Father, Tech-enthusiast,
Optimist, Mediator (INFP)

Every few months a new model arrives and someone announces that product owners, analysts, or quality people are finished. I have not seen that in the rooms I work in. I have seen something more interesting: the people who learn to cooperate with the model get their afternoons back.

Cooperation is a design problem. If you dump a pile of tickets into a chatbot, you get a pile of answers. If you give it a role, constraints, examples of a good work-product, and a human who still signs the result, you get a draft you can actually use.

In a regulated environment that last part is not optional. An AI can draft an SOP change, a test script, a stakeholder mail. A person still owns the decision. The future I want is not “the model ships quality.” It is “the model never ships without a person who can explain why.”

I build small agents for my own time, local tools, checklists, first drafts, because I want the four burners of life (work, family, health, recharge) to have fuel. That is not a manifesto against craft. It is a bet that craft is what remains when the copy-paste is gone.

When people hear “quality management system” they picture binders. Or a portal nobody likes. The actual product is trust: can an auditor, a regulator, or a colleague reconstruct what happened, who decided, and on what evidence?

That changes how you write a backlog. A pretty screen that hides the audit trail is a defect. A painful screen that captures the trail but makes people work around it is also a defect, because workarounds are where truth dies.

My job as product owner is to hold both ends. Integrations (identity, APIs, reporting) have to be boringly reliable. The workflow has to be short enough that people stay inside it. UAT is not a ceremony; it is where you find the lie between the process map and Tuesday afternoon.

I do not romanticise process. I do respect evidence. If we say we trained someone, the system should be able to show it. If we cannot, we do not have a quality system. We have a story.

Most failed AI drafts fail for the same reason most failed briefs fail: nobody said what “done” looks like. Tone, audience, constraints, source of truth, what must never be invented, all missing.

When I use a model for work I write the prompt as if I were briefing a new colleague. Role. Inputs. Output format. Examples. Forbidden moves (“do not invent system names”). Then I edit. The model is a very fast junior who never gets offended by red ink.

This is why I am not afraid of AI in product work. The scarce skill is still judgement: what to ask for, what to throw away, what to put in front of a stakeholder. The people who only generate will be noisy. The people who specify and edit will be useful.

If your prompt could not be handed to a person, it is not a professional prompt. It is a wish.

I enjoy working with people. Workshops, stand-ups, a messy Teams thread that suddenly becomes a decision, that is real work. I am also comfortable closing the door. Some problems need a quiet stretch, not another meeting.

Automation, for me, is not a personality. It is how I stop repeating myself so I can show up for the parts that need a human. A local agent that files things, a first draft of a status, a checklist that does not live in my head, that is time I can spend with family or on the next hard question.

I am always looking to improve, personally and professionally. That sounds like a poster until you treat it as a backlog: one habit, one skill, one system at a time. Karate with my son. A language. A better way to run UAT. Same muscle.

If we meet, I would rather do it in person when we can. The rest of the time I will be on the call, or in the quiet, doing the work.