Dreambeans and the Finite Feed: What a Google Storytelling App Teaches Edtech About Attention

An app that tells you when to stop is a strange thing to build in 2026. It may also be the most important design idea Indian education hasn't borrowed yet.

It is 9:40 on a Tuesday night in Indore, and Meher has promised herself twenty minutes.

She is in Class 9. Her science test is on Friday, and the chapter is Atoms and Molecules, the one with mole concept, the one that has been sitting in her stomach like a stone since the teacher said "this will come in boards also." Her plan is sensible. She will watch one short explanation of the mole, do five questions from the back of the NCERT book, and sleep by 10:15.

She opens a video app because that is where explanations live now. The explanation is good, genuinely good: a young teacher with a whiteboard and a gift for analogies. It is eleven minutes long. When it ends, the app does what it was built to do. A second video starts. Mole concept: 5 tricks toppers use. Then a third, a little less about chemistry and a little more about a topper's morning routine. Somewhere around the fourth, the algorithm notices she lingers on short clips, and the feed quietly shifts. A dance trend. A cricket highlight. A cat that sounds like it is saying "mummy."

At 10:52 her mother calls from the kitchen. Meher looks down. The NCERT book is still open at page 32. The five questions are still undone. She doesn't feel rested, and she doesn't feel entertained. She feels robbed, and she can't name the thief.

What robbed her wasn't laziness and it wasn't a lack of ambition. Her ambition was the reason she opened the app. What robbed her was a piece of interface design, one so ordinary that most of us have stopped seeing it: a feed with no bottom.

(Meher is a composite character. But if you're a parent, a teacher or a student in India, you've probably lived her evening.)

On 3 June 2026, Google Labs announced an experiment with one of the oddest names in recent tech history: Dreambeans.

The idea is simple to describe. With your permission, the app reads what's already in your Google life (your Gmail, Calendar, Photos, YouTube and Search history) and each day turns it into a small set of illustrated stories. One might remind you that a friend's birthday is coming up and suggest a gift. Another might notice you've been searching for trekking routes and show you a trail you haven't tried. Each story comes with a custom illustration, drawn by Google's image model, of the people and places you engage with most. Tap a story to go deeper and the app pulls in information from the web to help you act on it.

That isn't what makes Dreambeans interesting for anyone who cares about how children learn. What matters is a design decision that sounds almost reckless in today's attention economy: Dreambeans ends.

Each day's collection is a bounded set. You read today's stories and then there's nothing more to read. No autoplay, no "recommended for you," no bottomless swipe. Google's own announcement frames it plainly: "The goal is not to scroll forever."

Here is the fine print, because we should be honest about what this product is and isn't:

So why write 5,000 words about an American app that no Indian student can use?

Because the most important ideas in education rarely arrive labelled "education." The printing press wasn't invented for schools. Neither was the radio, the television or the smartphone, and each one reshaped how children learn whether educators were ready or not. Dreambeans matters for the idea it carries, and that idea has a name in Hindi that every schoolchild learns before they learn the word for "attention": पूर्ण विराम (purna viram) — the full stop.

For fifteen years, the most successful apps in the world have been built on the removal of the full stop. Dreambeans puts it back. And Indian edtech, which has spent a decade borrowing the habits of entertainment apps, needs to ask what it would look like to put the full stop back into learning.

To understand what Dreambeans is pushing against, go back to 2006.

A young interface designer named Aza Raskin, working at a small Chicago company called Humanized, was annoyed by something we no longer remember was annoying: the "Next page" button. Search results came in pages. Blogs came in pages. You read to the bottom, and if you wanted more you had to decide to click.

Raskin's fix was elegant. When you reach the bottom, quietly load more. No button, no decision. He called it infinite scroll, and it was built to be helpful.

What happened next is one of the most consequential stories in modern design. Infinite scroll spread from blogs to social networks, from social networks to video apps, from video apps to news, shopping and, yes, some learning apps. The companies that adopted it did so because it measurably increased time spent, and time spent is the currency of an advertising-funded internet.

Raskin has since become one of the pattern's sharpest critics. He co-founded the Center for Humane Technology, and in interviews he has said he regrets not thinking harder about how his invention would be used. He has estimated, admittedly roughly, that infinite scroll now consumes the equivalent of around 200,000 human lifetimes every day. Critics call the number imprecise, and it is. But nobody seriously argues with the direction of it.

The insight hidden in that story is worth slowing down for. The old "Next page" button wasn't friction. It was a choice point. Every page break was a tiny moment where the reader got their agency back: Do I want more, or am I done? Remove the break and you remove the moment. Remove the moment and you remove the choice.

Psychologists have a name for these moments: stopping cues. The end of a chapter, the end of an episode, the bottom of a bowl, the last page of a newspaper. They're natural signals that let the mind close a loop and move on. A newspaper ends. A textbook chapter ends. A school period ends with a bell. For most of human history, learning came with built-in full stops.

The phone in Meher's hand doesn't have one.

Now look at what this means for a country with more young people than any other.

The adult picture. According to EY's annual media and entertainment report, released in March 2025, Indians spent an average of five hours a day on their phones in 2024, about 1.1 trillion hours in total. Nearly 70% of that time went to social media, video and gaming. By total hours, India is the largest mobile market in the world.

The family picture. Vivo's 2025 Switch Off study put average daily smartphone use at about 4.4 hours for parents and 3.5 hours for children. The study described the shape of the day: morning notifications, afternoon streaming, late-night scrolling.

The adolescent picture. ASER 2024, Pratham's nationwide rural survey of over 6.49 lakh children across 605 districts, added a digital-literacy section for 14–16-year-olds for the first time. The headline findings:

Read those last two numbers again. Rural Indian teenagers don't lack access or skill, and they are using their phones to learn: more than half of them, in a single week. But the device that carries their learning is the same device, often the same app, that carries their entertainment. And entertainment has had fifteen years and billions of dollars of engineering to perfect its hold.

The attention picture. Gloria Mark, a researcher at the University of California, Irvine, has spent two decades tracking how long people keep their attention on a single screen before switching. In the early 2000s the figure was about two and a half minutes. In her more recent studies it has fallen to around 47 seconds.

That's the world Meher studies in. She isn't uniquely distracted. She's normally distracted, inside an environment designed to make distraction the default.

In One Device, Three Children I wrote about the shared family phone, the reality for millions of Indian students, where the same handset moves from a father's WhatsApp to a sister's homework to a younger brother's games. Add the bottomless feed to that picture and you get a precise description of the problem. The study window on a shared phone is short, contested and fragile, and every minute of it is up for grabs by apps that never end.

It would be comforting to believe that learning apps are the innocent party here: well-meaning tools that just happen to live on the same phone as the entertainment. The truth is less comfortable.

Over the past decade, much of Indian edtech has quietly borrowed the entertainment playbook. Video lessons autoplay into the next video lesson. Practice questions arrive in endless streams. Dashboards proudly report "minutes watched" and "daily time spent." Investor decks celebrate engagement, a word that in the attention economy has come to mean "how long we kept them here," not "how much they understood."

There's a reason this happened, and it isn't villainy. Time is easy to measure. Learning is hard. A platform knows to the second how long a child watched a video. It has almost no idea whether that child can solve a mole-concept problem three weeks later, which is the only thing that actually matters. When you can't measure the outcome you want, you optimise the one you can, and the one you can measure is minutes.

Public platforms face the same temptation. DIKSHA's public dashboard, to its credit, publishes its numbers openly for anyone to read. As of 3 November 2026 it reports more than 615 crore learning sessions and nearly 6,969 crore learning minutes across the platform's history. Those are extraordinary figures for reach, and few countries could produce them. But the honest question every platform, public or private, has to ask is the one minutes can't answer: How much of that time turned into something a child still knows?

I raised a version of this question in The Question Machine, asking whether AI is making Indian students think better or just think less. The finite feed is the design half of the same question. A tool can be brilliant at explanation and still be built to keep a child consuming rather than finishing.

Here's the part that should change how we build. The case for the full stop isn't only about wellbeing, sleep or screen time, important as those are. Endings are how memory works.

In the 1880s, the German psychologist Hermann Ebbinghaus memorised lists of nonsense syllables and tested himself at intervals. He found that forgetting follows a curve: steep at first, then flattening. Much of what we learn fades within days unless it's revisited. Every teacher in India has watched this happen. The chapter that was "done" in July is a stranger by December.

The cure for the forgetting curve is the spacing effect: short reviews spread over time beat one long session. A large 2006 meta-analysis by Nicholas Cepeda and colleagues pulled together hundreds of experiments and found the effect strong and consistent, from children learning vocabulary to adults learning facts. Twenty focused minutes on four different days will, in most cases, beat a single eighty-minute marathon.

Notice what spacing requires: stopping. You can't space your learning if the app never lets you leave. Spacing is, quite literally, a design built from full stops.

In a well-known 2006 study, Henry Roediger and Jeffrey Karpicke had students learn a prose passage. Some re-read it repeatedly. Others read it once and then practised recalling it from memory. On a test five minutes later, the re-readers did slightly better, and they felt more confident. A week later the picture flipped dramatically. The students who had repeatedly practised recall remembered roughly 61% of the material; the repeated re-readers remembered about 40%.

This is the testing effect, or retrieval practice, and it's one of the most robust findings in cognitive science. It also exposes a trap at the heart of feed-style learning. Watching one more explanation feels like learning, the way re-reading felt to those students. Closing the app and trying to recall what you just watched feels harder, slower and less pleasant, and that is exactly why it works.

Memory consolidation happens largely during sleep. A study session that ends at 10:15 and is followed by rest is doing more work than one that bleeds into midnight scrolling. Meher's lost hour cost her more than an hour. It cost her the part of the night when her brain would have filed away what she'd learned.

Psychologists since Bluma Zeigarnik in the 1920s have observed that unfinished tasks linger in the mind and nag at us. A clear ending ("you're done for today") isn't just pleasant. It releases attention so the next thing, whether sleep, play or family, can have it fully.

Put these five findings together and something striking emerges. The infinite feed is close to the exact opposite of how human memory wants to be trained. It favours re-watching over recall, length over spacing, late nights over sleep, and open loops over closed ones. A learning experience built like Dreambeans, finite, small, daily and then done, lines up almost perfectly with a century of learning science.

The entertainment industry removed the full stop because it was bad for business. For learning, the full stop was the business all along.

Readers of this Knowledge Garden know I like to give ideas a name rooted in our own languages, from DARPAN for the mirror between physical and digital classrooms, to ANKUR for personalisation, to UTSAH for curiosity and joy. This one is VIRAAM (विराम): the pause, the rest, the full stop. It's a set of six design principles for any learning tool, classroom routine or study habit that wants to respect a child's attention instead of harvesting it.

V — Visible end. The learner should always know where the finish line is. "Question 3 of 5." "Two cards left today." A clear done screen. Dreambeans shows you a bounded set, and so should a revision session. A visible end turns study from an open-ended chore into a winnable task, and winnable tasks get started.

I — Intentional start. Every session should begin with a purpose, not a feed. "Today: mole concept, five questions." Meher's evening failed in the first minute because she opened an app that decides for you. A learning tool should ask what are we here to do? before it shows anything at all.

R — Recall before replay. Before offering another explanation, ask the learner to retrieve what they already know. Explanation is the expensive, pleasant part; recall is the cheap, slightly uncomfortable part that actually builds memory. Design the order accordingly: try first, then watch.

A — Attention budget. Size the session to the real window a child has, not the window an investor would like. On a shared family phone in a two-room home, that window may be fifteen minutes between a sibling's homework and a parent's call. Ten good minutes that finish beat forty minutes that drift.

A — Adult in the loop. Parents and teachers should see what got done, not how long the screen was on. "Completed 5 recall questions, 4 correct, mole concept needs another look on Thursday" is information a parent can act on. "47 minutes of screen time" is information that only creates anxiety.

M — Memory over minutes. Measure success by what a learner still knows a week later, not by time spent today. This one principle, adopted seriously by platforms and procurement committees, would change the incentives of the entire sector overnight.

VIRAAM isn't anti-technology. It's a request that technology be built for the way children actually learn, and children, like the rest of us, learn in loops with endings.

The good news is that you don't have to wait for Dreambeans to reach India. Many tools already in Indian classrooms can be used, or built, the VIRAAM way. The difference is often less about the tool than about how we use it.

An assignment in Google Classroom is finite by nature. It has a due date, a set of questions and a Turn in button. That button is one of the most underrated stopping cues in Indian edtech. With Gemini now built into Classroom at no extra cost for Google Workspace for Education accounts, a teacher can generate quizzes, rubrics and differentiated questions in minutes.

The VIRAAM move: use that speed to set short retrieval assignments (five questions on last week's chapter, due tomorrow) rather than long consumption tasks like "watch these three videos." The tool makes recall cheap to create. Let it.

NotebookLM has a quietly radical property: it only works with the sources you give it. Upload a chapter, your class notes and a past paper, and the AI's world is that bounded library, not the whole internet. Indian students have taken to it in a big way. Google says Indian users generated more than 3 million NotebookLM outputs (quizzes, flashcards, audio and video overviews) in a single month earlier this year, and the tools now work in multiple Indian languages.

The VIRAAM move: favour the quiz and flashcards over the podcast. Audio overviews are wonderful for a first pass, but listening is still consumption. The magic happens when a student closes the notebook and tries to answer.

DIKSHA's energized textbooks, now more than 8,000 of them, carry a QR code that opens content linked to that exact chapter. That's a bounded doorway: one page, one concept, one purpose. It's a finite design that has existed in Indian classrooms for years, and its quiet genius is that it starts from the textbook a child already holds, the idea I explored in DARPAN.

The VIRAAM move: the next frontier for every QR journey, on any platform, is a clear ending. Three recall questions and a done screen before the child returns to the page.

AI tutors that ask questions step by step, rather than handing over answers, are a genuine leap for retrieval practice. Google says India leads the world in daily Gemini use for learning. But chat has its own bottomless quality. There's always one more question to ask and one more follow-up offered. Commentators have started warning that after infinite scroll, the next frontier of attention capture may be infinite chat.

The VIRAAM move: set the goal before you open the chat ("explain mole concept, then give me three questions") and close it when the three are done.

The category where VIRAAM thinking comes most naturally is revision. Tools in this space start from a different question. Not what can we show you next? but what have you already studied that's about to slip away? VidyaLoop, an AI study and revision app built for Indian school and college students, is one example: it's organised around short loops of recall on what a student has already covered, rather than another stream of new content. Old-fashioned paper flashcards in a shoebox, sorted by "know it" and "not yet", work on the same principle and cost nothing. The format matters less than the shape: small, spaced, recall-first, and finished.

There's one more lesson from Dreambeans, and it cuts the other way. Dreambeans is finite in what it shows you, but it's built on reading a great deal of what you are: your inbox, calendar, photos and search history. For adults in the US, that's a personal choice. For Indian children, it shouldn't be a design pattern anyone copies.

India's Digital Personal Data Protection Rules, notified in November 2025, treat everyone under 18 as a child and require verifiable parental consent before platforms process a child's personal data, with obligations taking effect in phases from November 2026. They also restrict tracking, behavioural monitoring and targeted advertising aimed at children, with limited exemptions such as for educational institutions acting for defined purposes.

So the real lesson is precise: borrow Dreambeans' full stop, not its data appetite. A finite feed powered by infinite surveillance isn't a humane design. It's the same attention economy wearing a calmer face. The best learning tools for Indian children will be finite in two directions: bounded in what they show, and bounded in what they take.

Any idea worth writing 5,000 words about deserves pushback. Here are the four strongest objections to the finite feed, and where I think they land.

"Curiosity doesn't come in boxes." True, and important. Some of the best learning happens in rabbit holes: the child who starts at photosynthesis and ends up, three hours later, reading about the Great Oxygenation Event. I've argued for exactly that joy in UTSAH. But there's a difference between a rabbit hole a child chooses and a slide an algorithm engineers. VIRAAM doesn't forbid exploration. It asks that exploration begin with intention and that the child, not the feed, decide when it continues.

"Streaks and daily limits create their own anxiety." Also true. Plenty of learners have felt the guilt of a broken streak, and badly designed "finite" apps can turn a full stop into a source of pressure. The fix is gentleness: an ending that says well done, see you tomorrow, not one that says don't you dare miss tomorrow. A full stop should feel like rest, not a deadline.

"Some students need more time, not less." Absolutely. A Class 12 student three weeks before boards may rightly study for hours. Finite doesn't mean short. It means structured: many bounded sessions with breaks between them rather than one shapeless evening. The spacing research suggests that structure helps the hard-working student most of all.

"This is a nice idea for rich countries." I'd argue the opposite. The finite design matters more in India, not less. When one phone serves three children, when data packs run out mid-month, when the study window is fifteen minutes long and the power may cut at nine, a bounded session isn't a luxury. It's the only kind that reliably finishes. I made a related case in When Online Meets Offline: design for the real constraint, not the ideal one.

It's Thursday night in Indore, two days later. Meher has a sticky note on her desk in her mother's handwriting: Mole concept — 5 questions — then sleep.

She does it differently this time. She tries the five questions first and gets two right. She watches only the part of the explanation she needs, the bit about Avogadro's number, and pauses the video before it can roll into the next one. She tries the three she missed. She gets two of them. She writes the last one on a flashcard for Saturday.

It's 10:08. The phone goes face-down on the stack of books. There's nothing left to scroll because she never started scrolling. There was a goal, and then it was done.

It's a small moment. It's also, I think, the whole future of learning technology in miniature: not an app that holds a child forever, but one that knows exactly when to let her go.

Google built an app that ends and called it an experiment. For Indian education, it shouldn't stay an experiment. It should be the standard. Every learning experience we build, whether in a classroom, on a public platform or in a startup's app, deserves a पूर्ण विराम.

If this piece was useful, share it with one parent or teacher who's fighting the same battle with the bottomless feed, and subscribe to Chalk & Code for the next one in the series.

Sources