Moses Xu

225 posts

Moses Xu

Moses Xu

@mosesxu

ML PhD, Startup Founder, AI Builder, Investor

Sydney Katılım Mart 2011
217 Takip Edilen54 Takipçiler
Philipp Schmid
Philipp Schmid@_philschmid·
I asked @GoogleDeepMind Gemini 3.1 Pro watch the launch video of @cursor_ai SDK and create a production script. Then tasked it to re-create the video 1:1 with @Remotion 0-shot. Video Understanding capabilities.🔥Original cursor video in answer thread.
English
12
10
244
23.8K
Moses Xu
Moses Xu@mosesxu·
everyone's optimising for autonomous agent handoff. but co-steer wins for some work. @danshipper's posts on writing in proof + browsing posthog inside codex name it — essay drafting + EDA need shared cursor + live state, not "agent goes off, comes back."
English
0
0
1
32
Moses Xu
Moses Xu@mosesxu·
@danshipper's other posts make this concrete: x.com/danshipper/sta… x.com/danshipper/sta… some deliverables fit co-steer better than the async handoff → review → refine loop. writing essays, exploring data — you want shared cursor + live state, not "agent goes off, comes back."
Dan Shipper 📧@danshipper

im writing an essay in proof using codex's in-app browser. i type directly in the document, and codex loops in parallel on the left, collaborating with me in realtime. this is obviously the future:

English
0
0
1
81
BradWMorris
BradWMorris@bradwmorris·
not sure im following this what's the difference here between agents doing browser/computer use and meeting you in your apps/browser? have you used the new codex computer use? would much prefer a model-agnostic workflow here, but maybe im missing the benefits of having a co-pilot style experience?
English
1
0
0
784
Dan Shipper 📧
Dan Shipper 📧@danshipper·
we might need to coin a new term: Codex-native Cowork-native Cursor-native apps designed to be used inside of the in-app browser of your agent of choice. the agent can use it and you can use it too, full context is shared, and you can both see what the other is doing. huge opportunity to build software here
English
10
2
63
35.1K
Moses Xu
Moses Xu@mosesxu·
@dabit3 or you can install Amphetamine
English
1
0
3
251
Moses Xu
Moses Xu@mosesxu·
@emollick CTO friend at a 500-person tech co. 6 months ago: anxiety mode. now: built agents, spent a month getting his CEO/COO off terminal phobia. adoption swept through once the c-suite became super-individuals themselves. top-down is the real leverage
English
0
0
0
375
Ethan Mollick
Ethan Mollick@emollick·
One change over the last six months is that in every big company I talk, at least a few senior people absolutely get AI — they experiment (a lot of OpenClaw, surprisingly) & they have an intuitive sense of the exponential curve — next challenge is translating that to the firm.
English
48
29
479
33.8K
Moses Xu
Moses Xu@mosesxu·
@MatthewBerman i've been a massive mech keyboard nerd but now, i really do think typeless is the best "keyboard" there is
English
0
0
2
604
Moses Xu
Moses Xu@mosesxu·
@BentoBoiNFT tbh the $20 CC entry was already borderline unusable — 10 min of real work and you'd hit the limit. but it was the only way to sample CC before committing to Max. that onramp is what actually gets lost, not the daily workflow
English
1
0
0
75
BentoBoi
BentoBoi@BentoBoiNFT·
Anthropic just cooked the average AI user. • Claude Code now removed from $20 plan. • OpenClaw & third party tools banned from subs. Now you need $100-200/m to do what $20 used to cover Meanwhile: Codex: $20/m Gemini: Free Qwen & Kimi: Open source Claude is taxing their users.
BentoBoi tweet media
English
24
6
41
7.6K
Moses Xu
Moses Xu@mosesxu·
@chalaska here. aussie builder density on X is thin — most still live on linkedin. good for deal flow, quite dead for the AI-builder conversation actually happening here and information lag is crazy. more of us should move
English
0
0
1
104
Chris Halaska
Chris Halaska@chalaska·
Genuinely curious, how many Aussie designers, founders and builders are here on X? Wanting to connect with you all!
English
467
13
780
55.6K
Moses Xu
Moses Xu@mosesxu·
one version of what personal software actually looks like. bespoke, worry-free, no longer needing "general applicability" to justify existing. could still be broadly useful, doesn't have to be. that's the whole freedom.
@levelsio@levelsio

Now just 1.5 hours into this mini project I have replaced ALL my Chrome extensions with my own vibe coded ones for safety After another one of my extensions after updating suddenly wanted to read my entire browser history (probably to sell off to an ad company) My own Chrome extension is called 🚀 SuperLevels and now has: 🗑️ Tab Cleaner, auto closes tabs after inactivity (with host-based exclusions) 🍪 Cookie Editor, lets you nuke all cookies or edit any 🔀 Redirect Tracer, lets you see any redirects a site took you through 🌗 Dark Mode, changes websites to dark, per site or all sites 🎚️ X Dim, changes X background back to dark blue 🎶 Music Finder, records music in a tab and submits to find the name of the song 🗺️ Add back a Maps link to Google in EU (they hide it!) 🌠 Add back a View Image link to Google Images in EU (they hide it too!) I've deleted all my other extensions (except uBlock Origin for ad blocking!), much much safer because now I control the source code!

English
0
0
0
55
Moses Xu
Moses Xu@mosesxu·
"personal software looks like openclaw" is the tell. SaaS survives but the targeted user flips from human-in-browser to agent-in-API, and per-seat pricing breaks the second that happens. the Salesforce-clone vibe coders are building logins nobody logs into.
Dan Shipper 📧@danshipper

counterintuitive things I believe about AI: SaaS is not dead, it will be stronger than ever The userbase for SaaS will 10x over the next 3 years because agents will become users Knowledge work will change dramatically, but everyone will still have jobs (except very specific categories like personal injury insurance for car crashes) Humans will continue to be the sandwich at the beginning and end of every AI process (h/t @kieranklaassen and @trevin who came up with this) Personal software looks like OpenClaw, not a vibe-coded Salesforce clone You'll use Codex and Claude Cowork through your SaaS software, and use your SaaS software through Codex and Claude Code Because you'll use Codex with your SaaS, all software users will start to look like extremely technical users All SaaS will need to incorporate the idea of 'presence' for agents—making what agents are doing legible in real-time to users Agents are currently built for 1-1 interactions, they'll need to be built for one to many and many to many interactions Specialization will be just as important in agents as it is in humans, therefore we will live in a many-agent world Most software companies are not building for a world where everyone has an agent, and they're wasting a ton of time and capital developing capabilities that a user with Claude Code doesn't need

English
0
0
1
60
Moses Xu
Moses Xu@mosesxu·
context management hygiene 101：先把内置命令用明白，再谈别的。简单朴素的东西往往最管用。
中文
0
0
0
20
anul agarwal
anul agarwal@anulagarwal·
pitch me your SaaS in 2 words let's see how creative everybody is
English
65
0
40
5.6K
Moses Xu
Moses Xu@mosesxu·
@steveruizok indie AI pin -- what does it take to set it up? $ and effort wise?
English
1
0
4
997
Steve Ruiz
Steve Ruiz@steveruizok·
this is a [deep breath] M5 Stick S3 acting as a frontend for Gemini Flash 3.1 Live, running via a CloudFlare worker, with tools to set device settings (brightness, volume, power), web fetch / search, and vector search (of tldraw docs) via cf's Vectorize. crazy low latency, push to talk on this hardware is great too
English
24
23
430
38K
Moses Xu
Moses Xu@mosesxu·
just read this and it got me thinking. a GPU is basically seed corn. you can sell the harvest (inference) or save it to plant next season (training), but it's the same grain. AI labs are deciding how much seed corn to eat right now. that tension doesn't resolve until compute stops being scarce, which isn't soon.
Ethan Mollick@emollick

Compute constraints are a double bind: On the inference side you need to either (a) raise prices, (b) ration use, and/or (c) serve worse models. This hurts current growth On the training side, you can't train the next gen of models to stay competitive. This hurts future growth

English
0
0
1
29
Moses Xu
Moses Xu@mosesxu·
@emollick GPUs are seed corn. sell the harvest now (inference) or plant next season (training), but it's the same grain
English
0
0
0
46
Ethan Mollick
Ethan Mollick@emollick·
Compute constraints are a double bind: On the inference side you need to either (a) raise prices, (b) ration use, and/or (c) serve worse models. This hurts current growth On the training side, you can't train the next gen of models to stay competitive. This hurts future growth
English
37
10
191
20.6K
Moses Xu
Moses Xu@mosesxu·
haven't thought about overloading in the OOP sense in a while, but that's exactly what happened to "agent." same name, completely different implementations. this article finally gives it a type signature: three categories, and 60-70% of what people call "agents" are just workflows. taxonomy alone is worth the read.
Lenny Rachitsky@lennysan

The word "agent" is the most overloaded term in AI right now. Your backlog probably has 5-10 agent ideas. But one agent idea is a 6-week automation you can build with n8n. Another is a 6-month bet requiring a dedicated ML team. Putting them on the same spreadsheet and hoping impact-vs.-effort will sort it out doesn't work. Hamza Farooq and Jaya Rajwani—who teach two of the most highly rated and well-respected courses on building AI agents—spent 50+ hours putting together a guide that'll help you make sense of the different categories of "agents," how to prioritize across them, and how to avoid common pitfalls—with recommended tools and many real-world examples. Read this post the next time your team is going in circles on your agent strategy: lennysnewsletter.com/p/not-all-ai-a…

English
0
0
1
40
Moses Xu
Moses Xu@mosesxu·
@emollick based on the Anthropic I know, hyping the market like this is uncharacteristic. they've always been the cautious ones. if they're saying Mythos warrants new safety measures I'd take that at face value before assuming it's marketing. can't say the same for other labs
English
2
0
7
363
Ethan Mollick
Ethan Mollick@emollick·
I am catching glimpses in my feed that there is a backlash against Mythos as "marketing hype," and it is a little confusing. I don't think anyone who has used the latest agentic coding tools, would think that expecting large-scale cybersecurity implications of increasingly good AI models is unbelievable, especially after reading the red team reports. It feels like a better place to start is to assume that there are new risks, and then we can all laugh at Anthropic and pat each other on the back if there are not. Also, while the AI labs certainly are impressed by their own accomplishments and benchmarks are flawed, I would note that both publicly and privately, Mythos seems to be taken seriously at a lot of large institutions and organizations filled with smart people who would rather not be worried about a new cybersecurity risk. Finally, I am not sure "our product is dangerous and we need to alert the government to that" is the sales pitch to the corporate world that critics seem to think it is.
English
90
45
687
62.5K
Moses Xu
Moses Xu@mosesxu·
intelligence as an external tool you can grab off the shelf is genuinely new. AI is hard-bent on aligning with us, but we're the malleable ones here. we just learn the sharp corners and adapt around them, same as we do with anyone we work with
Ethan Mollick@emollick

Things that make the jagged intelligence of AI harder to deal with than the jaggedness of humans: 1) Weaknesses are not always intuitive or identifiable in advanced 2) All LLMs have similar weaknesses, so you can't just hire a different one 3) Jagged frontier is moving outward

English
0
0
1
64

Keşfet

@N0t_Andreas @_philschmid @GoogleDeepMind @cursor_ai @Remotion @danshipper @dabit3 @emollick