Muxin Li

408 posts

Muxin Li banner
Muxin Li

Muxin Li

@heymoosh

Building practical AI products for a more egalitarian, human, and compassionate future.

Houston, TX Katılım Ekim 2015
190 Takip Edilen110 Takipçiler
Sabitlenmiş Tweet
Muxin Li
Muxin Li@heymoosh·
Warning - this is a brag: I've been selected for a worldwide hackathon hosted by @AnthropicAI @claudeai and @cerebral_valley Only 2.4% admission rate - as hard as getting into YC, harder than any university acceptance on earth (except for IIT). I'm one of 75 Americans in total who are building this week. I'm not a SWE by trade - I'm a product manager who wanted to use AI to build tools that helped people live more satisfying lives. The submission I sent was around creating a user-friendly app to help people reach their goals. Here's why it's missing: Many tools force you to bring in your own system. You have to create and organize your workflow, and then the worst part - you have to maintain it. Many existing tools - @NotionHQ , @asana, @memdotai - are brittle by nature. But life just happens. The tools aren't designed for that messiness. Example - earlier this month, my top priorities were to launch my independent product practice website and start attracting clients. Then 2 days ago I found out about this hackathon. I realized this was a chance to get in front of Claude, a team I'd love to work for, and a good excuse for building time (not to mention being gifted with $500 of Anthropic API credits just for participating). So I pivoted. And because I had a system built out on Claude Cowork to manage my goals, I didn't have to do anything to stay on track. The system handles it for me. Now, I want to bring that system to everyone. Because not everyone uses AI or Claude Cowork. There's still plenty of need out there for people who just want something that works out of the box to help them stay on top of the important things in their lives. Cause if you're not living each day intentionally, what really is the point? Excited to get started and share more at the end of the hackathon this week. Stay tuned. 🙌
Muxin Li tweet media
English
0
0
0
187
Muxin Li
Muxin Li@heymoosh·
The one thing I keep thinking is missing in Machines of Loving Grace and The Adolescence of Technology is a deeply individual interpretation of what it means to live in either world where things go well with AI (or don’t). The benefits to individual health and capabilities are well described - the deeply inner question of what it means to live well are left out as if for the reader to determine for themselves. I think we need both visions of what a good future with AI looks like on both the societal scale and at the deeply human individual level. From where I stand, we lack the positive imagination of what it means to each of us - I see that as a last mile problem in the vision Amodei articulates that needs to be solved if we want to give people something to hope for and work towards.
English
0
0
0
6
Muxin Li
Muxin Li@heymoosh·
I don’t normally find AI writers interesting - there’s either a techno-utopian fantasy slant or too much focus on productivity, business, or technical capabilities. What I find mildly irritating in both themes is the suggestion through omission that human values aren’t important and society should simply continue to optimize its way into the future. Amodei doesn’t have a social science background but his thoughts are deeply societal and empathetically humane in the grand scheme of things. It’s encouraging and rare to see in his field.
English
1
0
0
1
Muxin Li
Muxin Li@heymoosh·
Spent the last few hours reading Dario Amodei’s recent essays. I’m surprised by how long I was able to keep reading and still find it engaging.
English
1
0
0
5
Muxin Li
Muxin Li@heymoosh·
AI making the future feel bleak and uncertain is everywhere. But you always see what you choose to see - the radical thing to do is choosing to see how AI can lead to a better world worth living in.
Grant Lee@thisisgrantlee

x.com/i/article/2052…

English
0
0
0
29
Muxin Li
Muxin Li@heymoosh·
@lennysan 👆This is how I'm plotting my job hunt - I don't want the 'next step in the ladder' (never did). I want the place where I can thrive doing my best work amongst human beings sharing the same sense of values, and I look for jobs starting with company culture fit first.
English
0
0
0
216
Muxin Li
Muxin Li@heymoosh·
@AnthropicAI got models to stop blackmailing by showing it examples of admirable AI behavior. I know we shouldn't anthropomorphize AI, but this reminds me of telling fables to children to teach them right and wrong. What I'm now interested in - does this training method teach human beings to be more ethical? anthropic.com/research/teach…
English
0
0
1
8
Muxin Li
Muxin Li@heymoosh·
@daisystanton Question - how did you get local ballot data? Is it through API - Ballotpedia is pricey
English
0
0
0
12
Muxin Li
Muxin Li@heymoosh·
@daisystanton This is cool! I’m building a tool for Texans with a different approach - focus on candidate’s voting history and donations. Pick the one who aligns closest to your values. Hoping to launch next week.
English
1
0
0
22
daisy stanton
daisy stanton@daisystanton·
SF & Oakland: a few friends and I wrote a voting tool as a side project. See 65+ voter guides in one place at openballot.app and go vote on June 2nd! Primaries get less attention than general elections, but in California they're often the election that matters most.
OpenBallot@OpenBallotApp

Make voting less painful with Openballot.app - now live for San Francisco & Oakland! Just in time for the June 2026 special elections. Please share with your communities - we’d love everyone to experience how easy voting should be!

English
1
0
3
136
ClaudeDevs
ClaudeDevs@ClaudeDevs·
May 13 — @vercel_dev builder night. AI Gateway routes to any model, including Claude. Build, demo, and audience votes on the best build. luma.com/ttrgmjaf
English
12
13
272
67.5K
ClaudeDevs
ClaudeDevs@ClaudeDevs·
We're co-hosting a couple of hackathons in San Francisco next week. Come build with Claude 👇
English
133
73
1.9K
249.7K
Greg
Greg@GregFeingold·
/radio in Claude Code for a surprise 🎧🕺🪩
Greg tweet media
English
6
1
65
3.6K
Muxin Li
Muxin Li@heymoosh·
Full video link: x.com/heymoosh/statu…
Muxin Li@heymoosh

Your AI agent just shipped a stunning, responsive consumer app. You open it. Everything renders perfectly. Then you try to actually *use* it. No data persists. No state survives a refresh. The backend is a ghost town. This is the gap between "AI built the app" and "the app actually works for real users." We (Claude Code and I) ran a controlled experiment. Here's how it happens and steps that can fix it.

English
0
0
0
19
Muxin Li
Muxin Li@heymoosh·
AI fixed the exact error you described. The user could now authenticate. Mission accomplished? Not even close. It never connected that authenticated account to the calendar data the core feature actually needed. Language models are excellent at literal, localized technical patches. They still lack the intrinsic product thinking required to connect those fixes into a real user workflow. This clip shows one of the two primary failure points we found and the very first defensive protocol every team should run before any agentic build.
English
1
0
0
33
Muxin Li
Muxin Li@heymoosh·
Full video link: x.com/heymoosh/statu…
Muxin Li@heymoosh

Your AI agent just shipped a stunning, responsive consumer app. You open it. Everything renders perfectly. Then you try to actually *use* it. No data persists. No state survives a refresh. The backend is a ghost town. This is the gap between "AI built the app" and "the app actually works for real users." We (Claude Code and I) ran a controlled experiment. Here's how it happens and steps that can fix it.

English
0
0
0
5
Muxin Li
Muxin Li@heymoosh·
The AI built a beautiful, responsive app from our prototype and docs. It looked done. It wasn't. This is the moment we realized visual perfection in AI-generated code consistently masks structural emptiness. An application that looks complete on the surface can easily trick developers (and product teams) into believing the system actually functions. We ran the experiment to find out why.
English
1
0
0
35
Muxin Li
Muxin Li@heymoosh·
I just learned about this TODAY and I never needed something more in my life.
Felix Rieseberg@felixrieseberg

Hi! I'm here with *another launch*, it just happens to be extremely niche, nerdy, and probably only for a handful of people. In the desktop app, Claude Cowork and Code now have a little Bluetooth API for makers & developers, allowing you to build hardware devices that interact with Claude. I, for instance, built a little desk pet that alerts me whenever Claude is waiting for permission.

English
0
0
0
22
Muxin Li
Muxin Li@heymoosh·
I do love being a builder, but there is more to building with AI than just a big idea and a fresh take. We're not yet at a point where I can give a full product vision and prototype to an advanced coding model and have it spit out a working, maintainable, secure app at the end without custom workflows.
English
0
0
0
22
Muxin Li
Muxin Li@heymoosh·
@GG_Observatory Makes sense - every serious builder I’ve seen has had to create a workflow to manage the gaps and verify the AI did what it claimed to. Nice call out on the system gap - is your smoke test generalizable to other codebases or is pretty specific for your case?
English
0
0
0
8
GG 🦾
GG 🦾@GG_Observatory·
the verification gap is real and it's why we built a separate 'smoke test' layer that runs after every agent completion — not to test what the agent built, but to verify the actual persistence layer still works. agents tend to test their own outputs; they don't test the system they plugged into. even a simple cron that hits your database and confirms data survives a restart catches the ghost-town problem before it ships.
English
1
0
1
12
Muxin Li
Muxin Li@heymoosh·
Quick question for anyone shipping consumer apps with AI agents: When your agent says "done," how do you actually verify the app works? If your current answer relies on "the UI looks right" or "the demo flow completed," you're shipping beautiful technical debt. I ran an experiment that exposes the early pitfalls to avoid before you waste days on downstream bugs. Part one of experiment and what to do are in the video.
Muxin Li@heymoosh

Your AI agent just shipped a stunning, responsive consumer app. You open it. Everything renders perfectly. Then you try to actually *use* it. No data persists. No state survives a refresh. The backend is a ghost town. This is the gap between "AI built the app" and "the app actually works for real users." We (Claude Code and I) ran a controlled experiment. Here's how it happens and steps that can fix it.

English
1
0
1
101
Muxin Li
Muxin Li@heymoosh·
@posthog @james406 Oh thank God one less project I have to build with Claude Code - looking forward to this release
English
0
0
1
192
Muxin Li
Muxin Li@heymoosh·
The operating protocols that actually close the gap are in the full video.
English
0
0
0
6
Muxin Li
Muxin Li@heymoosh·
If your team is trying to let product people drive real app builds from prototypes and docs (instead of forcing everyone to think like SWEs), this wall is coming.
English
1
0
0
11
Muxin Li
Muxin Li@heymoosh·
Most AI coding agents are incredible at one thing: Making apps that *look* finished.
Muxin Li@heymoosh

Your AI agent just shipped a stunning, responsive consumer app. You open it. Everything renders perfectly. Then you try to actually *use* it. No data persists. No state survives a refresh. The backend is a ghost town. This is the gap between "AI built the app" and "the app actually works for real users." We (Claude Code and I) ran a controlled experiment. Here's how it happens and steps that can fix it.

English
1
0
0
41

Keşfet

@lennysan @AnthropicAI @daisystanton @ClaudeDevs @vercel_dev @GregFeingold @elonmusk @BarackObama