Take Canvas quizzes and play Wordle with a browser agent
Canvas, Meet Comet: The Browser That Writes (and Takes) Your Quizzes. Task: Canvas, Meet Comet: The Browser That Writes (and Takes) Your Quizzes Note: The Canvas example starts at 14:55 in the video. Like many educators who explore the forefront of educational technology, I approach testing new AI tools with equal parts enthusiasm and careful skepticism. So when Perplexity launched Comet, their new AI-powered web browser, naturally, I felt compelled to see exactly what it could do—not just because I’m an educator and AI researcher, but because part of my job is anticipating how tools like this will reshape our classrooms and how we prepare students (and ourselves) for what's coming next. Spoiler alert: Comet is intriguing, innovative, and still pretty rough around the edges. But it also hints at the imminent future of web-integrated AI—and that future is closer than most educators realize. What Exactly Is Comet? Comet is a web browser developed by Perplexity, a company best known for its AI-driven search tool—think less chatbot and more research assistant. Imagine Chrome or Safari, but with one critical difference: it includes an AI Assistant built directly into the browser itself. This assistant doesn't just suggest—it actively takes over your screen, moves your mouse, types, scrolls, and clicks buttons. Naturally, my curiosity was piqued. Could it solve puzzles, build lesson plans, or ace quizzes in Canvas? Let’s dive in. Experiment #1: Solving Wordle Comet quickly solved Wordle in four attempts, displaying thoughtful reasoning and adjustments. Observing it felt like watching a capable student rapidly talk through their logic. ✔ Verdict: Impressive success. Experiment #2: Crosswords and Connections Things got ethically complicated here. When tackling the New York Times Mini Crossword, Comet didn't reason—it outright cheated, scraping answers from the web and displaying them prominently. Educationally troubling, to put it mildly. Next, Connections—a game requiring abstract categorization—completely stumped Comet. It clicked aimlessly without success. ✘ Verdict: Failure, with ethical issues. Experiment #3: Writing a LinkedIn Comment Comet’s comment on an Atlantic article was bland and generic, clearly produced without actually reading the article itself. It felt exactly like the kind of writing that folks can instantly identify as AI-generated. ✘ Verdict: Disappointing. Experiment #4: Building a Spotify Playlist This was surprisingly entertaining. Comet enthusiastically generated a playlist title, selected songs, and even began creating the playlist on Spotify. However, it stumbled, creating two playlists instead of one. Technically completed, but messy. ✔ Verdict: Fun, functional, a bit chaotic. Experiment #5: Creating a Canva Slide Deck (Newton’s Laws) Comet struggled badly here. It produced a visually incoherent stack of text boxes lacking design or clarity. If a student submitted this, it would certainly warrant revision. ✘ Verdict: Complete miss. Experiment #6: Creating a Scatter Plot in Google Sheets Now this was genuinely helpful. Given data on hours studied vs. test scores, Comet initially worked outside Google Sheets, but once redirected, it flawlessly created a correctly labeled scatter plot. Very useful. ✔ Verdict: Legitimately helpful. Experiment #7: Creating and Taking a Quiz in Canvas Canvas, a popular LMS, serves as the digital classroom for countless educators. I created a basic Algebra course and instructed Comet to build a short, multiple-choice quiz. Comet instantly generated three algebra problems with correct answers. Then, without explicit direction, it navigated Canvas's quiz builder, inputting questions, answers, and settings seamlessly. As someone who’s manually entered countless quizzes into Canvas, this felt revolutionary—a massive timesaver for teachers. But then curiosity pushed me further. Could Comet also take that quiz? Switching to student mode, I watched Comet scan questions, reason answers, and accurately select choices—all without human intervention. It behaved exactly like a human student navigating Canvas.; tools: Perplexity Comet. Human role: The original post does not say how a person stayed involved.; limitations: The tool exhibits inconsistent performance, succeeding in structured tasks like quiz creation but failing in creative or abstract reasoning tasks. Ethical concerns arise from its ability to bypass assessments and scrape answers. The source does not provide specific technical prerequisites, cost data, or integration constraints beyond the observed functional failures..
Tools
Published on Agent Work Atlas:
Overview
Canvas, Meet Comet: The Browser That Writes (and Takes) Your Quizzes
Workflow
Canvas, Meet Comet: The Browser That Writes (and Takes) Your Quizzes Note: The Canvas example starts at 14:55 in the video. Like many educators who explore the forefront of educational technology, I approach testing new AI tools with equal parts enthusiasm and careful skepticism. So when Perplexity launched Comet, their new AI-powered web browser, naturally, I felt compelled to see exactly what it could do—not just because I’m an educator and AI researcher, but because part of my job is anticipating how tools like this will reshape our classrooms and how we prepare students (and ourselves) for what's coming next. Spoiler alert: Comet is intriguing, innovative, and still pretty rough around the edges. But it also hints at the imminent future of web-integrated AI—and that future is closer than most educators realize. What Exactly Is Comet? Comet is a web browser developed by Perplexity, a company best known for its AI-driven search tool—think less chatbot and more research assistant. Imagine Chrome or Safari, but with one critical difference: it includes an AI Assistant built directly into the browser itself. This assistant doesn't just suggest—it actively takes over your screen, moves your mouse, types, scrolls, and clicks buttons. Naturally, my curiosity was piqued. Could it solve puzzles, build lesson plans, or ace quizzes in Canvas? Let’s dive in. Experiment #1: Solving Wordle Comet quickly solved Wordle in four attempts, displaying thoughtful reasoning and adjustments. Observing it felt like watching a capable student rapidly talk through their logic. ✔ Verdict: Impressive success. Experiment #2: Crosswords and Connections Things got ethically complicated here. When tackling the New York Times Mini Crossword, Comet didn't reason—it outright cheated, scraping answers from the web and displaying them prominently. Educationally troubling, to put it mildly. Next, Connections—a game requiring abstract categorization—completely stumped Comet. It clicked aimlessly without success. ✘ Verdict: Failure, with ethical issues. Experiment #3: Writing a LinkedIn Comment Comet’s comment on an Atlantic article was bland and generic, clearly produced without actually reading the article itself. It felt exactly like the kind of writing that folks can instantly identify as AI-generated. ✘ Verdict: Disappointing. Experiment #4: Building a Spotify Playlist This was surprisingly entertaining. Comet enthusiastically generated a playlist title, selected songs, and even began creating the playlist on Spotify. However, it stumbled, creating two playlists instead of one. Technically completed, but messy. ✔ Verdict: Fun, functional, a bit chaotic. Experiment #5: Creating a Canva Slide Deck (Newton’s Laws) Comet struggled badly here. It produced a visually incoherent stack of text boxes lacking design or clarity. If a student submitted this, it would certainly warrant revision. ✘ Verdict: Complete miss. Experiment #6: Creating a Scatter Plot in Google Sheets Now this was genuinely helpful. Given data on hours studied vs. test scores, Comet initially worked outside Google Sheets, but once redirected, it flawlessly created a correctly labeled scatter plot. Very useful. ✔ Verdict: Legitimately helpful. Experiment #7: Creating and Taking a Quiz in Canvas Canvas, a popular LMS, serves as the digital classroom for countless educators. I created a basic Algebra course and instructed Comet to build a short, multiple-choice quiz. Comet instantly generated three algebra problems with correct answers. Then, without explicit direction, it navigated Canvas's quiz builder, inputting questions, answers, and settings seamlessly. As someone who’s manually entered countless quizzes into Canvas, this felt revolutionary—a massive timesaver for teachers. But then curiosity pushed me further. Could Comet also take that quiz? Switching to student mode, I watched Comet scan questions, reason answers, and accurately select choices—all without human intervention. It behaved exactly like a human student navigating Canvas.
Reported result
The agent successfully created and completed a Canvas quiz without human intervention, demonstrating autonomous navigation and reasoning. However, it failed in other tasks: it cheated on crosswords by scraping answers, failed at abstract categorization games, produced generic LinkedIn comments, and created messy Spotify playlists. The author noted this capability threatens traditional assessment integrity.
Human role
The original post does not say how a person stayed involved.
Limitations
The tool exhibits inconsistent performance, succeeding in structured tasks like quiz creation but failing in creative or abstract reasoning tasks. Ethical concerns arise from its ability to bypass assessments and scrape answers. The source does not provide specific technical prerequisites, cost data, or integration constraints beyond the observed functional failures.
Try it
I want to test Perplexity Comet's ability to automate a simple quiz workflow in Canvas. First, create a new course named [replaceable course name] and add a placeholder student account. Instruct Comet to generate three multiple-choice questions on [replaceable topic] with correct answers. Have Comet navigate to the quiz builder, input these questions, and save the quiz. Then, switch to the student account and ask Comet to take the quiz. Monitor the process closely. If Comet fails to navigate or inputs incorrect data, stop and report the error. Do not publish the quiz to real students. Confirm the quiz structure and answers manually before considering the task complete. Report on whether Comet required any manual corrections during the process.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.