Blog

Thinking in public

WEB DEVELOPMENT · 9 MIN

The database migration that should have been two migrations

Combining a schema change with a data transformation in one migration is a common shortcut that makes rollback nearly impossible.

AI CONSULTANCY · 8 MIN

Why I turn down AI projects with unclear success metrics

A project without an agreed definition of success can't fail, and it can't really succeed either. Both are bad for everyone involved.

SEO · 9 MIN

Why duplicate content penalties are mostly a myth (and what's real instead)

There's no actual duplicate content penalty in the way most people describe it. There is a real, related problem worth understanding instead.

AI AUTOMATION · 7 MIN

Why AI automation projects need a rollback plan

Every automation launch plan covers what happens if it works. Almost none cover what happens if it needs to be turned off fast.

MEDIA BUYING · 9 MIN

Why seasonal campaigns need a different measurement window

Judging a holiday campaign by the same weekly cadence used for evergreen campaigns routinely produces the wrong conclusion.

WEB DEVELOPMENT · 9 MIN

Why environment variables leak, and how to actually stop it

A surprising number of API key leaks trace back to the same handful of preventable mistakes, repeated across otherwise careful teams.

AI CONSULTANCY · 9 MIN

The consultant's dilemma: recommending the boring tool over the exciting one

The newest, most impressive tool in a category is rarely the right recommendation, and saying so is a harder sell than it should be.

SEO · 9 MIN

The XML sitemap mistakes that quietly waste crawl budget

A sitemap padded with low-value or broken URLs doesn't just fail to help, it actively directs a search engine's limited attention away from pages that matter.

AI AUTOMATION · 6 MIN

The difference between an automation and a script that nobody wants to admit

Calling something an automation implies a level of robustness a lot of "automations" never actually earned.

MEDIA BUYING · 7 MIN

Attribution windows: the setting nobody double-checks after setup

A default attribution window set once at account creation quietly shapes every performance number after it, and almost nobody revisits it.

Next step

Prefer the applied version?

A 30-minute discovery call. You bring the numbers, I'll bring the questions. If I'm not the right fit, I'll tell you who is.

Book a discovery call See the work