AI Experience

AI changed the way product management works for me. Not by making the same job faster, but by collapsing the distance between seeing a problem and shipping a solution. The old cycle was define, spec, hand off, wait. Now it's define and build, often in the same day.

That speed changed which problems became worth solving. Internal tooling, operational visibility, developer experience: work that would never survive a prioritization exercise against revenue features became viable once one person could ship it. Several projects went from concept to production-deployed in the same week they were conceived.

The same shift showed up on the customer and business side, just with higher stakes. When I could prototype against real data instead of a slide, or fully integrate our product into a third-party app in days instead of a quarter, conversations with stakeholders stopped being about whether an idea was plausible and started being about whether it was right. Feedback loops that used to take a sprint took an afternoon.

Domain Knowledge + AI = Dangerous Duo

AI compresses the coding. A five-vendor billing reconciliation pipeline took five hours to build, start to finish. What AI couldn't do was know the underlying data well enough to get the requirements right — that the internal system stores fees as negative values, or that one vendor bundles two fee types into a single line item. Miss either one and the tool ships broken. That's where I spend my energy now: not writing the code, but making sure AI is building the right thing.

Technical Understanding to Get It Done

An example of leveraging my technical acumen was this project: document extraction accuracy pipeline. It chained together Databricks SQL, OAuth2 authentication with automatic token refresh, a Playwright browser-automation fallback for when the API path was blocked, and eighteen type-specific comparison strategies — five distinct technical problems, unified into a single pipeline that runs unattended and turns eight days of manual validation into a fifteen-minute run.

Sharing is Caring

As I build my own tools and methods, ensuring results can be replicated is important for validation. My recurring pattern is focused on encoding this knowledge into a skill anyone can run. natural-language event testing skill cut analytics verification from ten to one minute, and is available for the team to install into their Claude Code instance.

Tools Adopted

#claude-code #claude #vsc #cursor #gemini #vertex-ai #m365 #datrabricks #amplitude #segment #extend.ai #railway #vercel #docker #xcode