it reliable
Testing with the real user, every day, became our research.
We added RAG for more context, and split the work across multiple agents to write fuller reports.
We sat with the person using it every day, and watched what confused the agent.
The reports got more reliable, with much less back and forth.
Round 3
One unified system
Everything came together into one connected system.
We moved everything into one Airtable-based CRM, with Merlin built directly into it.
Result:
This removed steps that used to live on paper or across different tools.
07- Impact
05- insight
Choosing tools that could change overnight
Most of the tools I used were brand new and still changing fast. I had to design around the workflow, not around one tool's quirks, so it would keep working when the tools underneath changed.
Building a real system without a dev team
No developers, no big budget. Just me, a few no-code tools, and a lot of testing. It showed me you don't need an engineering team to build something that actually works.
Learning where AI actually helps
AI demos look impressive. Real workflows are messier. This project taught me exactly where AI made things faster, and where it just added confusion.
06- Learnings
GDPR and the AI Act mattered more than I expected
Every piece of data Merlin touched had to hold up under GDPR and the coming AI Act. Checking for that became part of every decision, not an afterthought.
Don't get attached to a tool
AI tools change fast. What we picked could be outdated in months. Staying loose about tools let us swap things out without losing time.
