I am genuinely curious about this question and want to understand people's experiences. I'm a principal engineer with decades of experience, I am currently working on a massive codebase that does RPA (data entry automation for old software). I can get my agents to build complex features from scratch, However, if I don't describe exactly how it needs to be built (low level specs), or if I don't know how the current codebase needs to be updated to support this change, things quickly get out of control. So in other words, if I trust the agents to make any design decisions on their own, then a lot of nuance will be completely missed, leading to major bugs or issues with the SLA.
To summarize my problem, I am finding that the work required to build detailed enough specs, review and evaluate AI changes is far more time consuming (and cognitively demanding) than implementing the features myself.
To summarize my problem, I am finding that the work required to build detailed enough specs, review and evaluate AI changes is far more time consuming (and cognitively demanding) than implementing the features myself.
Where am I going wrong?