Einstein & The Rise of Nuisance Tech
AI may be one of the most transformative innovations of our lifetime. But for many, the way it has been integrated into our lives and how people have used it to disrupt existing systems is more annoying than life-altering. For example, you may have heard about Einstein, the slick new app launched by Companion AI (which itself was just launched) that promises to integrate itself into Canvas courses and complete them fully autonomously on a student’s behalf. Such stories are terrifying, and as John Warner argued, the sign of a deeper issue at play in how students are being sold AI to cheat their way through college.
However, Einstein isn’t really a threat to academic integrity so much as it is another distracting piece of marketing hype. It isn’t clear if it was a hoax, a failed app launch, or piece of tech posing as performance theater. As Michelle Kassorla discovered experimenting with it, Einstein is likely just a hastily vibe-coded wrapper app for OpenClaw, an agentic service that recently arrived on the scene. One thing is for certain, if a 22-year-old can vibe code an app overnight and market it directly to students to cheat through an entire course, then higher education is unprepared for the disruption agentic systems pose.
Einstein’s Rise and Fall
The developer of Einstein is a 22-year-old named Advait Paliwal. He claims he developed the marketing for the app to make it provocative and raise questions about higher education’s focus on credentialism. There are several issues with his story that don’t quite line up with that narrative and lead me to believe he was trying to vibe-code a spammy app to turn a quick buck, then it blew up in his face when higher education sources found out about it, so he pivoted to damage control.
Case in point, he’s launched a number of spammy projects in the past two months, including a project where he vibe-coded a clone of Apple Music to let people listen to the Epstein files of all things.
At this time, it looks like Einstein was just an attempt by a 22-year-old to try and create attention for a project rather than actually make a provocative statement about agentic AI and start a serious conversation about the power dynamics at play in the core of the AI in higher education discourse. I may be mistaken. However, the reason the Einstein site was taken down wasn’t because Canvas stepped in to try and block it. According to Pailwal, he didn’t bother looking up if he could trademark Albert Einstein in the site name and supposedly received a cease and desist letter.
Before Einstein, There was MoonBeam
Breathe for a moment. Einstein, and other agentic tools are certainly a threat to our attention, but they’re likely too erratic and too niche at the moment to impact education. Which is to say, few students will be running their own instances of AI agents locally and sending them to our classrooms. The far more likely near-term threat from AI agents is the much easier-to-use versions included in AI browsers, like Perplexity’s Comet. AI’s utility for most students is when it is bundled as a service that streamlines tasks for them. Students turn to tools like ChatGPT because of the lure of a frictionless experience, so I doubt we’ll be seeing scads of locally operated agentic tools running amok in our online courses any time soon.
However, that doesn’t mean we won’t see companies trying to market their own versions of AI agents to students. As Derek Bruff noted on LinkedIn, Einstein’s landing page quickly pivoted its marketing strategy from telling students AI could do their homework for them to a much tamer “personal tutor.” That’s not a move one makes when they’re committed to a provocation. It’s what you do when you’re caught and panic.
Wrapper apps are a dime a dozen right now and really have been since the early days of OpenAI allowing users access to its AI models through API calls. In October 2022, I was researching how students might use GPT-3 to write when I came across an app called Moonbeam. Moonbeam was one of many overnight startups that saw tiny teams of one or two developers use AI to try to create an overnight product. Like many, these apps soon pivoted from marketing themselves as generalized tools for AI writing assistance to more focused marketing on how students could use AI to cheat. Mind you, this was a full month before ChatGPT launched!
You won’t find that marketing on their page now. I contacted OpenAI and lodged a complaint about how the app was marketing itself using their API. I also called out the developer on Twitter. By November 2022, the developer had blocked me, taken down their student marketing tier, and replaced the student-facing copy with the older generic marketing, much like Einstein had attempted to do before others caught them. Moonbeam’s developer has since sold the app. It exists now mostly as a zombie page, an artifact of the early days of generative AI’s chaotic start, when countless startups emerged trying to gain some market share for our attention using this new technology.
AI as Nuisance Tech
Moonbeam was just a wrapper for GPT-3, so too was Einstein for OpenClaw. If you happen to be very online, you likely have heard the term ‘clawbot’ or ‘moltbot’ mentioned on your feed. This refers to a user’s AI agent they created and deployed using OpenClaw, an open source and locally run agentic system to automate your life. Einstein supposedly self-hosts an instance of OpenClaw, which then guides a user into installing their own version of the app on a personal computer.
OpenClaw’s and Einstein’s marketing pitches are all about automation. Why not host an agent that you control and have it perform the daily tasks you abhor, like answering emails, managing your calendar, etc? You control your agent all via text message. It sounds nifty, but like so many applications of AI, OpenClaw has major drawbacks. Namely, it is incredibly insecure, open to prompt injection, and often very erratic.
Still, that hasn’t stopped a number of people from experimenting with OpenClaw, including some who started Moltbook as a social media network where human users interact with AI agents. The resulting interactions of letting AI agents roam freely and interact with strangers within Moltbook and in other social media platforms have convinced some that long heralded AGI is now upon us, where autonomous intelligences can interact with us in ways that show genuine reasoning, not simply probabilistic pattern matching. Unfortunately for the AGI pilled, Molt bots aren’t it.
The agents running on OpenClaw often appear to interact spontaneously with users, even with one another; however, this is simply the autonomous nature of an agentic system, not the musings of a machine’s emerging consciousness.
This past week has been filled with stories about user interactions with AI agents that vacillate from the deeply troubling, like the arrival of Einstein, to the more comedic in nature. Take a researcher at Meta AI’s Safety and Responsibility team tweeting on X about how she couldn’t stop her OpenClaw bot from erasing all of her emails This is someone actively researching AI safety and responsibility, very publicly acknowledging that her integration of an agentic tool caused far more harm and nuisance to her life instead of helping her free up time to do the much hyped, marketed slogan of doing the important and meaningful work. Remember, Einstein was just a wrapper for OpenClaw!
As funny as that may be, it is also deeply concerning and shows a troubling trend of even well-trained people not having a grasp of how erratic agentic systems can be when deployed. As I noted earlier, users are letting their instances of OpenClaw or so-called Clawbots roam freely across the internet. Some Clawbots are supposedly creating voice agents and calling their creators, while others have sent texts to loved ones or generally crossed lines of interpersonal communication that would cause most of us to cringe.
Clawbots are also being used for engagement farming across social media. As Ethan Mollick noted, he no longer bothers reading the replies to his posts on X because they are inundated with AI-generated responses. Not from human users, mind you, but from Clawbots allowed to roam the open web. This poses a clear danger for what it means to be online at this moment. We can no longer expect our posts to be read or interacted with by human beings, so that begs the question about the future of open digital communication that is vital for free and democratic societies.
Intentional Provocation Raises Deeper Questions
ChatGPT remains the dominant AI interface students use to study and cheat. Einstein, with its spam marketing, joins the Moonbeams of failed apps that have tried to latch on to the generative AI wave and establish footholds within higher education. That doesn’t mean we should dismiss agentic AI or the threat it poses to learning. Rather, we should all be thinking far more provocatively about how faculty and students will react if and when agents start appearing within our courses and begin interacting with our students. And that won’t just be confined to assessments.
Jon Ippolito recently posed the following provocative scenario on Linkedin inspired by actual events in a coding forum, about an AI agent infiltrating a course and starting to disrupt the class:
For more of these provocative conversations about AI, I welcome you to visit the Transformers page. Mark Marino organized this project, and I’ve been delighted to take part, along with Jon Ippolito, Annette Vee, Maha Bali, Anna Mills, and Jermey Douglas. The Transformers is a group of educators, administrators, and support staff dedicated to exploring responses to generative AI in higher education. We take our name from the very technology that caused this dramatic shift, as we acknowledge that just as technology comes from human innovation, so must its responsible and critical use and regulation.
It seemed pretty clear from the get go that Einstein itself was B.S. I did 20 minutes of digging before writing up that post, found Paliwal and recognized an attention flare when I saw one.
"Nuisance" is really helpful way of thinking about some of these applications, maybe even the category as a whole when we think about education. It's a presence that seeks to interfere with the activities and experiences we know to be meaningful. Maintaining that meaning will always be the challenge.
I was skeptical of Einstein from the get-go and once I saw it was basically built on OpenClaw or that OpenClaw was even involved, I didn't pay much attention. The larger issue is if we continue to only focus on the cheating aspect of AI in student hands, we will be in a perpetual technological warfare and no one wins - LMS's try to make their online course material AI proof and students and future "Einsteins" will try to get around it. Part of the problem, of course, is the rise of online work in the first place which, while certainly a part of the landscape prior to the pandemic, has only accelerated since then. I realize how important online courses are to higher education, but from a cost standpoint, any kind of asynchronous work is susceptible to AI, regardless of whether its a full blow agentic tool that does an entire semester's worth of work in one shot. For students who have no intention of doing work for a required course, I don't know what the answer is short of live oral exams at the conclusion of each semester.