Suyu Zhang

Terminal Use (YC W26): Vercel for background agents. Terminal Use is hosting infrastructure for background agents that makes harness engineering easy. https://lnkd.in/gv99wN2t

51

2 Comments

The reality is that every engineer is becoming a tech lead with a crew of agents. In that world, the primary job shifts from writing code to coordinating work: tasks, context, handoffs, and verification. That likely calls for a new kind of UI, one that feels closer to Kanban, Jira, or Asana than a traditional IDE. We don’t know yet if Intent is the final winning approach, but I’m excited we’re shipping a research preview instead of standing still. PS: What’s one small, recurring pain in your daily workflow? Drop it in the comments. If it’s a good fit, I’ll build a prototype with Intent and share what we learn.

7

I can finally say that I have been working on a deep learning convolutional neural network (CNN) model for audio classification and the detection of gunshots in a split-second using an app running 24/7 on our phones. The real power comes from multiple users running the app so that warnings can be shared instantly to lead the user away from the danger. The classification is done in real-time on our phones, so no network connectivity is needed and there is no audio data sent out to maintain your privacy. The model was trained and optimized using a database of over 21,000 real world sounds, and with some very special (patent-pending) techniques to discern gunshots in real-world environments. I am happy to say that the model is working extremely well. If this app can provide alerts and early warnings or notifications to first responders to save even one life, it will all be worth it.

19

2 Comments

🏜️ Alpha Roundup #1 — Shotgun rides Shotgun is a local-first CLI that turns rough ideas into buildable specs—PRDs, user stories, architecture—and exports them to Cursor/Claude/Lovable. Howdy, partners! A quick gallop through what landed this week: • GPT-5 ready - Sharper specs on the first pass. • Exports, no yak-shaving - /export claude, /export cursor, /export lovable, /export markdown. • Spec Engine upgrades - Cleaner PRD structure, tighter user stories & AC, clearer architecture/flows. • Smoother first ride - Polished install & quickstart flow, first-run checks, “what to try” path. • UX improvements - snappier webview, reliable Ctrl+O, clearer logs/progress. Try it: follow the install instructions & join our Discord → links in comments 🤠

20

2 Comments

🚀 Built to Ship – Hosted by Tyler Postle, Co-Founder @ Voker (YC S24) 🎙️ Featuring - Garvit Chouhan, CEO @ 9AI In this episode, we dive into how 9AI is pioneering a new category: intelligent AI employees. After shifting from offering AI services to building ready-to-deploy AI team members, they’re now helping businesses scale faster with smarter, integrated workflows.

21

1 Comment

I’ve been designing directly inside Cursor. Here at ztgrowth, we are an AI-native company, and that changes how we think about design. Generative AI is far better at producing code than traditional design artifacts. So instead of polishing static screens, we iterate through real interactive mocks in code. The biggest shift is that design is now iterative and conversational. It feels a bit like that classic scene of a stakeholder standing behind the designer, asking for changes over their shoulder. I prompt Cursor to work with data mocks as part of the design process. Creating different states is just one prompt away. Empty states, edge cases, messy real-world data. Exploring them early dramatically reduces surprises later. We iterate until the experience feels right. Every aspect of the interaction, not just static UI snapshots. When it feels right, we start building the feature itself, replacing mocks with real data. In the past, I have seen many important details left out of design artifacts and eventually decided by a developer just trying to get things done: - Input masks and keyboard types. - Responsiveness beyond two fixed frames for desktop and mobile. - Loading states and shimmers. - Interaction states and disabled behaviors on submit or request. These functional mocks are also far more effective for user testing. You are not validating screenshots. You are validating real behavior. It also becomes much easier to respect and evolve a solid design system organically. Just specify the process in a Cursor rules mdc. This approach does not lead to generic interfaces. It does the opposite. It creates more room to explore novel interaction ideas. This has worked extremely well for web. I believe the same mindset applies to any frontend. PS: The image was generated by AI and does not represent our actual workflow. 🤣

14

One of the most interesting funding announcements in the AI tooling space lately: 🎙️ Wispr Flow has now raised $81M (with a fresh $25M led by Notable Capital, plus Steven Bartlett’s Flight Fund) to build a voice-native OS for work. 💸 Backers include investors behind Airbnb, Slack, TikTok, Anthropic, and more — a strong signal that voice interfaces are the next wave of computing. 👉 If you’re building with AI daily (agents, copilots, workflows), it’s worth experiencing what a serious Voice OS feels like: 🔗 https://lnkd.in/ga7tzxZu

3

2 Comments