Yuxi You

You can now try Gemini 3 Pro in Figma Make! What stands out to me from early testing is how useful Gemini is at exploring visual directions + ideas that designers can then shape with their craft. As AI models get better, Figma gets better and designers can go further. I share more thoughts in the attached video, but a few quick highlights below... Gemini 3 is excellent at 0->1 generation I've never seen a model do one shots like this so consistently! It's honestly very addictive... using it in Figma Make reminds me of play Civ 5 😅 Just one more prompt... ! A wide variety of styles and aesthetics With detailed prompting, you can consider different visual directions very quickly. This creates more understanding of the option space you're working in, giving you more time to deeply explore the path you want to take. It is a highly steerable model Of course we also tested Gemini 3's ability to work with Figma design files in Make. The model translates designs with precision + offers impressive interactions. Working with Design Systems I also wanted to see how Gemini 3 could work with Design Systems in Figma, in particular with Make kits and npm packages. Again, the results were impressive. It's great for exploring options and refining details while staying on brand. New possibilities I'm especially excited to see the creative possibilities skilled, high-craft designers will explore with models like Gemini 3 + Figma Make. Additionally, these types of tools help non-designers express ideas at higher fidelity — bringing new perspectives to the design process. If you want to try Gemini 3 in Figma Make, go to "Settings" in the upper right and click "Experimental models." Then toggle it on. Have fun and please let me know if you create something cool!

2,828

83 Comments

Continuing our technical deep-dive blog series. In this blog post, Ryan and Megan share insights about AI agent evaluation and how we built our own evaluators at Cresta. Welcome any feedbacks! https://lnkd.in/gAs4cf_W

57

2 Comments

🚀 Animoca’s Yat Siu on CNBC: The Next Phase of Web3 Adoption In his recent CNBC interview, Yat Siu, Chairman of Animoca Brands, shared powerful insights on the evolving crypto ecosystem and what it means for businesses and investors: 🔹 Bitcoin vs. Ethereum ·       Bitcoin = Digital Gold: A store of value, rarely spent. ·       Ethereum = Utility: With ~$100B in total value locked, it powers stablecoins, NFTs, and L2s. Because ETH is used and spent, its rise fuels broader ecosystem growth. 🔹 The Coming Altcoin Wave As ETH strengthens (with predictions of $6K–$10K ahead), capital will cascade into altcoins and Web3 ecosystems. Corporates moving treasury allocations from BTC → ETH → other tokens show how adoption will spread across the industry. 🔹 Internet Money → Internet Customers Accepting tokens isn’t only a payment play. It’s access to a loyal, liquid, and growing community. Just like selling in different fiat currencies, offering goods in ETH, SAND, or other tokens lowers friction and builds deeper engagement. 🔹 A New Global Wealth Class and 🌍 The Big Picture Over 500 million people worldwide now own crypto and digital assets. This isn’t just about asset prices – it’s a massive new customer base. Luxury brands were early movers, integrating NFTs and Web3 strategies to reach this audience. The token economy already represents $4T+ in market cap. Every cycle of adoption redistributes value across projects, creating new opportunities for builders, brands, and investors alike. 🙌 At True Global Ventures, we share Yat’s conviction: the real story isn’t just crypto prices, but the transformation of money into a new internet-native customer economy. 👀 Watch the interview: https://lnkd.in/grXu26Em

164

20 Comments

2025 had been a wild run. What started as a dissing conversation on how stupidly the AI agents worked turned out to be one of the best ideas to work on. Back in May, I was discussing with Lu Xian on how stupidly painful it is to build an agent that works seamlessly in production. MCPs were there but nowhere easy to integrate with the agents, on top of the fact most of them were local MCPs being hosted on npm. The entire infrastructure was off and the number of gotchas you have to get through while you ship an MCP to production was crazy. That’s when we built Leanmcp (initially started as Airtrain python sdk - reaching 100k+ downloads and then TRMX hosting platform) and shipped to get over 6000 folks signing up on the first day. Past 5 months: - we had shipped Leanmcp vibe coding platform which is by far the fastest way to build, test, deploy and monitor MCP, MCP Agents, and MCP Apps. - first to ship a virtualization platform that specifically targets MCPs and enables stateful MCPs to be hosted on Lambdas. - built MCP and MCP Apps (including ChatGPT Apps) SDKs that enable enterprises to build MCP Apps to reach 100 million+ users of ChatGPT and Claude. It had been a wild journey - our SDKs reaching 200k+ downloads organically. More than 8k developers shipping 10,000+ wildly creative MCPs on Leanmcp for different use cases including workflows, Cursor for X agents, browser agents. SDKs being used by developers from Google , Meta , Salesforce , ByteDance , NVIDIA, AWS and many more. 2025 has been a fantastic journey. But I’m more excited for what we are bringing in 2026. Happy New Year!

48

3 Comments

HK isn’t behind — we’re just underestimated. The only thing missing? More collisions between high-quality founders who are actually building. Here’s what happened when that finally happened in one room. I attended a What You Building? event this week. It wasn’t just another startup gathering. It was a room full of sharp, high-intent founders and operators, asking real questions, offering real insights. Here's what stood out: Edith Yeung from Race Capital: Traction speaks louder than tech. It doesn’t matter if you’re building the next-gen AI agent — if there’s no growth, there’s no story. For seed-stage companies, over 100% YoY growth is table stakes. And remember, investors bet on you, not just your product. Nils Pihl from Auki Labs: The decentralisation of perception networks isn’t just philosophical — it’s necessary. If the way we interpret the world is controlled by too few systems, innovation stalls. Decentralisation protects diversity in human understanding and opens up new ways to build truly intelligent systems. Andrew Keh from @Aisle8.ai: Before you scale, obsess over your core customer. Serve them with such intensity that they become your biggest advocates. Don’t dilute focus early — depth creates defensibility, especially when your product is still evolving. 🍴Hacene Taibi from Bistrochat: Beating giants is possible — by going where they won’t. Hyper-specialise and build deep, defensible integrations that global players can’t replicate easily. Local expertise, niche partnerships, and cultural fluency are your competitive edge. Precision beats power when you’re in the early innings. Saptarshi Mukerji from Curiominds AI: AI and kids? Tricky combo. Balancing innovation with data privacy is crucial, especially in education tech. Parents expect safety and results. Founders building in this space must bake ethical considerations into product DNA — trust is a moat you can't afford to lose. My reflections: - High-quality founders in HK need to collide more often. The ideas, energy, and potential for synergy when builders come together is unmatched. We need more rooms where sharp minds can jam, debate, and co-create. - The HK market is not playing catch-up — it’s full of opportunities. But you’ve got to start with one painfully specific problem. Get your hands dirty, go niche, and build trust brick by brick. - Being early feels like being wrong — until it isn’t. Many HK founders are ahead of the curve. Don’t let lack of immediate validation shake your belief. Stay close to your users, iterate, and trust your conviction. Big kudos to Casey Lau, Gene Soo, Phil Ingram, Naman Tekriwal and Jason Li for curating such an energising space. Let’s keep building. (Btw they shared this slogan during the event - and I love it!)

78

15 Comments

Here’s my Just Curious conversation with Csongor Barabasi, Co-founder & CEO of Bonsai Labs, an AI studio helping PE-backed companies, especially those focused on B2B software, move from slide decks to shipped products in weeks. What impressed me is how fast they turn ideas into products and into revenue. Their process took one SaaS client from a POC to 7-figure ARR in six months, showing that you can launch profitable AI products with the data you already have, instead of waiting a year to modernize your entire stack. A few things that stuck with me: [1] Execution over strategy decks. Bonsai’s bias for action means 4–6 week POCs tied directly to EBITDA, margin, and net-new ARR, not “innovation theater.” [2] Two adoption traps to avoid. 1) Inaction. Too much talk, not enough testing. 2) “We must finish our data transformation first.” Csongor’s advice: prove ROI first, modernize data later. [3] The People–Process–Product (PPP) model. Real transformation means up-skilling teams, redesigning workflows, and building products that fit the new process, not just bolting on a model. [4] Speed with discipline. 2Weekly sprints, eval datasets, and domain-deep dives let them ship in 4–12 weeks, even in complex, regulated verticals. [5] Case study 1 – Legal AI. A legacy legal-tech firm needed to compete with new AI entrants. Bonsai built an Azure-hosted legal research assistant in 3 months (beta), 6 months (GA). Result: 7-figure ARR, fundraising unlocked, and a credible defense against AI-native competitors. [6] Case study 2 – PE-backed legacy SaaS. Facing churn risk, they launched new AI features that boosted ACV and stickiness—proof that incumbents can out-innovate startups if they iterate fast enough. Full interview below.

15

2 Comments

Got featured in a magazine! PORTFOLIO Magazine Singapore was asking some folks for their predictions in 2026, and thought I’d chime in: Q1: What industry insights and trends do you see shaping 2026, and which of these hold the greatest opportunities in your field? “Fully autonomous AI agents are the future of B2B sales, but they won't be ready by 2026. Current AI lacks the nuanced judgment needed for high-quality personalization, platform compliance, and relationship navigation. Full automation is still 2-3 years away, probably. The near-term opportunity is human-in-the-loop systems. AI can handle research and draft messaging at scale, but human operators remain essential for final review, personalization, and quality control. This hybrid approach delivers AI efficiency without sacrificing results or risking platform violations. Most companies will either over-automate too early and fail, or stay fully manual and get outpaced. The winners in 2026 will be those who architect the right balance between AI capability and human judgment.” Q2: How do you see yourself/your organization leading the way in turning these forecasts into reality? “We book B2B meetings using AI for research and drafting, with humans reviewing every message before it sends. LinkedIn bans automation, so this hybrid approach keeps accounts compliant while scaling output. Most competitors pick one extreme: full automation that gets banned, or pure manual work that doesn't scale. We built proprietary tools that make the hybrid model efficient. Our engineering team tests new AI models and adapts our tools as technology evolves - competitors using off-the-shelf software can't move this fast. Current results: Twelve clients acquired in four weeks, 30% reply rates on outreach. The tooling and engineering velocity create our competitive edge.” Thanks for the feature and the amazing picture. Can’t wait to keep building in 2026! 🚀

117

35 Comments

Cross-border friction shouldn't be the tax on global growth. For too long, moving money across borders has been slow and opaque. In my piece for The Edge Singapore, I explore why stablecoins have moved past the "hype"phase to become a critical tool for business efficiency in 2026. We are seeing a real-world shift in how companies operate: programmable money is redefining the velocity of commerce around the world. With clearer regulation and stronger infrastructure now in place, the "bottleneck" is finally disappearing: https://lnkd.in/gpRzV6Nt

156

6 Comments

Some friends asked what has been happening in the Singapore web3 space, I wrote a short summary, happy to take feedback if I missed any, and feel free to reach out if you are building something interesting along those also: ==== The biggest shift in Singapore’s Web3 landscape recently has come on the regulatory front. In May, MAS announced that any Web3 player serving overseas users must either apply for a Digital Token Service Provider (DTSP) license or exit the market by the end of June. This move addresses a key regulatory gap—while the existing MPI license framework focuses mainly on domestic activities, it didn’t cover crypto services targeting users outside Singapore, which is bring some risks to the Singapore ecosystems. We saw several unlicensed Web3 startups being forced to relocate to other crypto-friendly hubs like Hong Kong or Dubai. In some ways, this can potentially been seen as turning point for SG and MAS—from nurturing the Web3 ecosystem in its early days, to now tightening oversight as the industry matures. At the same time, institutional players are becoming increasingly open to crypto. DBS has operated its own exchange for private wealth clients for some time, and is now more actively engaging with ecosystem partners. Standard Chartered, an early mover in the UK crypto space, has also expanded its presence in Singapore—launching new services and collaborating with startups. Stablecoins are another hot area. Partior—a blockchain settlement network spun out of MAS’s Project Ubin and backed by heavyweights like Temasek, JPMorgan, and Standard Chartered—continues to push forward with institutional-grade stablecoin use cases, and started to build a ecosystem around it by enabling startups. MAS established a formal stablecoin regulatory framework in 2023, and now we're seeing practical adoption: startups like StraitsX are facilitating stablecoin settlement between Alipay users and GrabPay merchants, using SGD-backed tokens. Looking ahead, with the passing of the GENIUS Act in the U.S., it is reasonable to expect USD-based stablecoins to scale rapidly. Other areas like RWA / STO seem generating a lot of interest but personally I have followed it for quite a while, but has not seen interesting ideas. Would love referral if anyone has.

111

10 Comments