How I work · AI-assisted systems · clear ownership
I use AI to move faster without outsourcing the important decisions.
A working method for framing decisions, testing assumptions, and making ownership visible.
I start with the decision, not the tool. I map the system around it, expose assumptions that could break it, test the result, and adapt the scope when reality disagrees. AI handles repeatable implementation; I decide what evidence is enough and what is ready to release.
Translate the business question before touching the tool.
A request such as “find the best campaigns” or “automate the search” sounds clear until someone has to decide what “best” or “automate” means. I start there, before touching a tool.
The reusable six-question loop has its own page.
This page stays about judgment: how I frame the work, where AI fits, and what I own. How I check the work carries the shared questions, controls, incidents, and readiness stages.
Think strategically. Build systemically. Test the story. Adapt on evidence.
These are not abstract values added after the work. They are the habits that show up across the portfolio, whether the output is a marketing measurement system, a paper-trading testing process, a game, or a content pipeline.
AI helps me build faster. It does not decide what is true.
I am open about being heavily AI-assisted in implementation. That makes the review discipline more important, not less. A fast wrong answer is still wrong, and a polished interface can make a weak assumption harder to notice.
I own the question. I decide what the work is for and which claims are worth publishing.
I own the boundary. I decide what the system may collect, change, send, spend, or publish.
I own the checks. I ask for tests, hostile examples, fresh-source checks, and a readable record of what happened.
I own the final explanation. If a reader needs a glossary to understand the headline, the headline is not finished.
What owning the final explanation looked like when the prediction was wrong.
In the marketing measurement study, I wrote down a prediction before the analysis ran: one week of data would not be reliable enough to agree with the final answer. That was the rule before the result existed.
The result said the opposite. One week agreed with the final answer most of the time, but not always. I kept the original prediction, labeled it partly contradicted, and kept it in the record rather than quietly rewriting or removing it. The result is shown here while the source project remains private. See the measured result and method →
Automation should make attention cheaper, not judgment disappear.
Every project on this site has a line it will not cross without me. Four examples, in the words that actually apply.
A good fit? Let’s talk.
These are the habits I bring to every project on this site, not just the ones with a public write-up.