I banned Claude from saying 'looks reasonable' and it started finding my blind spots
I'm in the middle of launching a SaaS product, and the deliverable from one particular session wasn't code. It was an execution plan, plus the design of the quality gates. In other words: writing down "how we're going to build this" and "how we're going to inspect what we built", as documents a later session (the implementer) could pick up and run with. As part of the quality gates, I defined a…
I recently launched a SaaS product, and a valuable outcome from one specific session was not code, but rather an execution plan and quality gates. Within these gates, I defined a review sub-agent named "skeptic," whose job was to disprove the claims in the design document. This was a challenging task, as the skeptic was instructed not to approve or close with "looks reasonable overall."
Instead, it had to list the angles it attempted to refute, even if it ultimately found no issues. Despite my initial reluctance, I decided to put the skeptic to the test by feeding it the design document I had just created. To my surprise, it uncovered four significant issues within the document that I had just authored. One finding was particularly alarming, as I had contradicted a rule I had set for the project: block all outbound communication, yet I had inadvertently introduced a loophole by reusing an existing procedure.
Other findings included unclear language, unfinished work phases, and circular dependencies between phases. These issues would have been immediately apparent to anyone else reviewing the document, but I had failed to spot them in my own work. The skeptic's effectiveness stemmed from its unique success condition: finding flaws instead of confirming correctness.
By reversing the success criteria and preventing the "looks reasonable overall" escape clause, the skeptic was compelled to dig deeper and uncover hidden problems. This experience taught me an important lesson: when developing development documentation, it is crucial to have a reviewer who is not the author of the source of truth.
The person who wrote the document is often the least qualified to review it, as their subconscious mind may overwrite the actual content. In this case, I unleashed a reviewer whose sole purpose was to disprove my work, and the results were astounding. The skeptic proved to be an invaluable tool, revealing critical flaws in the design that I had previously overlooked.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
