Essays, playbooks, and measured findings on building with AI, running agencies, and how work actually gets done - plus personal stories and the opinions I will defend. Much of it published here first, and all of it open.
In 2015 I was running the exact version of the agency everyone now predicts the death of. Thirty people, a pyramid, margin living in the gap between what a junior cost and what a junior billed. Predictions about the industry are usually written so they can never be wrong. These five are written the other way: dated, specific, and made to be graded in 2029. Hold me to them.
The audit tells you what is broken. This is the file that fixes it: a stack-agnostic instruction you hand to an AI coding agent or a developer, covering rendering, head tags, previews, schema, machine files, and the content structures that get quoted rather than just read. Published whole, no email gate.
Stuck agencies almost always reach the same conclusion: they need more leads. The symptoms point that way. Calls feel harder, proposals stall, win rates drift. But pipeline symptoms can come from four completely different places, and most of them have nothing to do with lead volume. More leads do not fix a broken offer, a differentiation problem, or a founder who is still the system. The diagnostic step is the one most agency advice skips.
My agency had thirty people, and every chair had a person in it. My current company has one person and more chairs than the agency ever did. AI fills the hands of every seat: it writes, codes, forecasts, drafts. What it does not hold is the head of the chair, the question each seat exists to ask. That part still has to live somewhere, and there is only one somewhere left.
Marketers credit AI with raising the quality of their output, and in the same breath name AI slop as their biggest fear. Both instincts are right, because they describe the same tool with and without a quality system around it. This is the quality system: how the four failure modes differ, the five checks that catch them, and why review stopped being the tax on production and became the product.
Every senior I ever hired was somebody's junior first. So was I. The junior desk always did two jobs at once: cheap production, and the slow apprenticeship that produced the next generation of seniors. AI is removing the first job fast, and the industry is quietly acting as if the second one never existed. It did. I am the proof, and so is everyone I ever promoted.
Every LinkedIn analytics export is capped at fifty posts. Ask for a year and you quietly get only your best fifty, which inflates every average you compute. Ask month by month and you get everything. That is step one of eight, and the rest is joining data LinkedIn keeps in separate places so it can answer questions the dashboard cannot.
In 2021 I deleted my LinkedIn account and moved to Mexico. I came back in January 2026 with one connection. Seven months and 135 posts later I have 139 followers, one post that reached 34,708 people, and enough data to answer a question I could not answer before: what actually builds an audience from nothing, and what only looks like it does.
A readable site is not a found site. After the technical work is done, there is a short list of things that happen on Google's and Bing's own consoles, not in your codebase - and one of them needs no account, no API key, and takes about twenty minutes to wire in permanently. Here is the whole list, and what to actually read once it is running.
There was a wall in our Amsterdam office covered in framed certificates, and a shelf with the serious hardware: a Cannes Lion, a Webby, a D&AD Pencil. When I mention them to clients today they are always positively surprised. That surprise is worth examining, because it means the awards impress everyone and brought us almost no one.
I write regularly on LinkedIn about what I'm building and learning: agency growth, AI development, product judgment, and the messy reality behind making things work.
Follow on LinkedIn