AI at work should cover effort, not capability, argues Zenn essay

The essay is the first in a three-part series aimed at engineers, and it builds from a concrete observation: pull requests written with AI where the author cannot explain why a given approach was chosen, or AI output passed along with the judgment left to the reader. The author says these scenes produced a vague discomfort that only became nameable after encountering t.wada's lecture material, "ソフトウェアエンジニア人生サバイバルガイド", and its line that effort can be outsourced but capability cannot.

The piece then separates AI use into two modes. In effort compensation, the user keeps the judgment and hands the labor to the AI; in capability compensation, the user hands over the judgment too. The author ties the choice to who bears responsibility for the result: within hobbies or personal research it stays with the individual, but at work it spreads to the team, the company, and other specialists. From there the essay traces the consequences — an inability to review or debug what you did not understand, reliance on other specialists for confirmation, and work that can end up slower than if those specialists had been involved from the start.

The author extends the argument beyond engineering to legal, medical, design and writing work, noting that engineers themselves tend to use AI as a capability substitute outside their own expertise, such as with contract wording. The recommended posture is to judge AI output within your own specialty and keep learning so that more tasks can eventually be delegated as effort; outside that specialty, the advice is to go to the specialist first rather than producing something with AI and handing it over.