//About
Twenty-plus years turning data into decisions. Today, agents in production.
I am André Silva. I work in process automation and agentic AI at Wise Pirates, in Porto, Portugal. This lab is where I document, with evidence, what I learn putting agents to work in real companies.
Today: Wise Pirates
PortoAt Wise Pirates, a data-driven digital marketing agency (since 2017, 100+ people and 500+ brands served), I lead the Process Automation & Agentic AI unit and sit on the operational leadership with the technology and infrastructure axis.
The unit's job is to make intelligence run instead of sitting in a report: business automation workflows, specialised agents with RAG, memory and supervision, data pipelines with observability, and the security layer that lets them into production. That is where I apply what I write here, and where you can find me to work together.
Work and proposals through Wise Pirates. Technical conversations on LinkedIn.
Path
four eras, one direction- //01
Campaigns and data
I started in Google and Meta Ads campaigns, then analytics, BI and forecasting. That is where I learned to read a business through its numbers, and to distrust dashboards nobody opens.
- //02
Performance marketing
Multi-market accounts, rule-based budgets, attribution that had to reconcile with the client's BI. The more campaigns I optimised, the more manual processes I found around them: reporting that took hours, data copied between systems, decisions by gut feeling.
- //03
Process automation
So I automated all of it. n8n workflows in production, multi-system API integrations, data pipelines with logs, metrics and alerts from day one. Idempotency, retries and runbooks stopped being engineering vocabulary and became what separates what holds from what falls.
- //04
Agentic AI in production
Today I design and operate AI agents that do real work: with tools, memory, supervision and explicit limits. I built a knowledge engine (Prism) that turns every project into reusable patterns, MCP servers for the platforms I work with, and the pipeline that produces this lab. The question stopped being "does it work?" and became "is it safe, observable, and does it prove value?".
What I build
- Agents with brakes
- Tool whitelists, output validation before irreversible actions, kill switch, per-agent access policies.
- Observability by default
- Structured logs, a trace per run, metrics and alerts that wake up before the client complains.
- Automation that proves value
- Every automation must show euros saved or hours recovered. Less theatre, more production.
- Productisation
- Turning what works for one client into a repeatable product: knowledge engine, MCP servers, reporting pipelines, quoting agents.
This site: Source code · Lab areas · RSS