Amber Zhang 张安博

Something pretty shocking happened recently: An open-source project called Art Craft showed up. The team used Rust to reverse-engineer Adobe’s whole suite. Photoshop, Illustrator, Premiere Pro, After Effects, Lightroom, InDesign, Acrobat — suddenly, all of it was free. Here are my thoughts on $ADBE , open-source Art Craft, and what we’re actually paying for when we buy software: A designer friend told me he’d uninstalled Illustrator and switched to Vector Craft. As someone considering whether to buy the dip in Adobe, what caught my attention wasn’t that the alternative was free. It was that Vector Craft connects to other AI models through MCP, so he barely needs to draw the old-fashioned way anymore. He tells AI what to do, and it fits into the rest of his already AI-powered workflow. Adobe’s setup, meanwhile, is closed and fully managed: you can’t switch models, and getting precise control over particular functions is difficult. His reason for leaving wasn’t just price; the alternative suited how he works now. I’m still skeptical of the idea that anyone can vibe-code a replacement for professional software. When you pay for software, you’re essentially buying a group of people’s time and attention: someone fixing bugs, maintaining it, and dealing with all the annoying details long after the launch excitement wears off. Even with AI, someone has to keep answering user requests, setting priorities, and exercising product taste. Those decisions don’t have one obvious right answer, and they never really stop. That tedious, unsexy work is the value I still see in software companies. But could an open-source community do that work just as well, or better, with AI speeding things up? That’s the part I’m trying to think through in this article:

I’m happy to share my weekday coffee-break side project!!! 3D visualization to decompose an AI datacenter (to educate myself so I can level up my equity research) I’m going to be quite frank upfront: as someone who has already invested in a variety of names, from Nvidia to Micron to even lesser-known names that specifically supply a tiny part or material that goes into a hidden corner inside the data rack, I still have trouble remembering what an HBM looks like and where the ABF substrate goes under. Good memory isn’t particularly my gift, so I had to go back to Google search again and again to search “What’s the competitive landscape of component XYZ?” This finally irritated me enough that it led to my side project idea for the workday coffee break: why don’t I just build a 3D visualization to decompose an entire AI data center and show the technical components from top to bottom? And I did!

@miramurati’s Thinking Machines just dropped Inkling, and it’s a massive wake-up call for the "closed-door" frontier labs. (@AnthropicAI @OpenAI ) It really feels like we are reaching a tipping point where open weights will structurally commoditize the API-only giants. Why I think the landscape is shifting: 1/ Sovereignty. I suspect we'll see serious enterprises push back on sending proprietary data or 1M-token context dumps to third-party APIs. Running open weights behind a private firewall simply feels like the safer, more compliant long-term bet. 2/ Instead of paying flat-rate per-token premiums, hosting open-weight models on private compute lets you adjust the thinking-effort dial dynamically. Over time, the cost savings become hard to ignore. 3/ Relying on a black-box API that can change or deprecate overnight is a major risk for production systems. Owning a frozen, open-weight base gives you the predictable control needed to build safely. The high-margin opportunity seems to be moving away from raw foundation training and shifting toward private post-training and domain-specific customization. If this plays out, renting out commodity access won't be the winning business model anymore. Helping enterprises own their intelligence will be.

5/ Feature #4: AI Daily Standup Solo founder schedules are a blessing and a curse—way too flexible. It’s so easy to procrastinate or spend weeks building in a vacuum. I teach my AI to be a strategic partner: For instance, If I’m hiding behind code, it suggests marketing tasks. If I’m stuck and procrastinating, it looks at my idea log and tells me exactly what’s strategic. It also: Prioritizes: Goes through all your ideas and turns the important ones into actual todos. Automates: All checked todos go straight to your “Build Log” without manual tracking. Unblocks: Suggests resources to learn or new ideas to explore if you're feeling uninspired. Drafts: Creates social media posts based on what you’ve actually done and reflected on recently. This is your strategic morning ritual—a daily standup to help solo founders stay motivated and intentional every single day. Github link: github.com/AmberYZ/my-build-…

6/ Feature #5: Automatic social media content based on your real progress Your Notion is a diary of everything you’ve thought, built, struggled with, and celebrated. That’s natural content right there! Often, we don’t even realize how much we’ve achieved or how valuable our "lessons learned" are to others. The AI analyzes your logs and drafts content for you. You pick the drafts that resonate, and as you edit and save them, the AI learns your specific tone and style. The result? You post your true self, not generic AI fluff. Github link: github.com/AmberYZ/my-build-…

🧵1/ Launching Build-in-public Headquarters: AI + Notion = automated sharing of your progress and smartly planned todos. I was building another project when I realized something: managing multiple projects and ideas as a solo founder is a different kind of chaos. Especially for me who has ADHD, ideas grow in my brain like an explosive web. But ideas aren’t todos planned strategically. Sometimes the problem isn’t coming up with ideas — it’s knowing which one to focus on. Other times, it’s forgetting ideas altogether, or how they connect back to your north star goal. That disconnect kills momentum. So I built Build-in-Public HQ — a simple, flexible, AI-enhanced notion dashboard to help solo founders and multitaskers like me orchestrate ideas, projects, and progress without drowning in complexity. Here’s what it does: 🗂 Keeps all your projects, ideas, build logs, diary entries, and social posts in one place 💡 Logs every idea—even the random ones that don’t fit anywhere yet 🔗 Tracks your work automatically via synced GitHub releases, pushes, and commits (no more manual logs) 🤖 Generates AI-powered daily standups with strategic todos & learning suggestions based on your goals & blockers 📝 Auto-creates social media drafts from your recent builds and thoughts Everything is notion-based: no reinventing the wheel; you control your data. Turn your chaos into a map, not a mess. Github link: github.com/AmberYZ/my-build-… Key features below ⬇️:

Memory chip stocks whipsawed last week when Google announced TurboQuant. But most financial coverage skips the technical “why.” Precisely, what is TurboQuant compressing? My latest article is for anyone who wants to get an intuitive understanding of the technical picture (but not too technical). I find the underlying math fascinating—so this post focuses primarily on that. But if you want my two cents on the TurboQuant-induced sell-offs in the equity market, they are at the end of the article.

Funny how productivity shifts. I used to find copying and pasting the same content across different channels a total chore and a waste of time—the kind of problem worth building a tool to solve. But now, it has become my sanctuary for relaxing after finishing work. My mental context window is completely burnt out because AI makes every second count. A bit of 'mindlessness' is the only way I can actually relax.

We need to touch grass and reset brain context window

This is the best intuitive explanation of Google's Turbo Quant I found online. reddit.com/r/LocalLLaMA/comm…

I put my dyslexia web reader demo on YouTube (cold start, no ads!) and was shocked to get five comments and 100% likes in a single day. Sometimes I have an existential crisis thinking that all the good business ideas have already been—or will be—done by the tech giants. But I re-anchor myself by focusing on helping people with real, tangible problems. I prefer ideas that make a small group very happy over viral concepts that grab mass attention but don't end up being useful. If a new model eventually supersedes what I’ve built, I simply move on to the next problem that needs solving. The secret to conquering AI existential dread is to stop building to AVOID being replaced and start building to EMBRACE the next chance to create a meaningful impact. Focus on where you can drive change right now. We really need to actively program our minds with proactive, motivational language; it truly is life-changing! Link to the Chrome extension: chromewebstore.google.com/de… Demo below ⬇️ youtube.com/watch?v=Xa1M8GWn…

Because I have dyslexia and couldn’t find one single good tool, I built this chrome extension to assist with reading long articles: chromewebstore.google.com/de… I don’t like using “AI summary” on materials I actually want to retain and internalize. So I built this tool that allows you to actively engage with the material, such as asking questions, writing reflections, and getting quizzes. More on product design and building process to come!

The "Harness Engineering" is the defining AI trend of 2026. What is a Harness? Think of the LLM as a brilliant brain in a jar. It’s smart, but it’s essentially passive. The harness is the "body" and "nervous system." It’s the software scaffolding that gives the brain hands to use tools, a memory to store context, and the logic to self-correct when things go wrong. We’ve spent the last few years obsessed with making the brain bigger (remember the “scaling law”?). Now, the industry is racing to make the system more autonomous. In 2024, you asked a model for an answer. In 2026, you give an agent a goal. The technical paths are already diverging. MiniMax is letting the model build its own harness from scratch. Anthropic is using "trajectories" (logs of agent behavior) to train their models on how to better navigate the world. Meanwhile, NVIDIA is moving toward the infrastructure layer with OpenShell, providing a secure "Operating System" for these agents to live in. What does it mean for AI Builders? For the rest of us, this marks the end of Prompt Engineering. We are moving toward Harness Engineering. An easier and intuitive understanding: The harness handles the "how," while you focus entirely on the "what." You won't spend your morning tweaking words in a chat box; you'll spend it defining high-level objectives.

We should value our “brain tokens” as much as AI compute tokens — our intention and attention are our most precious assets. Every time we pretend to care or work on someone else’s dream, we deplete our own. “Intention burnout” is real: we can have all the productivity in the world but no drive to do anything meaningful. How often do we spend our energy optimizing tasks that don’t align with our values?