Designing Editorial Freedom in the Age of AI

Everything began one afternoon, not with a breakthrough, but with a quiet realisation. I had just finished a collection of five books I wanted to publish — nearly two years of continuous work, revisions layered upon revisions, entire sections written and abandoned in silence. When I closed the final file, I did not feel relief. I felt a strange clarity.

The traditional path was obvious. Send the manuscripts to a publisher. Let editors reshape them, smooth the edges, adjust the structure so the books would “fit” the market. I had walked close to that road before, and I knew what it meant: architecture softened, tension reduced, voice translated into something easier to sell.

I did not want my books domesticated.

I had already published close to ten titles through Amazon KDP and IngramSpark. Those projects worked well, but always with technical intermediaries — layout experts, distribution specialists, people from Fiverr who helped solve problems I did not want to learn myself. It was efficient, but each technical layer created distance between the work and its origin.

This time I chose another path. I decided to do it myself and see how far artificial intelligence could support me — not as a substitute for authorship, but as part of a system I would design consciously.

I am speaking about three concrete models: a personalised ChatGPT instance, Claude, and Gemini. Not as abstract ideas, but as specialised language partners working inside a living editorial process.

And that distinction changed everything.

I was not trying to automate writing. I was trying to design the process.

When AI Stops Being a Tool and Becomes Architecture

At the beginning, I worked the way most people do. I asked scattered questions. I requested corrections. I allowed versions to grow organically without a clear structure. Sometimes the results were brilliant. Other times they felt strangely hollow, as if something essential had been polished away.

The problem was not technical ability. The problem was the absence of a system.

Clarity arrived when I realised that each model had a different “vocation.”

• A personalised ChatGPT functioning as a narrative interlocutor, preserving tone,........

© The Times of Israel (Blogs)