Point of view

AI isn't a product. It's a lever.

The point isn't using it. It's knowing where it creates value, where it destroys it, and how to turn that into a feature that holds up.

Book a call Back to portfolio

The conviction

Opening ChatGPT no longer sets anyone apart. What sets you apart is judgement: deciding where AI belongs in a product, and where it doesn't.

I never approach AI through the technology. I start from the product problem, the business value and the cost of getting it wrong. AI is only a good answer when the problem calls for it. Otherwise a simple rule, better UX or an API integration does it better, faster, cheaper.

Discernment

Where AI creates value

How I decide. Two areas, each with a single trigger. AI comes in only when that trigger is met.

05cases

In the product

Capabilities that change the user experience.

Assistants & copilotsWhen the user faces too much information or a repeated expert task.
Semantic search & knowledgeWhen the data exists but no one can find it across a fragmented IT landscape.
Document intelligenceWhen humans re-key or sort documents by hand.
Decision support & recommendationsWhen a recurring decision rests on many implicit signals.
Natural language & smart formsWhen the complexity of a flow drives users away.

04cases

In delivery & operations

Speed and quality gains across the delivery process.

Workflow automationWhen a chain of tasks is repetitive, high-volume, low-judgement.
Monitoring & anomaly detectionWhen signals are too numerous for the human eye.
Documentation & specsWhen quality depends on individual discipline. AI sets a floor.
Faster discoveryInterview synthesis, verbatim analysis, market watch. It accelerates, it doesn't decide.

And where I say no

Where a mistake is costly and irreversible, where compliance demands traceability, where volume doesn't justify the complexity: a deterministic system often beats a probabilistic model.

Step 01
Frame the problem, not the tech
What job to be done, what business value, what cost of error. The tech comes after.
Step 02
Challenge the use case
Is it really the best answer, or just the trend? Most AI ideas don't pass this step.
Step 03
Pick the approach and the model
By context: cost, latency, privacy, criticality. The biggest model is almost never the default answer.
Step 04
Specify the guardrails
Fallback, quality measurement, feedback loop, controlled cost. An AI feature without guardrails isn't ready.
Step 05
Bring engineering in
Feasibility, data, security, GDPR. I speak their language, nothing gets lost in translation.
Step 06
Drive adoption
Without adoption, the best AI feature is worthless. That's where the ROI is won.

Principles

My stance

An accelerator, not a substitute for product teams.

The problem first, the tool second.

Judgement and accountability stay human.

Privacy and compliance first, especially in enterprise.

What I already do

I don't theorise about AI, I use it every day in my own product process: structuring specs, synthesising discovery, analysing data, prototyping. Tools change fast; the reflex stays the same: hand the machine what doesn't need my judgement, so I can focus it where it counts.

My daily tools