
Serial founder. Previously ByteDance & Tencent. Built AI products to $1M ARR and $10M GMV. Now building AI memory hardware at @Memoket_AI.
Joined September 2026
- Tweets111
- Following42
- Followers9
- Likes72
Pinned Tweet
Since Memoket Desktop went into beta in August, user feedback has been pretty positive. Over 90% of recordings come from our automatic recording feature, and desktop users record nearly 2 hours a day on average. That’s almost a quarter of the workday. That’s wild. Why does everyone have so many meetings?
It reminded me of something my colleague @098uyiubgvgcss mentioned: peace of mind.
I wonder if that’s part of why people record. You don’t want to miss any details. When you need something, you can go back and listen—or, when you’re in a hurry, find the exact moment you’re looking for. There’s a sense of security in that.
That’s exactly what we want people to feel with Memoket: your data is safe, your memories are safe, and you have something to rely on when you need to think things through. That peace of mind means having less to carry. If meetings are an unavoidable part of your work, you should be able to enjoy them and get involved without worrying about what you might miss.
Recording is just our first step. What would you like us to do next to make meetings less stressful? Help you revisit past conversations? Give you reminders in real time? Or is there something else that’s been bothering you for ages?
3
122
a few things I've learned as a founder:
1. users want the outcome, not the AI. don't build a product just to show off the model.
2. users want it to work, not look pretty. make it useful first, then make it beautiful. not the other way around.
3. the fastest way to get someone to pay isn't "look what I can do." it's "here's what I can do for you."
4. if you can read and move how someone feels, you're already halfway there. the other half is the product.
7
a little bit about myself:
- graduated 6 years ago
- built products at ByteDance and Tencent
- helped take Trae from 0 to 1, to a million users
- later led an AI product, hit $1M ARR in 6 months
- now building AI memory hardware
- founder of several companies
- 10 months, $10M GMV
- still early, but I'm just getting started.
ask me anything.
17
The logic is wild, but it's not even new. People want to look good, all that effort turns into electricity, and that electricity powers the machines that replaced them. The loop is almost too perfect. That's been in sci-fi for years, and it already feels like the future.
17
Even when you're busy, you still need rest. An AI recorder isn't here to pile more pressure on you. This kind of hardware is supposed to take weight off, not add more. If a product just makes life harder, it probably shouldn't exist.
2
44
The strange part of living with an AI recorder: you start depending on the hardware. Walk into a meeting without it and you feel the pressure to stay perfectly focused. Online, a bot can join for you. Offline, the recorder shows up in your place. The roles flip. The device enters the world. You're just its Stand.
30
We don't just want you to pick Memoket. We want you to see how this space is changing. We'll keep following AI notetakers and AI recording hardware, and hope to lead the way one day. Got a great idea or product? Send it our way.
teapaw.org/Binyu63/status/2122673…
🎁 An AI note-taker for Christmas? We compared wearables and apps by who they’re best for and what they’ll cost over the first year.
Read the gift guide ↓
memoket.ai/blogs/news/best-a…
1
31
Building Gem taught me a trade-off I didn't expect: every gram you save in an AI wearable costs you battery or mic quality. We landed at ~11g, with 20 hours of continuous recording and 30 days on standby. It took dozens of versions to get there. What would you give up first: weight, battery, or recording quality?
2
3
43
Building an AI memory wearable taught me something unexpected: the hard part isn't recording, it's deciding what's worth remembering. This week we cut 5 hours of raw audio down to 12 moments you'd actually want to revisit. Still tuning it. What would you want your memory device to forget?
20
"Attention is all you need" isn't just true for models. It's true for every product.
Strip product building down far enough and it's one thing: earning human attention. I think every product has to do it three times.
1. Find it: positioning
A pain point is where people's attention already is. You don't create it, you find it. Solve it and you earn a seat at the table.
But a seat is not a win. Plenty of products solve real problems and still disappear.
2. Win it: brand
Now you have to get seen. App store rankings, shelf space, a push in an infinite feed. Every channel is a fight for a few seconds of someone's eyes, against everyone else who also solved a real problem.
3. Keep it: experience
This is the one most people get wrong. Keeping attention doesn't mean demanding more of it. It means people can hand it to you and stop worrying.
Two things make that happen.
First, you learn it once. The first use teaches you everything. Today, next month, a year from now, it works the same way. No relearning, no new menus to figure out. That one moment of attention keeps paying off.
Second, peace of mind. When people trust a product to just work, they stop checking on it. That trust is where their attention settles.
Easy to use. Built to last. That's how attention stays.
Building @Memoket_AI, this is the bar we hold ourselves to. An AI recorder you have to babysit has already lost. The best one is learned in a minute, trusted for years, and there when you need it.
Which of the three is hardest in your product?
1
2
30
One sells features. The other sells years.
One lists every mode, screen and smart function. The other basically says: this will last you 20 years.
What do people actually want from a product? 🧵
1
23
Same for software and hardware. Find the one or two things users truly come for. Make them simple and direct.
Same way today, next month, next year. Zero relearning.
Easy to use. Built to last.
1
3
This is what we're wrestling with at
@Memoket_AI right now, as we redesign our desktop app.
What's the first principle of an AI recording product? Capturing everything reliably? Finding it later? Something else?
Come think it through with us. Replies and DMs open 👇
7
Makes total sense for internal team brainstorms—having an AI build docs live while you talk is great.
But most real conversations can't have a bot sitting in the call: client calls, 1:1s, hallway chats. That's why we're focused on botless capture on desktop and a lightweight wearable for everything off-screen.
Different problems.
Introducing Claude Meetings: add Claude to your team’s video call and it builds docs, whiteboards, and other artifacts with you while you’re still talking.
Starting to roll out in beta on Team and Enterprise plans.
29
A CEO leaving doesn't change what builders need: cheaper inference, better models, more real use cases. If anything, it's a reminder that the AI wave is bigger than any one person. Back to shipping.
The best home for a personal agent isn't your phone. Your phone already carries your life, work, and every notification. A nudge from your agent there is just more noise. A true personal agent needs its own device that does two things: listen, and act.
2
58
When everyone who seems to be at the forefront of the industry is telling the same grand story, who actually wins—and what makes the difference?
1
15
If you wear your GEM for hours at a time, you’ll probably like this one from Memoket. Just keep recording—we’ll take care of splitting it up for you.
8 hours at a trade show. One continuous recording. 40+ conversations.
Good luck finding the one where that distributor mentioned pricing.
That's why we built Topics in Memoket (beta): it splits a long recording into the topics you actually covered, each with its own summary and the exact transcript lines. Jump straight to the part you need.
2
25









