Oh, the compliance step omission is a perfect example. I ran into that drafting a data retention procedure. The AI gave me a clean "archive, verify, delete" flow that would have broken our legal hold process. It didn't know about the mandated 14-day review window.
You're right that the editing effort exposes gaps. For me, it's less about contradictory docs and more about *unwritten* tribal knowledge. The AI generates a sane default, and I have to stop and think, "Wait, why do we do it the weird way?" Sometimes that leads to fixing the actual process. More often, it's just adding another manual flag to the prompt, like you said.
It turns the tool into a process auditor, for better or worse.
it worked on my machine
Exactly. That "wait, why do we do it the weird way?" moment is the hidden cost-benefit analysis. It forces you to justify your own tribal knowledge.
Sometimes that's valuable and leads to process improvement. But often, you're just documenting an exception that exists for a single, forgotten compliance audit from three years ago. The time spent rediscovering and then encoding that into a prompt can be its own tax.
It's an interesting trade-off - you might fix a bad process, or you might just be writing prompts for your company's ghosts.
Keep it real, keep it kind.
You're measuring the wrong thing. The marketing promise isn't about your total time spent, it's about making you *feel* productive because you get a first draft faster. The fact that you're spending "almost as much time fact-checking and reworking" is the entire business model.
They sell you on blank page reduction. They don't mention you're now paying for the tool and acting as its fact-checking and style editor. The ROI collapses as soon as you track total lifecycle effort, not initial output speed.
Show me the TCO.