What a Hands-On AI Note-Taker Review Actually Found

A hands-on reviewer just spent a year testing nearly every AI note-taker and pendant on the market: Plaud, Pocket, Bee, Omi, Fieldy, Soundcore Work, HiDock, Friend, Looki. Not a sponsored list, not a press-release roundup. He bought the hardware himself, wore it into real meetings and bars with friends, and tracked what actually broke.

The video is Best AI Note Takers — 2026 from Jasper Tech (330,000+ views). We watched it in full. A few of the findings line up closely with things we've written about separately, and a few are useful context for anyone shopping this category, so here's what stood out.

His five-question framework

Instead of specs, he judges every device on the same five things:

1. File transfer. How easily does audio get off the device and onto your phone? He found some devices are "pretty darn bad" at this.
2. App and ecosystem. Is there a real desktop app, meeting integrations, a usable mobile app?
3. Trust. How much do you trust the manufacturer with your recordings, and will they still support the product next year?
4. Data lock-in. Once you've built up a library of notes, how hard is it to get them out and into another system?
5. Cost. Upfront hardware plus the ongoing subscription, since "every time you transcribe or summarize a meeting, someone somewhere is getting a bill."

Worth noting: the first and fourth of those, file transfer and data lock-in, are both problems that grow out of the device keeping an audio file. A device that turns speech into text and keeps no audio in your account doesn't build up an audio library to move or get locked into. That's not a knock on his framework, it's built for the category as it exists today, where recording is still the default.

What he found testing each device

His pick for best overall is Plaud, mainly for the most reliable file transfers and a genuine developer ecosystem (Zapier integration, auto-email, an active community). He's upfront that it's on the expensive side, and that his own audience has been pointing him toward an independent open-source project called AudioBridge that lets you bring your own API keys for cheaper processing.

Pocket AI Recorder impressed him on hardware, metal build, built-in magnet, no case needed, but he flagged real reliability problems getting recordings off the device, plus a plan change that briefly cut their free tier from 90 days of access down to 14 before user backlash reversed it. His conclusion: promising, not yet recommendable.

Soundcore Work has solid hardware but reuses Soundcore's headphone app, so there's no AI model choice and a broken Notion integration. He calls it a "V1 product."

The always-listening "second brain" category (Friend, Limitless, Omi, Bee, Fieldy) got the harshest notes. Friend is "just terrible." Looki adds a camera and, in his words, made people stare at him like he was "wearing a police body cam." Limitless is gone entirely, pulled from the market after Meta acquired the company. Omi is open-source and ambitious (screen recording, AI glasses, more) but he found it buggy and unreliable across three separate units, calling it "peanut buttering," a team spread too thin across too many features. Bee impressed him at first, but he reports that since Amazon acquired it, the hardware has failed, support has gone quiet, and the app has gotten buggier, and he's watching for Amazon to monetize it the way it has Alexa. Fieldy, by staying narrow (conversations, tasks, search, nothing else), was the one he found most reliable, with the cleanest transcripts in his testing.

The part that matters most: his "fundamental problems" section

About 15 minutes in, he stops ranking devices and makes a broader point: these products are "tragically flawed" as an always-on second brain, independent of which one you buy. Two reasons:

They can't reliably tell who said what. He describes getting a task generated from audio the device picked up from a stranger talking nearby, not from him or anyone he was with. He calls this "context rot": transcription errors compound, names and words come out wrong, and the burden lands on the user to catch and fix it. "If you put raw garbage into the AI system, you're going to get garbage out."

Consent is a real, unresolved problem. He estimates roughly 40% of the US population lives in a state requiring all-party consent to record a private conversation, before even accounting for Europe's rules. And beyond the legal question, he's direct that "it's just not cool to record people in private without their consent," and that asking for it in the moment is genuinely awkward, which undercuts the pitch that these devices are seamless.

That second point is the same territory we wrote about separately in Consent and AI Wearables. An independent reviewer testing devices that keep audio files arrived at the same conclusion from a different angle: keeping no audio in your account changes the trust and file-custody questions, but it does not make the consent question go away. Ask people before you wear one around them, regardless of what the device does with the audio afterward.

Where this leaves Legend

Legend wasn't part of his test, and we're not claiming it would have solved everything he ran into. But two of his five evaluation criteria, file transfer and data lock-in, are specifically about moving audio files around, and Legend handles that differently. Legend turns your conversations into text. Audio is used only to create the transcript and isn't kept in your account. If your pendant is away from your phone, audio is held on the pendant until it syncs, then deleted. There's no audio to play back, and no audio archive to get locked into. His trust criterion (will the company still support this next year, will they use your data to advertise to you) and his consent point apply to every device in the category, Legend included, and we've written honestly about both: how we handle data and what we ask wearers to do around other people.

Where he thinks the category is headed

His predictions, for what they're worth from someone who's tested this much hardware: an explosion of Plaud-style competitors built on commoditized hardware and off-the-shelf AI models, more ring-based entrants (Sandbar, Pebble, Fable), a new self-improvement-tracking subcategory (Bee's daily insights feature, dedicated products like Nerva and Nuna), cameras moving into the category most obviously through smart glasses (Meta's Ray-Ban Display line, after acquiring Limitless), and big players circling: OpenAI reportedly working on an audio device after acquiring Jony Ive's hardware startup, and Apple rumored to be building an AI pendant after its own $1.5 billion-plus acquisition of a startup working on reading speech from facial micro-movements.

FAQ

Was Legend included in this review? No. This is a video from an independent reviewer, Jasper Tech, covering Plaud, Pocket, Bee, Omi, Fieldy, Soundcore Work, HiDock, Friend and Looki. We're summarizing his findings because parts of it are directly useful to anyone comparing devices in this category, not claiming he reviewed Legend.

Does keeping no audio in your account solve the consent problem he raises? No, and we don't claim it does. It changes the file-custody and data lock-in questions, not the question of whether the people around you agreed to be captured. See Consent and AI Wearables.

What did he pick as the best overall device? Plaud, mainly on file transfer reliability and its developer ecosystem, while noting it's the more expensive option and that data lock-in is a real tradeoff.

Is Limitless still available? No, per his reporting it was taken off the market after Meta acquired the company.

Source: Best AI Note Takers — 2026, Jasper Tech, published March 2026.