For the last two years, the conversation around artificial intelligence has followed a familiar script.
People worry about jobs. They worry about misinformation. They worry about deepfakes, hallucinations, and whether machines will eventually replace human creativity. Every few months, a new model arrives, the benchmark scores improve, and the discussion starts over again.
Meanwhile, something quieter has been happening.
AI writing has become normal. Students use it. Writers use it. Marketers use it. Consultants use it. Researchers use it. Professionals use it. In many organizations, AI-assisted writing has moved from experimentation to routine behavior in less time than it took most institutions to write policies around it.
The question is no longer whether people will use AI to help them write. They already do.
The more interesting question is what happens to authorship when AI-generated language becomes abundant.
That question has occupied a surprising amount of my attention over the last few years while building Thanis. What began as an effort to help writers improve difficult drafts gradually evolved into something much larger: an attempt to understand what role human judgment still plays once generating polished language becomes easy.
The first version of Thanis did not come from academia or enterprise software. It came from working with writers. Many of them were writing about intensely personal experiences — grief, addiction, trauma, family conflict, recovery. The kinds of subjects that people often struggle to put into words because the act of writing is part of processing what happened.
What I noticed was that most generative AI systems were remarkably good at producing clean language and surprisingly poor at respecting the emotional shape of the story. The systems wanted to improve everything. They wanted to smooth rough edges, soften difficult passages, reorganize emotional sequences, and rewrite language into something more polished. On paper, that sounds helpful. In practice, it often removed exactly the things that made the writing feel honest.
A difficult memory is not always supposed to sound clean. A painful experience is not always supposed to read smoothly. Sometimes the awkward sentence is carrying more truth than the polished one.
That realization stayed with me. The writers I was working with were not asking for a machine to tell their story for them. They wanted help understanding the story they had already written. They wanted feedback. They wanted perspective. They wanted another set of eyes. What they did not want was authorship quietly transferred to a machine.
Around the same time, I started having similar conversations with educators. At first glance, the environments could not have looked more different. One group was writing personal stories. The other was grading assignments. But underneath both conversations was the same concern: Who actually owns the work?
The public debate around AI in education often gets reduced into simplistic arguments about cheating. The reality is much more complicated than that. Most professors I spoke with were not anti-AI. In fact, many were actively trying to find responsible ways to integrate it into the classroom. They understood that students were already using these tools and that pretending otherwise was unrealistic.
The numbers support that reality. The 2025 HEPI/Kortext survey found that 88 percent of students in the United Kingdom had already used generative AI in connection with assessments. Globally, similar studies show AI use becoming routine across higher education. The technology is already inside the workflow.
The challenge is figuring out what role it should play. A student who uses AI to explain a difficult concept may genuinely learn from it. A student who uses AI to identify weaknesses in an argument may improve their writing. A student who uses AI to generate an entire paper, however, is no longer receiving assistance. The system has taken over the intellectual work itself.
That distinction matters. It is also where I believe the publishing industry is heading.
For years, writing tools have been evaluated primarily on their ability to generate content faster. Faster drafts. Faster summaries. Faster reports. Faster articles. Faster everything. Generation became the headline feature because it was easy to demonstrate.
Type a prompt. Watch words appear. Be impressed.
But generation solves only one part of the problem. The harder problem begins after the draft exists. Is the argument coherent? Does the structure support the idea? Is the evidence persuasive? Does the tone match the audience? Does the piece still sound like the person whose name will ultimately appear above it?
Those are evaluation questions. And evaluation is where human judgment lives.
The deeper I got into building Thanis, the more convinced I became that the future of AI writing would depend less on generation and more on evaluation. That belief shaped the architecture of the platform itself.
Instead of asking, “What can the AI write for you?” We started asking, “How can AI help you better understand what you already wrote?”
That sounds like a subtle difference. It is not. A generation-first system produces language. A feedback-first system analyzes language. A generation-first system optimizes for output. A feedback-first system optimizes for understanding. Those two design choices lead to very different outcomes.
Today, Thanis uses structured writing analysis, rubric-aware evaluation, revision scoring, and feedback systems designed to help writers, students, academics, and professionals strengthen existing drafts rather than replace them.
The larger question is what kind of relationship we want people to have with AI in the first place. Do we want systems that slowly move humans out of the creative process? Or do we want systems that help humans become more thoughtful participants inside it?
That question extends well beyond writing. As AI becomes capable of generating increasingly sophisticated language, images, code, research summaries, and business communications, authorship becomes more valuable, not less. Trust becomes more valuable. Judgment becomes more valuable. Responsibility becomes more valuable.
The internet does not have a content shortage. It has an authorship challenge.
The next generation of successful AI systems may not be the ones that generate the most language. They may be the ones that help people remain connected to the language they produce.
That is ultimately what I believe Thanis is really about. Not writing faster. Writing with enough clarity, ownership, and intention that the human being behind the words remains visible.
Because once AI can generate almost anything, the most important question is no longer what was written. It is who was still present when it was written.
