I Spent a Year Building an AI Editorial Staff. It Changed Me More Than It Changed the AI
I spent the last year building an AI-assisted publication.
Like a lot of people, I started with the assumption that the end goal was to make the AI a better writer but I don’t think that anymore.
Over the past year I’ve built specialized agents for research, editing, fact-checking, QA, scoring, and provenance. Every article goes through multiple rounds before it’s published.
The surprising part is that my editorial process got better.
Instead of asking, “Can the AI write this?” I found myself asking:
- Is this actually interesting?
- Is this true?
- Does this add something that wasn’t already on the internet?
- Would I still want someone to read this a year from now?
The AI became less of a ghostwriter and more of an editorial partner. It constantly proposes ideas, challenges assumptions, points out weak arguments, and suggests better structure. In the end, the finished articles aren’t “written by AI.” but AI is definately used.
They’re the result of an iterative conversation where the AI contributes breadth and speed, while I contribute judgment, taste, and accountability.
Ironically, the biggest improvement wasn’t the model, It was the questions I learned to ask.
I’m curious whether anyone else building with AI has experienced something similar. Has your product changed the way you think, rather than simply automating what you already did?
The line about the biggest improvement being the questions, not the model, feels right. A lot of people still treat AI like a ghostwriter problem when the real win is usually better structure, better pressure testing and faster iteration. That's how I think about DictaFlow too. The useful part isn't letting AI rewrite your voice, it's getting the messy first pass out fast, then cleaning it up without losing the original thought. Once you start seeing writing as a chain of decisions instead of one magic generation step, the workflow feels a lot more honest.
I really like the phrase “chain of decisions.” I hadn’t thought about it that way, but I think that’s exactly what’s happening. Over time I found myself thinking less about producing a draft and more about improving each editorial decision along the way. Ironically, that made me value the human part of the process more, not less. The AI got better too, but mostly because I got better at guiding the conversation.
This resonates a lot. I had a similar shift where the biggest change wasn’t what the AI could produce, but how it forced me to think.
When you break work into research, validation, structure, and clarity, you start seeing how much of “writing” is actually decision-making. AI just makes those gaps more obvious.
The part about better questions is spot on. Once you stop treating AI like a generator and start treating it like a system that reflects your thinking, the whole workflow changes.
Feels less like automation and more like amplification of taste and judgment.
Yes to “amplification of taste and judgment.” That captures what I’ve been struggling to articulate. So have you found that changes how you design AI features? I’m starting to think the goal isn’t reducing human involvement, but making human judgment more effective at the things humans are good at.
I like the shift from "Can AI write this?" to "Is this worth reading at all?" That's a much harder question, and it's one AI can't answer on its own. The best AI workflows seem to improve the human's judgment, not replace it.