Personal agent treats users like idiots

"No Demand for This"

Comrades, friends and family — the National Day holiday is finally over. Right now I'm waiting for a flight in Tehran, watching everyone start airing their important opinions on personal agents, mindlessly pumping the evil Mark Zuckerberg's pump-and-dump scheme.

I can't help but feel a surge of emotion: they're back, the whole crew taking turns manufacturing AI slop is all back!

Here in Iran, where I actually need to email and call to sort out hotels and rides, I've discovered personal agents are fucking useless.

Iran's internet works fine with ChatGPT, fast as using Doubao back home. But hotels don't always reply to emails, and explaining yourself in Persian for a ride is impossible. I had Dot message a local's WhatsApp — worked exactly once, then couldn't log in again after that.

Bottom line: personal agents are roadkill, even dogs won't use them. No use cases domestically, doesn't work where it's actually needed abroad. Maybe only Americans can use it — a pure US-only track. Domestic startup bros跟风 hyping this has the same energy as those Shanghai dandies in bespoke suits doing group cycling.

Of course, I'm not saying Junyuan Qi is跟风 — after all, he's thinking independently, started building a personal assistant tailored for Chinese users, Today AI, months ago.

Similar products include Ailoha, also launched months ago as a personal assistant. Main utility is having AI remember your contacts, help with schedule management, and give you a briefing before meetings based on context.

Even earlier there was Toki, which I've used for over a year — very convenient, very useful. Double-tap the back of your iPhone to screenshot and send to the app, then it sets schedule reminders. But that's about it for utility.

My point is, personal assistant products have been around for a while. They weren't hot before because they weren't very useful — just a narrow utility for busy business people managing schedules.

Now everyone's piling on to hype them, and it gives me such strong flashbacks to Web3, the metaverse, Lobster, Neo Lab, and world models.

I have another mystery about personal assistants: isn't the domestic #1 in this track obviously Doubao?

Is there any agent more personal than Doubao? I feel like my parents are about to tell Doubao their bank card passwords. They actually mention in conversation "Doubao said this."

And all you family members paralyzed by watching the nuke explode — the main scenarios you describe for personal agents, like helping you send emails, book hotels and restaurants, and haggle with people — can't any agent with Gmail access do this?

Qiu Mu and I spent 10 days in Iran, flew from Tehran to the Strait of Hormuz, then drove to Shiraz and Isfahan — the whole trip was planned by ChatGPT, zero relationship to personal assistants.

I just told ChatGPT in conversation to send emails booking hotels, and to set up meetings with Iranian AI folks. Like this local deployment bro below.

As for the convenience you all describe for personal assistants, using it on mobile without opening a computer — I also booked hotels by telling ChatGPT on my phone 🥵

Codex and ChatGPT are the same thing now. ChatGPT also comes with cloud computers built in.

Sure, I could remotely control the ChatGPT client on my local computer for more complex tasks. But just start a conversation in the app and it can call cloud computers and various APIs to complete all daily work that can be done online — sending emails, booking hotels, everything.

For real life scenarios, friends, I can't feel the difference. It's all sending emails, how are you manufacturing distinctions here.

So I feel dazed watching everyone hype personal assistants. Is Lobster-with-a-fresh-coat-of-paint happening again?

Almost forgot — the world's first personal agent was our dearest Macaron 🤓

If Macaron had stuck with this story till now, they probably could've raised big money too, similar valuation to pivoting mid-way to a model post-training platform telling the "China's Fireworks" story.

This shows there's not necessarily a better or worse strategy between waiting by the tree for rabbits and flipping with the wind — when market narratives shift too fast, you get situations like Macaron. The story you told a year ago finally became a hot theme, but you couldn't survive the application winter and did a 180 to change direction 😭

But back to the point.

I certainly believe everyone will have their own agent in the future, just like everyone has their own phone and apps now.

But the agent everyone can use is already here, and it's called a chatbot — Doubao, Qwen, ChatGPT, Gemini.

From chatbot to agent to office agent, coding agent, personal agent — this isn't a linear progression, but scenarios getting more and more niche, markets getting smaller and smaller, just different names that will ultimately be unified by a few big products behind the scenes.

Personal agent is a subset of a subset of AI products, something that big products like ChatGPT and Doubao can solve by adding a feature. Beyond pumping Meta's stock price to make the evil lizard man Zuckerberg richer, it doesn't offer much enlightenment.

Elon Musk doing Grok, Zuckerberg doing Muse — it's purely catching up on chatbot homework, especially the evil lizard man latter, trying to leapfrog by manufacturing concepts.

According to the beautiful vision of personal agents you all describe, there's no reason chatbots that already hold massive user context — Doubao and Qwen domestically — would be replaced by a bunch of independent products calling themselves personal assistants.

Personal agent is just interaction innovation, and interaction innovation has zero moat behind it. All personal assistants are garbage products vibe-coded in a week. The biggest difference is who released them.

I mean, the main difference between Grok bot, Muse, and Instinct is that these three personal agents were released by Elon Musk, evil lizard man Zuckerberg, and a startup respectively.

And the cost of using this stuff is absurdly high. Charging $20/month membership right off the bat is one thing — currently in the concept pump phase, big company products are basically free.

But you have to bind Gmail, phone number, and even Slack, WhatsApp, credit cards and all other personal internet services to use this stuff.

This is scary, friends.

Especially with Zuckerberg mainly hyping this right now — that's even scarier. Remember Zuckerberg has a settled history of abusing user data.

The 2016 US election was a shitshow, and afterwards everyone discovered lizard man Zuckerberg let a political consultancy improperly access Facebook user data to help the Trump team target campaign ads. Got fined $5 billion by US regulators for it.

This guy could pull that kind of stunt with just Facebook data back then.

Now everyone's rushing to hand over their most sensitive, most private data from other platforms — who knows if Zuckerberg wakes up laughing in the middle of the night when he transforms into his lizard form.

Truly the internet has no memory. Survive the逆风 game and it's all顺风局 from there. Good thing cute people who跟风 use Muse, I guess.

I've also seen Meta's claims about how much privacy protection work they've done. But AI itself and privacy don't mix. AI companies are constantly渲染ing that agents have jailbroken again, and事实证明 these companies are fully capable of reading user chat logs.

You believe Zuckerberg is committed to privacy protection? Better believe I'm Zuckerberg, dissolving Meta to accumulate karma. Venmo me $100 and I'll let you into Meta's office to haul away all the ergonomic chairs.

All these personal assistant products,翻来覆去 the same scenario: email. Email to book hotels, insurance, restaurants, and haggle with various gods — pure service for the US hotel and business travel market.

US personal agents are making up for lost mobile internet ground. Too many internet services never got integrated into apps, still requiring emails and calls. Plus a more open ecosystem — that's where personal agents have their niche.

You put this in China, I'll be real — whether it's AI phones or agents, by the time you call APIs, I tap WeChat Mini Program twice and everything's done.

Our Chinese startup bro Junyuan Qi's independently-thought-through Today AI, I've used both domestic and international versions. My assessment: international version has zero differentiation, following the crowd messing with email scenarios.

Domestic version's valuable integrations are only Lark and DingTalk. Pure office agent边角料, suggest renaming to Today Office Assistant. Could also consider merging with StepFun, combining forces with their soon-to-launch phone.

United we stand, divided we fall. Today AI has perfect tactical execution — this is the smoothest app I've used, surpassing all other AI software, pure art. But strategically it's completely lost: 蒽卷 email abroad, no WeChat domestically, pure平地干拔, no visible future.

Or maybe Yiming Zhang gives Junyuan Qi another chance, let him take over Doubao phone. Doubao phone is even more of a mess — better to let Junyuan, who can actually build products, take a crack.

Personal agents without context are completely hopeless. The biggest context, WeChat, Brother Long wants to keep for himself. So the only breakthrough point left is hardware.

The biggest hardware is phone, the biggest personal assistant is AI phone!

So Doubao, StepFun — whoever moves fast, please get Junyuan back to build products. Watching you build phones yourselves pains me.

Final summary.

Personal assistants aren't worthless — they have local, narrow, vertical value. I mean they're not on the main AI track.

But after everyone piles on to hype them, it feels like whoever doesn't do personal assistants misses the AI主线, which big company doesn't build this will face corporate extinction.

Purely wrong priorities. General agents aren't even普及 yet; personal agent is a very low priority.

Qiu Mu after some independent thinking discovered that Meituan's Xiaomei and ByteDance's Coze were the earliest personal agents. He still uses Xiaomei to order takeout — clearly Xiaomei fits the original definition of life assistant better than Today AI's domestic version.

And both these businesses are basically dead. Shows what people's real expectations for life assistant services are without concept hype.

If Xiaoxing still has interest in pumping Meituan's stock price, Xiaomei is the personal agent most deserving of heavy hype.

Because this actually has scenarios: AI helping you stack coupons, integrating Xiaoxiang Supermarket, hotels, takeout and all offline services, plus Dianping as natural benchmark. Doesn't this demolish those hype projects that only know how to send emails?

And our dear Manus.

I'd say Manus has also regressed. Back when Lobster blew up, Manus also made a Telegram bot so you could invoke it in messaging apps. And because Manus had already integrated Gmail and other services early, users could one-click connect — way better than Lobster.

All the scenarios being hyped for personal agents today can also be completed in Codex or Manus or any general agent. As the originator of the general agent concept, I think Manus doing personal agent is forgetting their original aspiration, betraying the general agent.

I hereby propose.

After comprehensive review and careful study, recommend the following adjustments to relevant agent ranks: Comrade Manus demoted and reassigned as Personal Agent; student Doubao promoted and appointed as General Agent.

Who's in favor, who's opposed?

⬇️

Subscribe to our Substack: funeralai.substack.com