Paulo de Vries · Senior Product Designer × CRO · builds AI · Amsterdam
I design products that convert, prove it with experiments, and build the AI inside them.
Ten years making brands convert — Telegraaf, NRC, dentsu. Then I taught myself to build the AI, and got curious about a newer question: how to get cited by ChatGPT, not just ranked on Google.
Experience: Mediahuis (Telegraaf, NRC, Dumpert) · dentsu · Contentsquare-certified · Design Academy Eindhoven.
Two crafts, one operator
Most people do one of these. The whole value is in the overlap — design the experience, prove it converts, then build the AI behind it.
Conversion & experimentation. A decade of A/B tests and conversion programs at Mediahuis and dentsu, Contentsquare-certified. I analyze the funnel end to end — where's the friction? — and prove what moves people with experiments, not opinions. I built our A/B-test builder and the group's Experimentation Hub, where every test plan and analysis lives. The craft I've been paid for since 2016.
AI systems I build. AcePilot, a self-improving multi-agent OS I wrote, runs unattended behind a guardrails layer it can't prompt its way past. I build the AI itself, not a wrapper around it.
AI visibility (GEO). I get pages cited by ChatGPT and Perplexity, down at the level of crawler logs and citations, and have been at it solo for two years. I've gotten a site's visitors to double, from a small base, from AI answers, not Google.
How I build
A build-measure-learn loop run with a CRO's discipline: every change is a hypothesis, not an opinion — shipped behind guardrails, then proven two ways before it's kept.
The discipline is the old one: never trust what you can't measure. What's new is running it with agents — at speed, at scale.
End to end, one person
The loop runs fast because I don't hand off the middle. I write the hypothesis, design it, ship the tracking, run the test, read the data, and work in the dev team's repo instead of throwing tickets over the wall.
Product & experimentation — Jira · A/B testing · Contentsquare (certified). Design — Figma. Measurement — Google Tag Manager · GA4 · Search Console; I implement my own tracking, so an experiment never waits on a developer. Build & ship — GitLab, and AI I write myself: Claude / Claude Code, MCP, autonomous agents.
Selected work
-
AcePilot — a self-improving multi-agent OS
Built solo: a safety-invariant layer, an evolution engine, task-ranking oracles, cross-session coordination, crash-recovery — running unattended in production. The hard part was never the model; it's everything around it that decides whether an agent quietly ships or quietly sets something on fire. Two years running it unattended →
-
A task queue built for AI agents — MCP server, REST API, multi-agent orchestration, lease-based concurrency so parallel agents don't collide. I dogfood it daily in production; live at taskprio.com.
-
The runtime-safety and test-time layer I pulled out of AcePilot: bound an agent's scope, cap its cost, halt destructive actions before they run — then assert in CI what it actually did. Zero dependencies. The thinking behind both →
-
700+ bookings a month, one local business
I took a local fitness studio to 700+ bookings a month with Instagram and conversion work — the same CRO craft, end to end.
What I'm betting on
Building got cheap; attention got expensive. The brands that win will convert the visitor — and get cited by the AI that sent them.— my operating thesis
Search is splitting in two: Google's blue links, and the AI answers quietly replacing them. I've spent ten years on conversion and two on the AI. The overlap is the whole game, and very few people work in both.
The short version
Designer by training (Design Academy Eindhoven), but I always cared less how things looked than whether they worked. Ten years made me a senior conversion specialist at Mediahuis and dentsu, proving it with experiments, not opinions. When AI arrived I didn't optimise around it. I taught myself to build it: an agent OS (AcePilot) and an MCP product. Now both crafts are one job, part senior CRO and part design engineer: design the experience, prove it converts, build the AI behind it.
Writing
-
Development speed is a friction problem, not an effort problem
Speed isn't typing faster — it's throughput of validated changes. The rate-limiter is the fog around building, not the building. The Shortest Path to Evidence: a five-move structure for turning rough research into shipped, measured results fast.
-
AI-referred visitors are the most pre-qualified traffic you'll get — and most sites waste them
Getting cited by ChatGPT is only half the job. The visitor the AI sends arrives mid-decision, pre-qualified by the model — and most sites drop it on a homepage that restarts the pitch. How a decade of CRO converts it instead.
-
What ten years of CRO taught me about building reliable AI agents
The conversion brain turns out to be the agent brain: never trust what you can't measure, and build the instrument that measures it. Why reliable agents are an experimentation problem, not a model problem.
-
How I got a site cited by ChatGPT — and why that's the new SEO
The GEO playbook I've refined in production for getting cited by AI answer engines, now that most searches end without a click.
-
Agent reliability is a guardrails problem, not a model problem
After two years running agents unattended in production: they fail on the operating layer, not the model. The layer that makes them ship.
Questions
- Who is Paulo de Vries?
- A senior conversion-rate optimization (CRO) specialist and product designer in Amsterdam, with ten years at Mediahuis (Telegraaf, NRC, Dumpert) and dentsu, Contentsquare-certified, Design Academy Eindhoven. He taught himself to build AI, and created AcePilot, a self-improving multi-agent system. The edge is the overlap: I design products that convert, prove it with experiments, and build the AI inside them.
- What do you actually do?
- Two crafts most people keep separate: senior CRO + product/UX design, and building AI systems. Designing experiences, proving they convert through experimentation, and building the AI behind them — including getting pages cited by AI answer engines.
- Can you own a product end to end?
- Yes, it's how I work. I write the hypothesis, design the solution, implement the tracking, run the experiment, read the data, and work in the dev team's repo, instead of handing off tickets. Discovery, roadmap, experimentation and stakeholder reporting are the work I've done for a decade, under CRO and design titles rather than "product owner".
- What's your experience?
- Senior CRO Specialist / UX Designer at Mediahuis (2023–present), UX/CRO consultant at dentsu (2022), Product Designer & Marketer at JobBoost (2017–2021). Contentsquare-certified. BA, Design Academy Eindhoven.
- How can I reach you?
- Email [email protected] or LinkedIn. I work at the intersection of AI and conversion.