About
Peter Jonker builds AI and analytics systems for Semrush, the kind you can trust with real budgets.
In today’s world you can build almost anything with AI, like the 3D melon below. While it might look impressive, it is practically useless. Just like a lot of stuff people are building with AI right now.
The articles here aim to be the opposite: they show, step by step, how to build reliable systems and ways of working that actually make an impact. Hopefully they help you more than this melon.
Applied AI
Digital marketing is going agentic. The practitioners who treat that as an engineering problem will build what a manual workflow cannot keep up with. The work is closer to machine learning engineering than to traditional marketing: eval sets, error analysis, feedback loops, version-controlled prompts, and tests that pass before anything touches a live account. Deploying agents adds its own work: what each agent remembers between runs, which tools, accounts and data it is allowed to use, and a record of every step it takes.
Marketing engineering
As a specialist, you can become a marketing engineer. You already know what a good outcome looks like and how to tell whether the data behind it is right. That is what lets you steer a system and validate it end to end yourself.
You build marketing tools, not software products. A coding agent such as Claude Code writes most of the code. What you add is systems thinking and a working grasp of the technical concepts: how data flows, how to test, how to publish, and how to keep it secure. The articles explain these as they go.
As models take on more of the daily work, more of the tools that do it get built by the people who know the domain.
The future of work
Marketers now build with coding agents, new AI tools keep appearing, and the platforms they already use are opening their APIs and MCP servers to agents. More of the job is now within an agent’s reach.
There is no playbook for this yet. Teams are trying very different things. Over the next three to ten years this will change how marketing teams work: who builds, who reviews, and how a team keeps its knowledge so that people and agents can work from it.
Topics
This site covers three areas.
Performance marketing: how to run and automate it, mostly paid search in Google Ads, with paid video on YouTube as well. That means search term mining and negative keywords, keyword research, and the data engineering underneath: pulling data from ad platform APIs with custom queries and pushing changes back to a live account safely. Bidding logic too: how clicks and spend move when you change a tROAS target, and value-based bidding on predicted lifetime value.
Marketing analytics: the data work in BigQuery. Pipelines, tables documented so people and agents read them the same way, reporting that accounts for conversion lag, and alerts when spend or conversions drift. Predictive analytics, such as budget forecasts by channel and checking predicted lifetime value against what customers actually paid. Embeddings too, for building topic trees, comparing targeting and reporting on search demand at a scale nobody can read by hand. And analytical agents, the kind that answer performance questions in plain English.
Applied AI engineering: the engineering practice behind AI in production. Context engineering, testing and evals, specs and code review, publishing and running what you build, and the coding tools that speed it all up, such as Claude Code. Building agents and LLM pipelines where a person approves the output. And the day-to-day: how to structure your work around AI, how a team keeps the knowledge its agents read, and what a new model or tool release changes for you.