2. Anticipate to iterate
As I discussed earlier than, the extra express and protracted your directions are, the extra seemingly you’ll get one thing resembling your intentions. That mentioned, it’s unlikely you’ll get precisely what you need on the primary, second, third, and even fourth attempt—not even for any single facet of your program, not to mention the entire thing. Thoughts studying, not to mention precisely, continues to be fairly a method off. (Fortunately.)
A specific amount of back-and-forth to get to what you need appears inevitable, particularly in case you are re-implementing a mission in a special language. The profit is you’re pressured to confront every set of modifications as you go alongside, and ensure they work. The draw back is the method might be exhausting, and never in the identical method making iterative modifications by yourself can be. While you make your individual modifications, it’s you versus the pc. When the agent is making modifications for you, it’s you versus the agent versus the pc. The determinism of the pc by itself is changed by the indeterminism of the agent.
3. Take full duty for the outcomes
My closing takeaway is to be ready to take duty for each generated line of code within the mission. You can’t resolve that simply because the code runs, it’s okay. In my case, Claude might have been the agent that generated the code, however I used to be there saying sure to it and signing off on choices at each step. Because the developer, you’re nonetheless accountable—and never only for ensuring every thing works. It issues how nicely the outcomes make the most of the goal language’s metaphors, ecosystem, and idioms.
