Jiayao Yu

@veryfarnaz @smart @cognition Great, love to exchange notes at some point. We use Devin quite a bit for cross repo changes and autonomous workflows.

@smart @cognition @jiayao Steve is coming for you!

I’m in the top 200 @cognition / Devin users. At $400m+ ARR,the enterprise contracts must be large and concentrated. I’m a huge fan of the pace of innovation, the multi repo support, automations, Devin review, wiki tools, computer use and now it can run and test android apps. The pace we can move at with @joinsuppco is unmatched so far. Check it out.

Cancer is cured by AI. GitLab founder Sid Sijbrandij was diagnosed with stage 4 spinal cancer. Every trial rejected him. His doctors had nothing left to offer. So he stopped being a patient. He built an AI research team. Fed them 25TB of his own medical data genomics, scans, treatment history, everything. The system found a treatment his entire oncology team had missed. Then engineered 19 custom vaccines from his own DNA. Relapse-free since 2025. Then he uploaded the entire blueprint. Free. For every person sitting in that same room, hearing the same verdict, with nobody left to call. Medicine runs on averages. AI runs on you.

@lennysan Staff PM overseeing AI product experience and platform @Linktree_ jobs.gem.com/linktree/am9ic…

Who’s hiring PMs right now? Reply with the role, company, location.

@lennysan AI has much higher capacity of attention. They can research alternatives much more deeply. Product builders just need to show real proof of value efficiently.

As AI eats more of product building (autocompleting code -> writing 100% of code -> reviewing its own code -> deciding what to build in the first place -> judgement to know what is great), where will human brains be most needed moving forward? (Until AGI 😵‍💫) I think distribution becomes the biggest challenge/opportunity. Getting people's attention in the ever-louder market. This is good news for incumbents and people with a platform. Tough for new startups. You'll have to get ever-more creative to break through. And then...what happens when the users choosing products are also AI?

@xdrudis It was very smooth in the beginning, but I do notice it starts to slow down with large number of tasks, taking a couple of seconds to start a conversation.

@lennysan @openclaw Collect and summarize answers from cool X threads like this.

Do you use @openclaw in some super impactful or fun way? I'd love to know. Please share your 1-2 favorite use cases in the comments. If it's awesome, I'll feature you in the newsletter. Big bonus points for screenshots of what it looks like in action (and any tips for setting it up).

@cpaik Love the analogy. X is a bubble full of people who are so sensitive to model quality, like the video engineers that easily tell the difference with 8k TV.

Kimi K2.5 being on par with Opus 4.5 is a critical moment. If 4K TVs were free, what demand would there be for 8K+ ones?

Made a mini app on @wabi to help you decide if you need a Mac mini to set up Clawdbot @jiayao/do-you-need-a-1024439" target="_blank" rel="nofollow noopener">wabi.ai/@jiayao/do-you… (You probably don’t)

@GoogleResearch Cool technique of LLM fine tuning

Today we announce a new approach to user intent understanding for mobile devices. By decomposing trajectories into individual screen summaries, our small models achieve results comparable to large models at a fraction of the cost. Learn more: goo.gle/3LEZOqX #GenerativeAI #EMNLP2025 #MachineLearning

@BlasMoros On discovery: Now we can create apps as easily as writing a tweet. Most just enact a thought, remove a small friction or fulfill a simple purpose for ourselves.​​​​​​​​​​​​​​​​ We need to stop thinking about them all as products waiting to be discovered.

wilfutw (what i learned from users this week) creation unlocks more creation. creating the first 1–3 mini-apps is the trickiest. after that, behavior shifts and you start seeing that every hunch, idea, problem, or curiosity can come to life as a mini-app. once you feel that agency, the world looks different and we couldn’t be more excited for what that unlocks. twitter gamifies communication; wabi gamifies creation. on twitter, ideas spread because the medium makes expression effortless. wabi does the same for apps - software, play, and creativity all become lightweight, fast, and social. apps at the speed of culture. people start building the moment a thought appears and remixing what others have built, expanding the surface area of possibility. the long tail is way longer than expected. once friction collapses, people start solving the tiniest, most personal problems: micro-utilities, niche workflows, “i wish this existed” apps. the long tail was always there - it was just invisible until creation became easy. it’s never been that people lack ideas; it’s that every step from “huh, i wish this existed” to fully functioning app is too hard and we at wabi are here to dissolve those frictions. discovery needs to catch up. when the long tail explodes, finding great apps and great creators becomes harder. we added search recently - an obvious step in the right direction - but we have a lot more to build to make discovery fun, personal, and alive. “wabi gives people instantaneous solutions to their little problems.” one of my favorite quotes from the week. we don’t use ai for the sake of ai - we use it to make life slightly easier, more delightful, more in your control. last thing: we’re teaming up with creators. if you missed the early days of youtube, tiktok, or the app store… this is your chance to be early on the next platform. want in? dm me.

@rauchg @vercel Snapshot is cool enhancement. Make it much easier to build agents with skills.

The future of computing is trillions of digital AI workers. The @vercel Sandbox is how: cloud computers purpose-built for agents. But computers have state, like Git repos (and petabytes of 𝚗𝚘𝚍𝚎_𝚖𝚘𝚍𝚞𝚕𝚎𝚜 😅). You can now clone & launch millions of agents with 1 API:

@benhylak claude —high-judgement

i wish there was a claude --less-dangerously-skip-most-permissions

@wabi create a mini-app that turns brainstorming into a turn-based strategy game. Users start with a central question, then explore a small set of “next moves” suggested by AI—expanding, constraining, inverting, combining, or taking a bold wildcard. Each move creates a new branch on a visual board so they can see how choices reshape the space. They discard weak lines, explore promising strategies, and gradually arrive at a strong position they feel confident in.

@janinaparker_ Get button not working

@jiayao what’s wrong? seems to work fine for me.

why is a pond full of koi always so calming? built with @wabi 🫧

@eshear This is why play usually precedes mastery.

Learning new things requires vastly more available degrees of freedom than making use the stuff you’ve already learned does. The more general the knowledge, the more true this becomes.

@gilgNYC @Shopify @Replit @Claude We’ve stopped using gpt-4 class models a while ago. Just curious if there is some unique capability of gpt-4o-mini that I missed.

@jiayao @Shopify @Replit @Claude What would you have used? I haven't built the eval framework yet so it may be a poor choice but I use the OpenAI playground to compare performance, cost, and accuracy.

So I built a Google Lens mobile app using just the @Shopify Catalog API (publicly available), GPT-5, @Replit and @Claude code. It’s insane how this was just built by telling it what I wanted. Here’s a demo. Should I open source? All APIs are publicly available. The app works like this: 1. Ask Replit to generate initial working app prototype. 2. Bring into Claude Code and iterate using screenshots and high level vision. The app works by: - user takes a photo or selects one from camera roll - crop photo to maximize focus area - send to GPT-5-mini to extract tags - send extracted product tags to gpt-4o-mini to generate search query - send search query to Shopify Catalog API Basically your camera becomes a shopping search bar.

@BlasMoros @pmarca @wabi Built a few mini apps this weekend. The surprising part wasn’t how easy it was. It was realizing how many ideas I’d previously dismissed for not being “useful enough.” When the cost is this close to zero, utility stops being the filter.

WILFUTW (what i learned from users this week) - Twitter got a new toy last week: after @pmarca kicked off a flywheel with us, anyone can tag @wabi with an app idea and get it built. at this stage, one of our biggest jobs is educating and inspiring people on what’s possible - and this flywheel does that in spades. every idea is public, there’s social proof, engagement, feedback, and community. you can find your tribe around a problem. it feels like early YouTube: public learning, public building, a new playground forming in real time. cracking 0→1 distribution is fun :) - lowering the friction to create shows how creative people really are: there’s a common trope that the average person isn’t creative. bullshit. everyone has a unique vantage point and a set of problems they want solved - they just never get to it because the friction is too high. As wabi makes creation easy, ideas pour out: crazy ones, useful ones, niche ones, deeply personal ones. many will look small or even stupid at first - but time and again, what starts as a toy evolves into something much bigger. our core focus is to systematically crush every point of friction from “huh, I wish this existed” to a fully functioning app, so anyone, anywhere, can bring any idea to life. - that’s why we’re starting in the skeuomorphic era of personal mini-apps. “this app but free.” “this app but no ads.” “this app but psychedelic colors.” it’s the natural starting point - the value is obvious - but it’s only the beginning. what’s more exciting is what comes next: software as self-expression, apps as content, disposable apps (as @illscience calls them), alongside highly retentive daily tools. every new platform creates a new content format and a new type of creator - and wabi is doing that with software storytellers. - people want to play: getting featured twice on trending news isn’t hype - it’s proof people are ready to play. people don’t just want better apps; they want a new creative medium and the ones who crack this new format will be this generation’s youtubers. - the magic moment is building your first app. once someone creates their first mini-app, they get it. oh… i can solve my own problems now. that idea i’ve wished existed for years can be real in minutes. wabi becomes a new way for me and my friends to just do shit we like together. - clarity and control are the next unlock: one of the most common pieces of feedback is that it’s not yet clear what you can do on wabi today. that’s on us - and the team is cooking. soon you’ll have much more visibility into capabilities, integrations, and building blocks, making it faster and more fun to bring any idea to life. - last thing: we’re teaming up with creators. if you missed the early days of youtube, tiktok, or the app store… this is your chance to be early on the next platform. want in? dm me.

@janinaparker_ @wabi Vibe coding at your desk is so 2025.

always be @wabi-ing 🫧