According to one of my favorite journalists, Katie Parrott, so much of AI writing happens in a black box. The critics are imagining the laziest possible version of AI-assisted writing, and the writers who use AI seriously haven’t been showing their work, though that’s starting to change. That silence lets the worst assumptions fill the gap.
Writing with AI is not a shortcut. It is a system of pressure.
Here's what I have learned after 2.5 years of using AI to craft my content.
The assumption is that AI replaces thinking. The reality is the opposite. If used well, it forces more decisions, more judgment, and more accountability than writing alone.
1. Writing with AI is not binary
Every writer’s process is different, and most of them would sound unhinged if described in detail. But throw AI into the mix, and suddenly everyone has opinions about the “right” way to get words on a page.
It is not “AI wrote it” vs “human wrote it.”
It is a layered process of thinking, structuring, arguing, rejecting, refining.
The real work lives in:
➢ What I keep
➢ What I reject
➢ Why
That “why” is where thinking happens.
2. The real shift is from tool → system
My writing process has about as much in common with that as cooking does with throwing something in the microwave and calling it dinner.
It didn’t start here.
Back in 2023, I was basically a relay system. Prompt into ChatGPT, copy the output, drop it into a doc, tweak a few lines, repeat. It worked, but it felt mechanical. I was moving words around, not really building anything.
By 2025, I started getting more intentional. I stopped treating AI like a blank box and started feeding it context. My past writing, my tone, my preferences. I experimented with automations, custom GPTs, early workflows. The outputs got closer to how I actually sound, but I was still stuck in that back-and-forth loop, still negotiating with a chat window instead of directing a system.
Now it’s different.
I don’t have one assistant. I have a set of writing agents, each with a job. One is tuned for my memoir, one for LinkedIn newsletters, one for long-form articles, others for specific topics and content streams. Each one runs on a defined set of instructions inside Claude, and each one follows the same disciplined flow from first idea to final line.
It starts with pressure. The agent interviews me, pulls the thinking out, forces clarity before a single paragraph exists. Then we move into structure, where the outline gets shaped, challenged, rearranged until it holds. Drafting comes next, and that’s where things get fluid. Some sections survive. Some get rewritten from scratch. Every line earns its place.
Then comes the part most people skip. Review.
I built a panel of critics into the system. Different lenses, different attitudes. One cuts anything excessive. One questions the argument. One reads like an impatient audience member. One looks for voice and tone. They don’t agree with each other, and they’re not meant to. They’re there to push the work until it either holds or breaks.
From there, it goes through a final tightening pass. Style checks. Pattern detection. Line editing. Sentence by sentence, stripping out anything generic, anything that doesn’t sound like me.
At that point, what’s left is something I stand behind.
It’s not a prompt. It’s not a shortcut. It’s a structured editorial system with built-in tension, and I’m still the one making every call that matters.
Most people use AI like a chat box. That’s entry-level.
The advanced move is building a repeatable workflow:
➢ Brainstorm
➢ Interview
➢ Outline
➢ Draft
➢ Review
➢ Final polish
This is beyond AI prompting, it's process design.
3. The interview phase is where thinking starts
When I work with AI, I don’t start by asking for answers. I start by forcing the thinking out of myself.
The first phase is always an interview. Before a single line gets written, the system pushes me to articulate what’s actually going on in my head. Not the polished version. The real one.
It asks me the questions I would usually skip:
- Why is this on my mind right now
- Where has this shown up in my work, in real situations
- What am I reacting to or pushing against
- What do I believe that others might disagree with
- What do I want someone to walk away seeing differently
Then it keeps going.
- Where is this still messy or unclear
- What part of this idea isn’t fully formed yet
- Am I speaking from experience or assumption
- What happens if I’m wrong
And it doesn’t just take the first answer. It pushes. If I drift into something generic, it pulls me back. If the thinking feels thin, it asks for more. If I’m circling around the point instead of hitting it, it forces me to land it.
By the time I move into structure or drafting, I’m not guessing anymore. I’m working with something that’s been tested, challenged, and clarified.
That first phase does most of the heavy lifting. Everything that comes after is execution.
Instead of asking AI for answers, I ask AI questions. That flips the dynamic:
➢ AI becomes a probe
➢ I become the source of truth
If you cannot answer the questions, the idea is not ready.
4. Structure is negotiated, not generated
Structure is where I spend most of my time, and most people underestimate how much work lives there. After the interview phase, the agents propose an outline based on everything I’ve said. I don’t accept it. I never do. That first version is a starting point, not a direction.
On average, I go through six to eight outlines before anything feels right.
Sometimes more, especially with my memoir. That work is not about organizing sections. It’s about finding the spine of the piece. What stays. What moves. What doesn’t belong. What hasn’t earned its place yet.
I shift sections around, collapse ideas, expand others, cut entire beats that feel thin, and bring in pieces the AI didn’t surface. There’s a constant back and forth until the structure matches what I can feel but haven’t fully expressed yet.
By the time the outline holds, I already know the piece works.
The writing becomes execution.
The real work is still in the decisions.
AI suggests structure, which I always end up fighting, ending up into a back-and-forth shaping of:
➢ Flow
➢ Sequence
➢ Emphasis
The outline is no longer a deliverable. It is a battlefield for clarity.
5. There is no fixed “AI vs human” ratio
Then comes drafting, and this is where people try to measure something that doesn’t behave in clean percentages.
I build the draft section by section off that outline.
Some parts come back close. I take them apart, reshape the language, break the rhythm, inject my phrasing, and make the tone land the way I want it to. Other sections don’t survive at all. I throw them out and rewrite from scratch, guided by the feeling of what the section is supposed to carry, not what the AI produced.
The mix shifts constantly. Paragraph to paragraph. Sentence to sentence. There is no fixed ratio of what stays and what goes, and trying to define one misses the point entirely.
With my memoir, this goes even deeper. I’ve lost track of how many times I’ve rewritten certain sections. I keep tight version control, so nothing is lost, but the number of iterations would surprise most people. Entire passages get reworked not because they’re wrong, but because they’re not precise enough, not honest enough, not mine enough.
At that level, AI is not writing for me. It’s giving me something to react to, something to refine against, something to push back on.
Some sections survive. Some get rewritten completely.
The ratio shifts constantly:
➢ Paragraph to paragraph
➢ Sentence to sentence
Anyone asking “how much did AI write?” is asking the wrong question.
6. The real power is in adversarial review
Then comes review, and this is where the “outsourcing your cognition” narrative falls apart completely. The more I did this the more I realized I need a panel of AI critics:
➢ Minimalist editor (cuts excess)
➢ Narrative tension reviewer
➢ General audience clarity check
➢ Aggressive critic attacking weak arguments
➢ Humor enhancer.
Each one is a set of instructions that tells the AI to read my draft from a specific angle, and none of them are nice about it.
This is critical:
AI is attacking my thinking from multiple angles.
The biggest AHA moment: AI becomes valuable when it disagrees with you.
7. AI exposes weak thinking faster
Because drafting becomes faster.
The criticism around AI being lazy work doesn’t bother me as much as people think it should. Mostly because I know where I cut corners. There are nights where I look at a draft at 5 am and decide it’s good enough. Not great. Not final. Just… good enough. And I move on. But I’ve done that long before AI was in the picture.
I’ve pushed through drafts I knew weren’t fully there because I’d already taken them apart too many times. At some point, you hit fatigue. That’s not a technology problem. That’s part of the process.
What matters is this. The work still demands something from me.
There’s nowhere to hide behind effort anymore. No credit for time spent. Just the output.
➢ Speed removes excuses.
8. AI writing has predictable failure modes
There are moments where I step back and question the whole setup. Did I build a system that actually sharpens my thinking, or did I just make it easier to move faster without going deeper? That question sits there, and I don’t ignore it. AI without intervention becomes:
➢ Too smooth
➢ Too generic
➢ Too “perfect”
Fix that is required here:
➢ Style enforcement
➢ Pattern detection
➢ Line-by-line tightening
Polishing your output is not optional. It is identity.
9. The work still hurts
I still sit there chasing a sentence that won’t land. I still read things out loud to hear if they hold rhythm or fall flat. I still question whether what I’m saying is actually worth someone’s attention or just filling space because I can. I still check tone constantly. Am I being too sharp, too soft, too guarded, too exposed. Is this honest, or does it just sound like it is.
And when something goes out that I know could have been better, I feel it. Immediately.
None of that disappeared when I started working with AI. If anything, it got louder.
Because now I get to a draft faster, which means I get to the real question faster.
Not “did I write this,” but “is this actually good.”
Nothing about the emotional process disappears:
➢ Searching for the right word
➢ Doubting the argument
➢ Reading sentences out loud
➢ Questioning if it’s worth posting or publishing.
Just like any other training,
if it doesn’t feel uncomfortable, you’re not doing it right.
10. AI removes the excuse of exhaustion
Before:
“I’m too tired to refine this.”
Now:
You get to the hard part faster, so you must confront quality directly.
This is the part most people don’t articulate:
You are no longer just a writer. You are an editor of systems.
You design:
➢ The questions
➢ The structure
➢ The critique loops
➢ The standards
Then you operate inside that system.
AI does not remove thinking. It removes friction around thinking.
Which leaves you exposed to one question:
➢ Is your thinking actually good enough to hold up?
So yes, I still consider myself a writer. Not because I type every word from scratch, but because I stay in the loop. I take the hits, I rework the thinking, I hold the standard, and I ship.
AI is part of my process now. It sits in the room with me. But it doesn’t make the decisions.
That part is still mine.
About The Author
Anca Platon Trifan, CMP, WMEP is an award-winning speaker and AI strategist who helps leaders navigate complexity with clarity, energy, and strategic control. Her work focuses on reducing cognitive overload and building systems that support sustained high performance.
With over 20 years of experience in event production and event technology, Anca brings practical, real-world insight to every stage. As CEO of Tree-Fan Events Productions LLC, she helps teams implement technology with intention, improve decision-making, and lead more effectively under pressure.
She is the host of Events: Demystified and founder of the #FIT4EVENTS™ framework, known for her grounded, no-hype approach to AI, AV, and industry leadership.
