By: Stephen Woodard
How a career in enterprise systems led to a fundamentally different approach to AI and the written word.
Before building Thanis, I spent a year working on a problem that had nothing to do with writing. The question was simpler, and in some ways more unsettling: how do you create trust when a machine becomes capable of generating outputs that look increasingly authoritative?
That work eventually produced two patented frameworks, the Symbolic Containment Firewall and the Canonical Containment Protocol. Both were built around the same core idea: AI-generated outputs should not automatically become real-world actions. There needs to be a verification layer between generation and execution. A place where people can review, challenge, and audit what a system is producing before it influences decisions that matter.
When I turned my attention to AI writing tools, I kept seeing the same gap. Systems were generating content faster and faster. Very few were helping anyone evaluate whether that content was actually clear, accurate, or reflective of the person behind it. The verification layer was missing.
That observation, not a market opportunity or a product roadmap, is what became Thanis.
For most of my professional life, I worked on systems. Over twenty-five years in technology, including five years at Amazon Web Services, I became accustomed to solving problems by removing friction, simplifying complexity, and helping organizations move faster. When large language models arrived, I was fascinated, but my interest quickly moved beyond generation. The first time you watch a machine write a coherent article, summarize a complex report, or answer questions conversationally, it is difficult not to appreciate the significance. These systems represented a genuine technological breakthrough. What seemed to be missing was any equivalent mechanism for evaluating what they produced.
The Voice Problem
I first noticed it while working with writers.
Some were writing memoirs. Some were documenting deeply personal experiences. Others were trying to make sense of difficult periods in their lives through writing. These were not people looking for content generation. They were looking for understanding. They wanted help shaping a story, organizing ideas, strengthening structure, and finding clarity without losing ownership of what they were trying to say.
What I found was that many generative AI tools were remarkably good at producing cleaner language and surprisingly poor at preserving the voice behind it. The systems wanted to rewrite everything. A rough paragraph became polished. An uncomfortable sentence became softer. A personal observation became more generic, technically cleaner, but no longer sounded like the person who lived it.
That observation stayed with me.
The Learning Problem
Later, I started hearing a similar concern from a completely different group: educators.
The professors I spoke with were not anti-AI. In fact, many were actively exploring ways to bring AI into their classrooms. They understood that students were already using these systems and that trying to ban them entirely was unrealistic. Recent studies show that generative AI adoption among students has become widespread. AI is no longer an emerging technology in education. It is already part of the workflow.
But the educators I spoke with kept returning to the same concern: if the AI writes the paper, where does the learning happen? If the AI builds the argument, where does the student’s thinking appear? If the AI improves every paragraph automatically, how do you know what the student actually understands?
The more conversations I had, the more I realized I was seeing the same problem from two directions. Writers were worried about losing ownership of their voice. Educators were worried about losing visibility into the learning process. Both groups were really asking the same question: what role should AI play in writing?
The industry’s answer seemed obvious: generate more, generate faster, generate better. But I began wondering whether the industry was solving the wrong problem. We already know AI can generate language. The harder challenge may be helping people evaluate it.
Building a Feedback Layer, Not an Authorship Layer
That realization became the foundation for Thanis.
I did not set out to build another AI writer. The world already has plenty of those. Instead, I became interested in whether AI could function as a revision and evaluation layer rather than an authorship layer. Could a system help a writer identify structural weaknesses without rewriting the work? Could it help a student understand where an argument breaks down without generating the argument itself? Could it help someone improve a draft while keeping ownership of the thinking process intact?
What emerged from those questions was not another AI writing tool focused on generating content faster. It was a writing feedback platform built around a very different goal: helping people better understand, evaluate, and revise the work they had already created.
Using Thanis is intentionally straightforward. A student, writer, researcher, or professional uploads a draft and receives structured feedback around clarity, organization, argument flow, consistency, tone, evidence, and revision quality. Instead of replacing the writing, the system evaluates it. The goal is not to produce a finished document. The goal is to help the writer produce a stronger one.
For students, this becomes particularly useful when working against assignment requirements and grading rubrics. A student can upload an essay, research paper, discussion post, capstone project, or thesis draft and compare their work against the expectations of the assignment. Thanis helps identify where arguments are underdeveloped, where structure weakens the paper, where evidence may be insufficient, and where revisions could improve alignment with the rubric. The student still has to think, research, and write. Thanis simply illuminates where the work can improve before submission.
What surprised me was how quickly educators understood the value of that approach. Most professors I spoke with were not asking for another AI system that could complete assignments. They wanted a way to support revision, critical thinking, and learning without encouraging dependency on generation. In many ways, Thanis became less of a writing assistant and more of a structured feedback environment.
Deep Trace: The Patterns Writers Can’t See
As the platform evolved, another challenge emerged. Most writing tools evaluate a single draft in isolation. But writers rarely operate that way.
We develop habits over time. We repeat sentence structures. We rely on familiar endings. We return to the same explanatory patterns and pacing decisions. Editors often notice these patterns quickly. Writers often do not.
That observation eventually led to Deep Trace, an archive-aware analysis layer inside Thanis that compares current work against previous writing and identifies recurring patterns across a writer’s body of work.
Most writing tools react to the draft sitting directly in front of them. That is useful up to a point. A single-pass system can identify a confusing paragraph, a weak transition, or a conclusion that collapses under its own summary. What it often misses is recurrence. Writers repeat ourselves. We repeat sentence rhythms, emotional framing habits, pacing decisions, explanatory shortcuts, image choices, and structural preferences. A single document rarely reveals the pattern. An archive often does. Deep Trace was designed to surface those patterns.
Building it turned out to be significantly harder than building a standard AI writing tool. Getting a model to comment on a paragraph is relatively straightforward. Building a system that can retrieve historical context, analyze recurring patterns, avoid generic observations, and return feedback that is genuinely useful required a very different architecture. There were days spent staring at logs, questioning whether I had made the problem more complicated than it needed to be. In some ways, I had. But that complexity was the point. The patterns writers return to are rarely obvious in a single draft. They only surface across time.
The Scarcity That Actually Matters
That challenge taught me something important about where AI writing is headed.
We are already reaching a point where almost anyone can produce competent language on demand. Students can generate essays. Professionals can generate reports. Marketers can generate campaigns. The scarcity is no longer language. It’s judgment. It’s authorship. It’s knowing whether the words in front of us still reflect the thinking and intention of the person whose name appears at the top of the page.
That is the future worth paying attention to. Not a future where AI replaces writers, or where collaboration becomes so seamless that nobody can tell where the human ends and the machine begins. The future that interests me is one where AI helps people become more thoughtful writers, stronger communicators, better researchers, more deliberate students, and more careful editors.
A future where AI can help evaluate an argument without owning it. Help improve a story without rewriting it. Help a writer see their work more clearly without taking it away from them.
That is ultimately what Thanis was built to do. Not generate more language, but help people stay connected to the language they create.
Once AI can write almost anything, the question that matters most is no longer whether the machine can produce the words. It’s whether the human behind them remained present in the work.
I believe that question is going to matter far more than most people realize. And I think we are only beginning to understand why.
