Banning AI in Schools Is as Smart as Banning Calculators

01 /The Concept

The rise of ChatGPT and similar AI assistants has made one thing clear: students are using them to write essays, papers, and assignments at scale. AI detectors followed, attempting to help teachers distinguish machine-generated text from student work – with mixed results.

Rumi's founders take a different position entirely. They argue there's nothing wrong with students using AI to write academic work – just as there's nothing wrong with using a calculator to solve math problems.

So they built a writing platform where students compose essays and assignments with or without AI assistance, and teachers review the results. The key differentiator: Rumi's editor records every step of the writing process.

If a student copies text from ChatGPT and pastes it into their work without proper citation – the editor flags that passage, both in the student's interface and in the teacher's review view.

Inside the editor, students have access to Rumi's built-in AI assistant – functionally a wrapper around a popular LLM. It currently handles two tasks: answering questions and paraphrasing text. But because it's embedded in the platform, every prompt the student sends is logged. Teachers can see the full list of queries alongside the completed assignment.

This tells the teacher not just what the student struggled with, but how creatively and strategically they used AI to work through those difficulties.

When assigning work, teachers set an AI usage policy: no AI, AI for grammar and paraphrasing only, or unrestricted use. When the student submits, the work appears in the teacher's review queue.

The platform then automatically summarizes how the assignment was produced – something like "significant time was spent on the text, few external passages were incorporated, only a handful of draft versions were created" – giving the teacher a useful starting point for evaluation.

Teachers can annotate any part of the text with comments, visible to students after grading. They can also replay the assignment's creation step by step, watching how the text evolved – which surfaces the student's actual thinking process.

Pricing is $3 per student per month (or $2 billed annually). School and district licenses are negotiated directly.

Rumi was founded last year and took first place at the University of California's startup accelerator in November. It's still early stage – the company has raised $300K in initial funding.

02 /Why It Matters

Rumi's founders frame their core distinction simply: most AI assessment tools evaluate *what was written*; Rumi helps evaluate *how it was written*.

This echoes a Socratic principle that's worth revisiting: the real goal of education isn't to produce correct answers, but to develop the capacity to think. Tracking a student's reasoning process – not just their output – is how you actually teach someone to think.

To develop that capacity, teachers need visibility into the process. AI makes this newly possible, in the same way calculators changed math education. Before calculators, most teaching time went to drilling arithmetic. After calculators, teachers *could* have shifted to teaching students what to calculate and why – most didn't, and instead started drilling harder arithmetic.

The same dynamic is playing out with AI. Teachers discover that their old assignments can now be completed in seconds by a machine. Rather than using that as an opportunity to assign more substantive work, most are trying to restrict AI use on the same old assignments.

But not everyone.

EdLight – [covered here](/review/uchit-nuzhno-na-salfetkah) in March – built a learning platform whose slogan is "See how students think." Students submit work as scanned handwritten pages showing their thought process on math problems. Teachers give feedback on the process, not just the answer. That startup has raised $7.25M.

Packback, like today's Rumi, puts an AI assistant in both students' and teachers' hands – helping students produce better work and helping teachers evaluate it. They've raised $1.5M.

The underlying thesis across all these platforms: AI should help students and teachers spend less time on mechanical tasks, freeing both for more substantive engagement with the actual content of learning.

03 /Opportunities

The broadest opportunity here is rethinking the learning process for the AI era. Not how to restrict AI, but how to allow it in ways that push students toward more meaningful work – and give teachers the tools to teach thinking rather than just procedures.

This is a genuinely new problem space, which means it's not yet littered with tried-and-failed attempts. That's rare.

More specifically, you can extend Rumi's core insight – recording and tracking how students solve problems, whether writing essays or working through math – in several directions.

One direction: AI that gives students hints and asks guiding questions in real time during problem-solving. Sizzle – [covered here](/review/ii-vzorvjot-obrazovanie-sovsem-s-drugoj-storony) last summer – does exactly this for homework, raising $7.5M.

Another gap: Rumi currently requires teachers to manually scrub through version histories by dragging a timeline slider. That's tedious. An AI layer that automatically analyzes the sequence of drafts and surfaces key insights would let teachers spend time giving substantive feedback rather than doing the forensics themselves.

What makes this space attractive right now is that the problem is newly defined. You can build here without worrying about reinventing something that already exists and failed – unlike most startup categories where every idea has been tried three times already.

So for once: build something genuinely new.