Eric “Buzz” Rubenstein argues that the next generation of AI tools shouldn’t police student writing—they should illuminate the thinking behind it, giving teachers a clearer understanding of how students learn.
INTERVIEW | by Victor Rivero
When generative AI burst into classrooms, much of the conversation centered on detection: How can teachers tell whether students wrote their own work? Eric “Buzz” Rubenstein took a different path. Rather than trying to determine whether AI was used after the fact, the Stanford GSB alumnus and co-founder of Revision History set out to make the writing process itself visible—from first keystroke to final draft. The result is a rapidly growing platform now used by more than 200,000 teachers, alongside a new student-facing tool, Byline, designed to encourage responsible, transparent AI use. In this conversation, Rubenstein discusses why process matters more than prediction, how thoughtful AI can strengthen rather than replace learning, and where writing instruction is headed next.
You’ve positioned “writing process visibility” as more important than AI detection—what did you see in classrooms that made you confident that’s the right wedge, not just a nicer framing?
Helen, an English teacher in Florida, put it best, referring to our Chrome extension, Revision History, and its writing process visibility approach versus AI detection tools:
“It’s a much saner way of assessing the originality of student work while still allowing students to have their own writing process and allows me to feel like I’m teaching again rather than being a warden in a prison.”
What is writing process visibility? What is AI detection?
AI detection tools use probabilistic models to predict how likely it is that a piece of written text was AI-generated, expressed as a percentage. Like a polygraph, AI detection tools are notoriously unreliable and stress-inducing.
Writing process visibility looks instead at the full creation journey of a document, from first keystroke to final draft—typos, revisions, and all. You can watch a lovely video made by the TVDSB tech team that demonstrates our “video replay” feature. This approach doesn’t guess whether you did the work; it shows the entire step-by-step journey that led to a document’s final form.
‘Writing process visibility looks instead at the full creation journey of a document, from first keystroke to final draft—typos, revisions, and all.’
In terms of what we saw in classrooms, we really just tried to listen to teachers, who kept sharing the same story about their experience with AI and students’ written work.
Part one: Teachers could no longer gauge whether a student authentically wrote a document or outsourced the work to AI, and initially turned to AI detection models as a potential solution.
Part two: Teachers quickly found that AI detection models were unreliable, biased against non-native learners, and offered zero pedagogical value. They didn’t like the “gotcha” feel. And nobody knew what an AI score of 78 percent meant.
Listening to this story again and again, we knew AI detection wasn’t the right approach. We also heard great teachers say that, in a world of AI, a document’s final, finished form couldn’t be trusted to reflect a student’s learning level. Instead, great teachers were increasingly focused on the writing process and writing scaffolding to help students make their thinking “visible”—moving from a 2D snapshot of a movie’s final scene to a 3D video showing every frame.
The more we dug, the more confident we became that writing process visibility—the approach we embrace at Revision History—was more fair, transparent, and pedagogically valuable for teachers and students.
So we built a free-to-start Chrome extension for teachers. Teachers add Revision History, open student work completed in Google Docs or Slides, and our extension automatically makes a document’s process “visible.”
With 200k+ teachers using Revision History, what’s one surprising pattern you’ve seen in how students actually write in the AI era that most people would get wrong?
This is an easy one for us, actually! We collect zero student or document data through Revision History. All data is processed locally in the teacher’s browser, so we don’t see any student writing on our end.
In that sense, the surprising pattern we keep seeing is just how relieved teachers, students, IT leaders, and administrators are that no student data makes its way to our servers.
Anecdotally, teachers tell us they’re seeing students “stare at the cursor” less because the writing process is valued over the final, polished output. In other words, students feel they get credit for showing their work, so they’re more motivated to just start.
‘…students feel they get credit for showing their work, so they’re more motivated to just start.’
At the same time, we’re launching a new student-facing writing tool, Byline by Revision History, which a few districts are currently piloting. It pairs with Revision History on the teacher side.
For teachers, Byline enables deeper visibility into students’ writing process, including AI use and active site and extension logging.
For students, with teacher permission on a per-assignment basis, Byline offers AI writing coaching that’s guardrailed, grade-appropriate, and unlocks only after every 50 typed words. Alternatively, AI coaching can be turned off entirely, meaning zero AI coaching is offered, while extension use and site browsing are still logged and visible to the teacher while the document is open.
The AI feedback students unlock is not an open chatbot and is not always available. Every 50 typed words, students can unlock AI feedback in the form of pedagogy-aligned, process-focused prompts: Does my writing flow? Am I staying focused? What else am I missing?
We make sure the AI feedback doesn’t do the work for the student. Instead, it asks questions, encourages a growth mindset, and follows a set of core instructional principles.
You’ve built products that people want to check—like Netflix Wrapped. How are you applying that same engagement mindset to something as traditionally painful as grading writing?
Teachers are busy.
Instead of asking them to adopt an entirely new workflow that makes life more difficult, we wanted to meet them where they already were. And for much of their day, that’s in Google Docs and Slides.
Once a teacher downloads Revision History’s Chrome extension, the tool automatically appears as a layer on top of Google Docs and Slides. Students continue writing exactly as they normally do, so there’s no change on the student side either.
You mentioned Netflix Wrapped—you did your research! That was a Chrome extension that let anyone see their Netflix usage data in a Spotify Wrapped-style dashboard. This may shock you: people watch…a lot of Netflix.
Like Revision History, Netflix Wrapped was born from the idea that unless you absolutely have to make users visit a new website or learn a new interface, there’s no reason to make life more difficult. You might as well offer a solution where they already are, whether that’s Netflix or Google Docs and Slides.
Grading is often seen as the “unpaid chore” of teaching—made even more difficult now that teachers also have to assess writing authenticity. By transforming a written document into a 3D video, we’re trying to turn written work into a story rather than a forensic investigation—and turn a 20-minute investigation into a two-minute review. We simply want to give teachers a little time back.
There’s a fine line between giving teachers insight and creating a sense of surveillance for students—how do you design for trust on both sides?
Given just how unguardrailed AI is today—and how easily accessible it is for students—teachers need deeper insight into students’ writing process. If a teacher doesn’t know how much work was a student’s own versus how much was offloaded to AI, they can’t accurately assess where that student is on their learning journey.
That said, we also have to be fair and transparent to students. To that end, we strongly recommend that teachers let students know they’re using Revision History. Revision History also collects zero student or document data whatsoever; all data processing occurs locally in the teacher’s own browser.
Looking ahead, we want to make sure students aren’t fully disconnected from the world of AI, which will inevitably be part of their lives and careers beyond school. That’s a big part of why we built our new student-facing tool, Byline by Revision History, which encourages responsible AI use.
To us, responsible AI means guardrailed AI assistance on the student side and full teacher visibility into how the student leveraged AI. Trust gets built when students don’t feel like they need to hide their AI interactions, and teachers have full visibility.
‘Trust gets built when students don’t feel like they need to hide their AI interactions, and teachers have full visibility.’
A lot of AI tools try to “assist” student writing directly. You’re focused on helping teachers understand it instead. Do you think most edtech is solving the wrong problem right now?
Coming out of Y Combinator and prior acquisitions, how did you decide this was the problem worth going all-in on—and what did you explicitly choose not to build?
ChatGPT was released in November 2022. I was lucky enough to be just a few weeks into Stanford’s MBA program at the time, so within days AI was impossible to ignore.
Dozens of my early-adopter classmates began leveraging ChatGPT to help with schoolwork—mostly for brainstorming ideas, organizing notes, creating study guides, and similar tasks. But by MBA year two, models had improved so rapidly that a majority of my classmates were using ChatGPT heavily, especially to do written work for them. The “easy button” of AI was simply too tempting.
It also became clear that the AI detection tools that quickly emerged were the wrong solutions to an important problem.
For me, learning how to write was learning how to think. As a college history major, admittedly my modern Egyptian history knowledge hasn’t come in handy much since graduating. But that degree did equip me with the ability to think critically and lean into productive struggle.
Today, teachers have minimal insight into how much work is a student’s own versus AI-generated. That means AI, in its current form, fundamentally threatens the education system’s ability to teach critical thinking and productive struggle, because neither can be reliably measured.
I gravitated toward this problem out of concern for students, a love of writing, and an interest in a really complicated question: How do we get a handle on AI to make sure it helps students learn, helps teachers teach, and preserves trust between students and teachers?
If Revision History works at scale, how does it change what good writing instruction looks like three to five years from now?
The current moment is really challenging. But over the medium and long term, I’m extremely optimistic if we ensure that the AI resources students leverage are guardrailed and fully visible to teachers.
With Byline by Revision History, AI should act as a mini teaching assistant—an unlimited resource students can check in with when their teacher isn’t available, nudging them in the right direction within the rules and guidelines established by their teacher.
That idea—an always-on teaching assistant, available and adaptable to every student—is deeply exciting. And that’s only on the student side.
On the teacher side, that same teaching assistant can turn around and help the teacher learn much more about the student—and ultimately become a better teacher.
The future is definitely scary but also very bright.
—
Victor Rivero is the Editor in Chief of EdTech Digest. Write to: victor@edtechdigest.com
0 Comments