The only AI advice I actually give people

People ask me how to get good at this, expecting a prompt library. There isn't one. Here is the entire method.

Use Claude. Opus 5, Medium or High. Then do every task with it for a week or two, including the ones you could finish faster yourself today.

That last part is where everyone bails. The first week is slower. You are paying tuition. Pay it anyway.

The rule that makes it work: never route around a blocker. This is what separates people who get 10% better from people who get a different job. When it hits a wall, your instinct is to go do that part by hand. Don't. Ask it to overcome the problem instead.

Can't reach the data? Connect it to the data. No integration exists for the tool you need? Have it build an MCP server. Have it stand up a small local server if that's what the job requires. Blocked on permissions? Have it tell you exactly what to unblock and who to ask. Every one of those is a five-minute conversation that permanently removes a category of work from your life. Doing the task by hand removes exactly one instance of it.

Around week two you look up and most of the work you thought was hard is running without you. The gains compound because each unblocked thing makes the next thing possible.

What that actually unlocked for me

The obvious stuff first: researching Jira tickets, digging through years of backlog nobody has read since it was filed, and reconstructing the history of a decision from scattered threads.

Then the stuff I didn't expect. Pulling customer sentiment out of recorded calls where customers vent to their CSM, which is the most honest product feedback that exists and the least read. Synthesizing thousands of support tickets into an actual ranked picture of where a feature hurts, with the volume attached. Work that used to be a quarter-long research project and therefore never happened.

Defensive product management

That is all defensive work, and I mean it as a compliment. Defensive product management is understanding what already exists: the history, the customer sentiment, the accumulated damage. It's unglamorous. It's also where most of the value in a mature product is hiding, because nobody has time to read 4,000 tickets and work out which 40 matter.

Once you can synthesize that fast, the job changes. You stop guessing at priorities and start knowing them. You look at a list of complaints and see the three root causes underneath. That is the "make our existing things better" half of the job, and AI is absurdly good at it.

Offensive product management

The other half is green field. Blue skies. Net new, genuinely fun experimentation.

Offensive work is not about certainty, it's about progress and momentum. You find the highest-leverage thing, solve it, show it, and repeat. Do that enough times and the product becomes undeniable. It becomes obvious that it's on the way to making a difference. Then something great happens: everyone wants in, including the best people. Momentum recruits better than any pitch does.

The reason I care so much about the defensive half is that it buys the time for the offensive half. If you're drowning in "why is this broken," you never get to "what if."

What this costs you

A week or two of feeling slower, and the ego hit of watching something else do work you were proud of being good at. That's the whole price. I'd pay it again.