My job isn’t just to generate ideas, it’s to de-risk them.

In my recent planning work, I used two specific techniques to catch failure points before we wrote a single line of code:

The “Single Source of Truth” Challenge:
My stakeholder wanted a “single source of truth.” It sounded good. Instead of just accepting it, I asked the AI to challenge the framing.
The Discovery: The AI pointed out that for our specific offline-first mobile requirement, a rigid “single source” would break user experience. We reframed the entire problem because I invited the AI to push back.

Checkpoints, Not Just Tasks:
I don’t create a plan that is just a list of “To-Dos.” Every phase must have a Verification Step.
Bad Plan: “Step 3: Update the database schema.”
My Plan: “Step 3: Update schema AND Execute the VerifySchema skill. If this tool reports an error, do not proceed to Step 4.”
The Insight: Errors compound. If you let an AI (or a junior dev) stack 10 tasks on top of a shaky foundation, you don’t have a project, you have a tower of cards.

Originally posted on LinkedIn