AI Didn’t Make Me a Faster Writer. It Just Made Me a Frustrated Editor.
Disclaimer: This article was written with Levee, my personal AI-writing assistant, a SaaS tool built to enjoy writing again while improving the quality of the articles. In the end, it wasn’t written by AI, it was written by me with AI.
I am not a blogger. I am an SAP tech freelancer.
Having crossed the 10-year mark in my professional career this year, I wanted to look back on my experience and share what I’ve learned along the way. Like anyone starting to write online today, I naturally turned to Claude to help me shape my thoughts. And, just like everyone else, I was deeply disappointed by both the process and the result.
We all experienced the same frustration when we first try using AI to write. We spend an hour setting everything up: project, context instructions, target audience, voice and documentation. Then, we shoot over the headline, wait 30 seconds and watch the article materialize.
Get Michel Escoda Pro’s stories in your inbox
Join Medium for free to get updates from this writer.
It feels magical. ‘Wow. Am I a real blogger or what?!’ It feels like we cracked the content creation game. Disclaimer: We did not
The perfect setup is never enough
Then reality hits. When we realize the article is full of fluff, repetitive phrases, and that generic “AI voice”. Of course, as everyone, I first thought ‘I may have mismanaged the settings’. So I looked up online to refine all I did initially, instructions and context. However, after three times and a result a bit better, this is still not totally what I wanted. What followed was a brutal, soul-crushing two-hour rescue mission. I call it the “de-slopping tax,” and it comes in three distinct flavors of frustration:
The Ping-Pong Chat Trap
This one is the productivity trap, and when we finish with it, we really wonder if this article written with AI was really worth it. We try to fix with prompt like “rewrite this section to be more specific” or “Make it punchier” or “More human”, blablabla.
Over time, the article is slightly improving while your energy drains just as fast as the power grid running the data-center hosting our AI queries. Once we finish this step, we have more or less won the battle for meaning. The ideas that we want to communicate are clear, but it is far from being over.
AI is AI, however you prompt it
Now, we start the style battle, it is the worst and most gruesome. It feels like trying to sculpt a statue with a sledgehammer: messy, exhausting, and utterly hopeless. (This punchline was written by the auto-complete feature, I love it) .
We remove the overused vocabulary, the em-dash addiction, the AI structural rhythm like ‘It’s not just X. It’s also Y; That’s the real X; Here’s why that matters,’. We delete them, rewrite some sentences. The style is better but something is off, isn’t it? It is good but not perfect, it is clear but tasteless. At this stage, it is however good enough to move on to the next step in our endless corrective journey.
(And yes, I see you, “AI humanizer”-skill experts. Sure, those tools help a little, but an overused hack just makes every single article look identical. Never forget that.)
The Fiction-as-Fact Nightmare
This part does not apply to every piece, only the ones that require accurate data. The AI will provide the data point but quite often when we click on the source link, it leads to 404 error or studies that do not exist. And that’s when the real detective work begins. We start hopping from tab to tab, cross-referencing claims, trying to salvage something that isn’t a hallucination. Half the time, we end up scrapping the whole section and writing it from scratch anyway, wondering why we didn’t just do that in the first place. In fact, when an hallucination takes root, it flows to the entire paragraph as AI is using it to generate the argumentation. And that’s when I finally admitted the obvious and maybe you did too: if I’m going to fact-check and rewrite every paragraph anyway, I might as well skip the AI middleman entirely. I was spending more time fixing AI output than I ever spent writing from scratch. So I stopped.
Well, did I really stop?
Not quite, it was at this point that I wondered: ‘Can AI help me write better articles without the gruesome and meaningless editing loops? Can it just help me write faster while keeping me in control?’ I made Levee for that reason, a tool that help writes the article for my personal blog. When I finished it after several weeks of work, I was so proud I wanted others to use it. Not to make money — I am a well-paid freelancer — but to help other writers break free from the same cycle of frustration. To let them experience the joy of writing again. At the same time, I truly believed AI is an amazing and powerful tool. It’s just the narrative pushed by AI labs, the ‘AI will replace you, especially in writing’ has done far more harm than good. AI has a limitation that you cannot overlook : it is not you, and it never will be. So, should you use AI the ‘old’ way ? Letting it generate everything and then fighting it back and forth until you get something acceptable but soulless? Yes, if you’re just writing a sterile corporate memo. But if you want to share your unique view of the world, there will never be a better way than writing the article yourself, perhaps just with a little bit of AI help along the way.
