Big Tech Just Bought Your Memory: AI Wearables, Lifelogging, and Perfect Recall

The Year Big Tech Bought Memory

For most of 2024 and early 2025, the AI wearable space looked like a scrappy startup race. A handful of companies made small recorders and pendants that promised to capture your conversations, transcribe them, and let you search your own life. The pitch was seductive: never forget a name, a promise, an idea you had in the shower.

Then the consolidation hit, and it hit hard.

On July 22, 2025, Amazon acquired Bee, the company behind a $49 always-listening wristband and pendant. By CES 2026 in January, Bee was being shown off as part of Amazon's Alexa and Echo roadmap, which means the always-on microphone you wear is headed toward the same ecosystem as the always-on microphone in your kitchen.

On December 5, 2025, Meta acquired Limitless, maker of the Limitless Pendant that had become the darling of the productivity crowd. The Pendant is no longer sold to new customers. Meta didn't buy it to keep selling a $99 gadget. It bought the team, the data pipeline, and the head start.

The acquisition came with a casualty. Two weeks later, on December 19, 2025, Limitless shut off desktop capture in Rewind, its original product, as part of winding it down under Meta. Rewind had taken a different approach: instead of a wearable, it recorded everything on your screen and made it searchable. Existing recordings stay readable for about a year, then they're gone. A product that promised to be your perfect memory now has an expiration date on the memories it already holds.

In the span of about five months, the field went from a competitive startup scene to something close to a duopoly of platform giants plus a single independent holdout. That holdout is Plaud, and we'll get to it. But the bigger story isn't which gadget won. It's that the two companies with the most detailed profiles of human behavior on the planet just decided that recording your spoken life was worth buying.

The Four Pendants: A Field Guide

Before the philosophy, the hardware. Here's where the four most discussed devices stand as of June 2026.

A few things stand out.

Limitless was arguably the best-loved of the bunch among knowledge workers, which is exactly why Meta bought it. Its disappearance from the new-customer market is the clearest signal that the independent premium pendant is an endangered species.

Bee competed on price. At $49 it was the cheapest way into always-on capture, and Amazon's acquisition makes sense as a hardware on-ramp to Alexa's broader ambient computing play. The privacy tradeoff is the one you'd expect from any always-listening device tied to a retail and advertising giant.

Plaud is the independent survivor and the most interesting case. The lineup runs from the original Plaud Note to the NotePin ($159) to the current NotePin S ($179), which reviewers consistently rate as the best standalone AI recorder available. Crucially, Plaud's default is press-to-record, not always-on. Its subscription scales from 300 free minutes a month up to a $239.99-a-year tier. It's a recorder for meetings and intentional sessions, not an ambient surveillance device, and that design choice matters more than it looks.

Friend is the outlier and the lightning rod. At $99, it's an always-on "AI companion" pendant designed less for productivity and more for, well, company. It listens constantly and talks back. It generated a wave of criticism for normalizing a wearable microphone whose entire purpose is to be in every moment of your social life. Whatever you think of it, Friend makes the always-on premise impossible to ignore.

The pattern across all four: the always-on devices got bought or stayed cheap-and-ambient, and the one that asks you to press a button to capture is the independent that reviewers respect most. That's not a coincidence. It maps onto a much older question about what capture is for.

Lifelogging Is Older Than the Pendant

The dream of recording your whole life didn't start with a venture-funded pendant. It's a decades-old idea with a clear lineage, and the lineage is worth knowing because it shows how often this dream has run into the same wall.

In 1945, Vannevar Bush described the "memex," a hypothetical device that would store all of a person's books, records, and communications and let them retrieve any of it through associative trails. It's the founding fantasy of personal information management: a prosthetic memory you could query.

In the early 2000s, Microsoft researcher Gordon Bell ran a project called MyLifeBits, an actual attempt to digitally capture everything from his life: documents, photos, phone calls, web pages, even a wearable camera that snapped images automatically. Bell's conclusion after years of living it was telling. Capturing was easy. Making the archive useful was the hard part. An ocean of recorded data isn't the same as memory, because memory is about retrieval and meaning, not storage.

The AI wearable wave is the memex with two new ingredients: cheap always-on microphones and large language models that can transcribe, summarize, and answer questions about the capture. The hardware finally caught up to the fantasy. What hasn't changed is Bell's lesson. The bottleneck was never recording. It was turning the recording into something your mind can use.

This is the same tension that runs through the broader consumer AI memory landscape, where every assistant now wants to remember your context. The pendant is just the most literal version: a microphone that treats your entire spoken life as input. Worth asking before you clip one on: what does the system you're augmenting actually need?

Why Your Brain Forgets on Purpose

Here's the premise the entire pendant industry quietly assumes is true: forgetting is a malfunction, and a perfect record is therefore an unambiguous improvement. The neuroscience says the opposite.

In 2017, Blake Richards and Paul Frankland published "The Persistence and Transience of Memory" in Neuron. Their argument cuts against the intuition that memory should aim for perfect retention. They make the case that transience, the active process of forgetting, is not a failure of memory but a feature of it. Forgetting serves cognition in at least two ways.

First, it promotes good decision-making by clearing out outdated and irrelevant detail. If you remembered the exact contents of every conversation, every parking spot, every version of a plan that later changed, retrieving the version that actually matters now would get harder, not easier. Forgetting prunes the noise so the signal stays findable.

Second, forgetting supports generalization. When the brain lets go of specifics, it keeps the gist, the pattern that transfers to new situations. A person who remembers the precise wording of every example but not the underlying rule has stored data, not understanding. Transience is part of how the brain abstracts.

There's a clinical reminder of what perfect memory actually looks like. People with highly superior autobiographical memory (HSAM) can recall the details of nearly every day of their lives. It is not, by their own accounts, an unmixed gift. Some describe being unable to let go of painful events, reliving them with the same vividness years later. The reason most of us forget the texture of an ordinary Tuesday is that forgetting it is healthy.

So when a pendant promises to remember everything you said and heard, it's promising to override a system that evolved to forget on purpose. That might be useful for specific tasks. As a general upgrade to how you think, it's solving a problem your brain doesn't have. The interesting question isn't "what if you could remember everything?" It's "what does it cost to never decide what's worth keeping?"

The Testing Effect: Recording Isn't Remembering

Even if you set aside the value of forgetting, there's a second problem with the perfect-recall pitch. A recording you can search is not a memory you possess. And the difference between them is one of the most robust findings in the science of learning.

In 2006, Henry Roediger and Jeffrey Karpicke published "Test-Enhanced Learning" in Psychological Science, building on a long tradition of research into what's called the testing effect, or retrieval practice. The setup was simple. Students studied material, then either restudied it or were tested on it. The restudy group felt more confident. The tested group remembered far more on a delayed final test a week later.

The mechanism is that retrieving information from memory, effortfully, without looking, strengthens the memory in a way that passively re-encountering it does not. Every act of recall is itself a learning event. The struggle to pull something up is not wasted time. It's the mechanism.

Now apply that to a life log. A pendant captures a conversation, transcribes it, and files it away. At no point do you retrieve anything. The archive does the holding so you don't have to. Which means that by the logic of the testing effect, the pendant doesn't help you remember the conversation. It guarantees you won't, because you never practice recall. You've offloaded the exact cognitive act that would have made the memory yours.

This is the same trap that shows up in reading with AI, where a summary feels like comprehension but leaves no residue, and it connects to the broader research on cognitive offloading. When you know a machine is holding the information, you encode it less deeply. Sparrow's "Google effect" research showed this in 2011, and a wearable that records your spoken life is the Google effect aimed at your entire day.

Contrast that with intentional capture. When you highlight a passage, take a note, or write a two-sentence summary, you're doing something the pendant can't do for you: deciding what matters and re-expressing it. That selection and effort is encoding. It's the reason a book you highlighted sticks while a podcast you half-listened to evaporates. The act of choosing what to capture is itself an act of thinking, and it's precisely the act that a capture-everything device removes.

The pendant optimizes for completeness. Learning optimizes for effortful selection. Those are not the same goal, and a tool built for one will quietly undermine the other.

Who Owns Your Life Log Now

Set the cognitive questions aside for a moment, because there's a colder one. If a device records your conversations, where does that recording live, and who controls it?

For two of the four leading pendants, the answer changed in 2025. Your Limitless data sits under Meta. Your Bee data sits under Amazon. These are the two companies whose business models are most directly tied to knowing what people do, say, and want. The ambient audio of your meetings, your dinners, your offhand remarks is now flowing into infrastructure owned by an advertising and commerce giant respectively.

Here's the timeline of how that happened.

The Rewind shutdown is the part everyone clipping on a pendant should study. Rewind sold itself as a perfect memory. Then the company changed direction, switched off the capture, and put a roughly one-year clock on the recordings users had already made. The "perfect memory" had a corporate off switch, and the users didn't hold it.

That's the structural problem with renting your memory from a company. Your recall is only as durable as that company's business plan. A product pivot, an acquisition, a shutdown, a change in subscription terms, and the archive you trusted to hold your life can become read-only, then unreadable. You experienced the events. The record of them belongs to someone else.

There's a desktop parallel worth naming. Microsoft's Recall feature, which periodically screenshots your PC to make it searchable, raised the same concentration-of-data alarm when it was announced: a single, continuously updated, highly detailed record of everything you do, sitting in one place, is a target and a liability whether or not it's ever misused. The wearable version just follows you out the door.

The defensive questions are concrete. Can you export your data in a usable, portable format? If the company is acquired or shuts down, what happens to your archive, and on what timeline? Is the capture default-on or default-off? Who can subpoena, sell, or train on it? For ambient always-on devices owned by platform giants, the honest answers range from "unclear" to "not in your favor." For a tool where you trigger the capture and can export it freely, the answers get a lot better. Ownership is not a footnote here. It's the whole game.

Passive Capture vs Intentional Capture: A Framework

None of this means recording is bad. It means recording is a tool with a narrow set of genuinely good uses and a wide set of seductive bad ones. The useful distinction is between passive capture, where a device records ambiently and you sort it out later, and intentional capture, where you deliberately mark what matters in the moment.

Here's when each one actually wins.

The line is clean once you see it. Passive capture is for situations where the record is the deliverable and your own memory of it is beside the point: a meeting transcript you'll reference, a doctor's appointment you need to recall accurately, a real-time accommodation for hearing. In those cases a pendant or recorder earns its place, and Plaud's press-to-record model fits that use far better than an always-on companion does.

Intentional capture is for everything you actually want to learn. The friction is the feature. When you stop to highlight a sentence, you're doing three things a microphone never does: deciding it matters, isolating it from the noise, and giving your brain a retrieval hook for later. That's why people remember the books they marked up and forget the audiobooks they ran in the background. This is the heart of the science of highlighting, and it's the part no capture-everything device can do for you.

Treat the pendant like the calculator it is. Great for the arithmetic you don't need to internalize. Useless, or worse, for the math you're trying to learn. The mistake is reaching for ambient capture as the default for everything, including the cases where the effort you're trying to skip is the entire point.

Highlighting as the Memory Layer You Own

If the cognitive argument lands, the practical question follows: what does intentional capture look like as an actual system, not just a virtue?

This is where highlighting earns its keep. When you read something and mark the passages that strike you, you're building a memory layer made of decisions. Every highlight is a small act of selection, the same selection your brain uses to decide what's worth keeping. Unlike a pendant's transcript, a highlight is already filtered through your judgment. It's signal you chose, not noise you'll have to sort later.

Glasp is built around this premise. Glasp's web highlighter lets you mark passages as you read across the web, and those highlights persist, stay searchable, and accumulate into a library of things you actually engaged with. That's the difference from a life log: it's not everything you encountered, it's everything you decided mattered. The encoding happened at capture time, because you did the choosing.

The same pattern extends across formats. YouTube Summary lets you pull the moments that count out of a long video instead of letting the whole thing wash over you, turning passive watching into deliberate capture. Kindle highlights carry the passages you marked while reading into the same searchable library, so the books you read deeply stay with you instead of fading after the last page.

And here's the ownership point that ties back to the acquisition story. Your highlights are exportable. They're a layer you control, not a recording held on a platform giant's servers under terms that can change with the next acquisition. When the question is "what happens to my memory if the company pivots," the answer for an exportable highlight library is simply: nothing, you take it with you. That's a categorically different relationship to your own memory than renting recall from Meta or Amazon.

The contrast with the pendant is the whole argument in miniature. The pendant captures everything and you own none of it. Intentional highlighting captures what matters and you own all of it. One optimizes for the company's data pipeline. The other optimizes for your understanding. For more on building a system around context you control, see personal context management, and for the voice-capture angle specifically, voice AI note-taking covers where ambient recording genuinely fits.

Frequently Asked Questions

Are AI memory pendants like Limitless and Bee worth buying in 2026?

It depends entirely on what you want them for. For capturing meetings you have to act on, or for accessibility, a recorder is genuinely useful, and the Plaud NotePin S ($179) is the best-rated standalone option right now. For learning, research, or anything you want to actually remember, a pendant works against you, because it removes the retrieval effort that builds memory. Also note that the Limitless Pendant is no longer sold to new customers after Meta's December 2025 acquisition, and Bee is now an Amazon product folding into Alexa.

What happened to Limitless, Bee, and Rewind?

All three changed hands or shut down in 2025. Amazon acquired Bee on July 22, 2025, and showed it at CES 2026 as part of the Alexa and Echo roadmap. Meta acquired Limitless on December 5, 2025, and the Limitless Pendant is no longer sold to new customers. As part of that same deal, Limitless wound down Rewind, its desktop capture product, turning recording off on December 19, 2025, with existing recordings readable for about a year. Plaud is the main independent company still selling AI recorders.

Does recording everything actually help you remember it?

No, and this is the central misconception. Research on the testing effect (Roediger and Karpicke, 2006) shows that retrieving information from memory is what makes it durable, not storing or re-encountering it. A searchable archive you never actively recall from doesn't build memory, it replaces it. The effort of intentional capture, like highlighting or summarizing in your own words, is the part that makes something stick. Passive recording skips exactly that step.

Isn't perfect memory always better than forgetting?

The neuroscience says no. Richards and Frankland's 2017 paper in Neuron argues that forgetting (transience) serves cognition by clearing outdated detail and supporting generalization, the ability to extract patterns rather than store every specific. People with highly superior autobiographical memory often describe it as a burden, not a gift. A device that never forgets is overriding a system that evolved to forget for good reasons.

Who owns the data from an AI wearable?

For the leading always-on pendants, increasingly a platform giant. Limitless data sits under Meta, Bee data under Amazon, as of their 2025 acquisitions. The Rewind shutdown showed the risk clearly: a company can change direction and put an expiration date on recordings you already made. Before buying any capture device, check whether you can export your data in a portable format and what happens to your archive if the product is discontinued. Tools where you control and can export your captures, like a highlight library, avoid this problem entirely.

Conclusion

The pendant pitch is built on a confusion: that capturing your life is the same as remembering it, and that remembering everything is better than forgetting on purpose. Both halves are wrong. Your brain forgets by design, retrieval is what makes memory durable, and a recording you never recall from teaches you nothing.

The 2025 consolidation made the stakes plainer. When Meta owns one leading pendant and Amazon owns another, the ambient audio of your life becomes their asset, governed by their terms, with the off switch in their hands. Rewind already showed how that ends. Plaud's survival as a press-to-record independent is a hint that the market itself values intentional capture more than the always-on version, even if the headlines went to the acquisitions.

The useful move isn't rejecting capture. It's matching the mode to the job. Record the meeting you have to act on. Use a recorder for accessibility. But for the reading, the research, and the ideas you actually want to think with, capture intentionally, because the effort of choosing what matters is the part that becomes yours.

That's what highlighting is for. Glasp's web highlighter and Kindle highlights build a memory layer out of your own decisions, searchable, durable, and exportable, which means you own it no matter who buys whom next. Mark a passage today that you want to remember a year from now. Then come back to it. That act of returning, of pulling it up again, is the one thing no pendant will ever do for you, and it's the only thing that turns capture into memory.