I build the systems that do the marketing.
Most marketing advice ends at a recommendation. Mine ends at something running on a schedule that you did not have to remember to start.
I came up through marketing. What changed is that I can now build the thing instead of briefing it, which turns out to change what is worth recommending in the first place. When building costs a fraction of what it used to, the answer to "should we do this manually or automate it" moves a long way toward automate.
What I actually run
I own and operate three businesses, and I use all of them as the proving ground.
Wilow is a creative analytics tool for Meta advertisers. Around it I built a free public ad research tool, programmatic brand and vertical pages, a research index covering 91 UK brands and 5,460 ads, and an MCP server so AI assistants can query the data at the moment they answer a question.
The Keyword is a marketing and technology publication with 7,000 subscribers on a DR 63 domain. It has one writer and one assistant. Agents do the research, the drafting, the hero images, the publishing, the SEO and the internal linking, working from a knowledge base that keeps a consistent editorial point of view rather than producing generic summary.
Best Medellin is an English language property portal, currently pre-launch. It went from decision to a live, fully templated site with structured data inside a single day, built against the API rather than by hand. Agency supply is won through automated outreach.
Across the three, more than 30 scheduled agents run in production across three separate runtimes, reporting into one dashboard that rebuilds twice a day.
How I think about agents
Two objections come up every time, and both are fair.
The first is control. Every system I build runs against a written autonomy ladder that states exactly what an agent may do alone, what it must draft for a human to approve, and what it must never touch. Routine work runs unattended. Anything that spends money, sends to a customer, or changes a live page waits for a person. The leverage is real and the keys stay with you.
The second is decay. Automation usually gets worse over time, because the world moves and the script does not. So every system I build writes back what it learned: when a job hits an error, needs a clarification it should not have needed, or solves something the hard way, the fix goes into its own instructions before the job closes. There is also a check that compares what the documentation claims against what is actually scheduled, and reports the difference as a number that should read zero.
How I work
- Decision first, build second. Every case study on this site is something I decided to build, not a spec someone handed me. That is the part worth paying for.
- Build rather than buy, where building is cheaper. A lot of what I run replaces a SaaS subscription with a script and a free tier.
- Ship, then measure. "The page loads" is not a result. Impressions, rankings, replies and revenue are.
- Leave the documentation better than I found it. A system nobody can operate after I leave is not finished.
Outside the work
I am based in Bogota, Colombia, which means I overlap with both UK and US working hours. I write about marketing and AI most weeks, and a fair amount of what I build starts as something I wanted to exist and could not find.