The short version
I work in search and growth marketing, and I build AI workflows. Those two halves are the same job: find the thing that repeats, understand why it repeats, and either automate it or teach someone to do it well.
The workflows come out of client work. Not demonstrations, not prompts that impress in a screenshot and fall over on real data — the automations I actually deploy, documented well enough that somebody else can run them. The courses exist because clients kept asking for the version where their own team could build the next one without me.
Search is still the foundation. AI answers changed what visibility means — ranking and being cited are now separate outcomes that move independently — and most reporting has not caught up. A good part of my work is rebuilding measurement so it reflects what is actually happening.
How I think about automation
The interesting question is not whether a model can do a task. It usually can. The interesting question is whether the output can be trusted, and that is an engineering question, not a prompting one.
So every workflow is built the same way: an input that gets validated, a step that is allowed to fail loudly rather than quietly, and an output a human can check cheaply. The failure modes get documented alongside the instructions, because the failures are where the value is — a workflow that is silently wrong is worse than no workflow at all.
What I won't do
I won't generate content at scale. It is penalised, it damages the sites it is used on, and it makes the internet worse. I will automate the brief, the analysis and the monitoring; the writing stays human.
I won't use fake urgency to sell. No countdown timers, no seats that magically reappear. If something is limited, the limit is real and I will tell you what it is.
And I won't take an engagement that does not fit. A declined project costs us both less than a wrong one, and saying so early is part of the job.