Alex Castillo

I’ve been playing with AI design tools for a while—initially as a way to move faster from intent to interface, and lately as something more. What caught my attention this past week wasn’t a new feature, but a pairing: Lovable running on Supabase. That detail shifted the frame for me. I care deeply about the front end—what our customers actually touch that improves their lives—and how my design teams can help define it faster, with greater clarity. But when tools like Lovable start partnering with infrastructure like Supabase, AI prompt design is no longer just about accelerating prototyping. It’s about collapsing entire layers of traditional ops and redefining where engineering effort gets applied. Supabase gives Lovable a real backend—live data, auth, persistence—without the drag of building scaffolding from scratch. Suddenly, you're not just seeing screens. You’re interacting with product-like behavior on day one. As a design org, that makes us force multipliers in a way that used to take a sprint cycle to even approximate. But the deeper shift is this: AI-native tools are no longer operating in isolation. They're forming ecosystems. Lovable on Supabase isn’t just a clever tech decision—it’s a systems alignment. And that alignment reshapes the economics of internal investment. You start thinking less about handoff and fidelity loss, and more about which tools can actually compose into one another in real time. That’s a very different kind of stack planning. It’s easy to say this is just where the tooling is headed. But to me, it feels more like a return to something I’ve missed. In the Web 2.0 era, people talked about this stuff—openly, publicly, and often imperfectly. We wrote about what it meant for tools to be modular, or for the web to behave like a platform instead of a destination. We imagined lightweight, expressive infrastructure that supported creativity instead of slowing it down. And we made decisions—sometimes naïve, sometimes visionary—based on what felt possible in that moment. This feels like one of those moments. We’re watching the scaffolding of product creation become intelligible across functions. The tools are increasingly fluent with each other. And instead of fighting for fidelity between design and build, we’re moving toward something far more fluid: a product surface that responds to intent, supported by infrastructure that doesn’t need a dedicated team to stand up. Call it Web 2.5 if you want. I’m less interested in the label than the posture. And right now, the posture feels open again—like the best tools are quietly building toward something bigger than themselves.

7

4 Comments

New n8n template by me: Supabase Storage Tutorial Workflow Template! I’m excited to unveil an all-new, beginner-friendly workflow template on n8n that makes accessing Supabase Storage simpler, smoother, and more streamlined than ever. What this template does Teach you to: Upload Files — Users can upload files via an n8n form, which are saved directly into a Supabase storage bucket. Fetch Files — Easily retrieve files by filename. Generate Temporary Access — Create signed URLs with custom expiry for secure sharing. List Stored Files — View all files currently stored in the chosen bucket. Why this matters: Ideal for developers, teams, and educators looking to explore Supabase Storage through hands-on automation. Simplifies secure file management in automated workflows. Serves as a practical demo to accelerate onboarding and internal tooling. How to get started: 1. Set up a Supabase project and enable Storage. 2. Create a storage bucket (e.g. test-n8n). 3. Grab your Supabase Project URL and Anon Key. 4. In n8n, configure a Supabase API Credential using those keys. 5. Import the workflow template. 6. Test it out via the n8n form—to upload, retrieve, sign, and list files. https://lnkd.in/dG7FqCmb Happy automating!

6

1 Comment

Recreate iOS Messages in one step using a Rive layout that fills the space. Simulate a real conversation with 10 messages and placeholders for icons. Include the header and input field. Use the correct color scheme

96

2 Comments

Anthropic has made the 1M context window for Claude Opus/Sonnet 4.6 generally available in Claude Code, Claude.ai, and removed the price premium (ie. no extra charge for large context usages): https://lnkd.in/dikp8BEh For coding, this is huge. On a recent project involving a few thousand lines of C code, I was constantly hitting the context - the language was just too verbose and the problem had too many important details to comfortably fit under 200k. I've been playing with this for a few days and it doesn't seem different from before, except I don't have to keep an eye on context usage (aside: they added a tiny tweak to Claude Code's "clear context and implement plan" where they show how much context is in use - very helpful when you're trying to determine whether to clear or not). It remains to be seen whether the context is good in practice. Benchmark say recall is high (78% on 1M tokens), but practice often differs. But I am hoping this also improved the recall on smaller context sizes (200-400k), which is a huge win in of itself. Another goodie (especially for folks in Europe, Africa and Asia timezones), Anthropic temporarily raises the US "off peak" limits 2x: - 2x usage on weekdays OUTSIDE 5–11am PT / 12–6pm GMT / 1pm-7pm CET - 2x usage all day on weekends This is likely an experiment to see if they can "bribe" people to offload usage to off-peak hours (similar to how many electricity companies operate). This can make Claude Pro ($20/mo) more usable, esp. if we start to (ab)use the 1M context windows. It'll be interesting to see if the "temporary 2 weeks" ends up permanent.

42

The New Standard for Rapid Prototyping If you aren't pairing Claude Code with Google Stitch, you're working too hard. The secret sauce is the MCP (Model Context Protocol). It allows you to bridge Claude’s reasoning power directly with Stitch’s design engine. The Architecture: Claude Code is the Brain: It handles the complex logic, manages your codebase, and structures the architecture of your app. Google Stitch is the Visual Nervous System: It interprets design intent and generates high-fidelity, responsive UIs in real-time.  Why this matters: Traditionally, terminal-based AI coding agents were "blind" to the front end. By using the Stitch MCP, you're giving Claude a pair of eyes and a professional drafting table. The result? You ship production-ready code and pixel-perfect UI simultaneously, without the constant back-and-forth between design tools and your IDE.  It’s effectively a Full-Stack Autopilot. Are you sticking to manual UI builds, or are you moving to an MCP-driven workflow?

3

GenAI tip: identifying when / how to use Claude cowork. Ask Claude to interview you to learn about a typical working week. Explain that the purpose of the interview is for Claude to identify ways to use Claude cowork to help you. At the end of the interview ask Claude to create a downloadable document Witt the top three use cases for cowork and a step by step guide to start using it for each use case.

7

5 Comments

If you're a designer interested in prototyping using AI, but are unsure how to get started, Adobe Design Prototyping manager CJ Gammon just released a new article on adobe.design that might just help you out. In it, he shares a few excellent tips and tricks on how to successfully prototype using AI tools, based on real experience in how designers at Adobe are using AI-based prototyping today. https://lnkd.in/guytfQwv

47

2 Comments

AI tools like Cursor and Claude Code can now "read" Figma via MCP. On paper, that sounds like the end of the painful design-to-code handoff. In practice? It's complicated. When it works, it's genuinely impressive. Figma's MCP paired with Code Connect can produce results that feel like magic. When it doesn't work — and that happens often — the AI picks up DS components but gets the props wrong. Or misses them entirely. Or generates something that looks right at a glance but breaks edge cases. The deeper issue is that AI works best with well-known component libraries it's already been trained on: ShadCN, MUI, AntDesign. Ask it to understand a custom, company-specific design system with hundreds of components and nuanced variant logic? That's expensive in context, tricky to teach, and inconsistent in execution. So is AI handoff realistic? Getting there — but not yet. The flow is improving fast, and when the tooling catches up, it will genuinely change things. Just don't cancel your visual QA sprint yet. 🔗 Read more: https://lnkd.in/dyb6QbqB

8

2 Comments

Challenge: Blur the background of a UI Toolkit Visual Element Took about 4hr~, and It works really well * VisualElement can be resized/moved at runtime - completely dynamic * Easy control over blur amount * Performance is great and this implementation allows for multiple panels to enable blur on the background without a performance hit. Foresight Sports, XGOLF America Inc. GOLF+

4

Pokémon GO was never a game. It was a 24/7 unpaid internship. It‘s 2023 and we all are playing it on our phones. Turns out: While collecting monsters, we were actually training a huge 3D model with our lidar scanners in our phone. This model is now actively monetized. It was a highly effective way to collect structured, real-world data at scale, because people were motivated to generate it without thinking about it. This is exactly what’s happening in Healthcare right now. Every time we log a run, "close a ring," or track our sleep, we think we’re leveling up our personal stats. In reality: We are the Sensors: We’re providing the high-fidelity biometric data needed to train massive health algorithms. The "Rewards" are Bait: Badges and streaks are just the incentives used to ensure we never stop submitting our data. The Model is the Product: The goal isn't just "helping you"—it’s building a predictive health engine that companies will eventually monetize and sell back to us. Just as Niantic turned our weekend walks into a 3D map, the healthcare industry is turning our biological "output" into a Digital Twin. Niantic proved that if the UX is engaging enough, the "product" will build itself for free. In Healthcare, we’ve reached the same inflection point: we aren't just users of health tech; we are the R&D department. But the real value is the aggregated data layer that gets extracted, modeled, and monetized. And today, most providers don’t own that layer. That’s the uncomfortable part. If you’re not actively building on top of your data, you’re just supplying it. The question isn’t whether data is being collected. It’s who captures the value. #AI #Healthcare #Gamification #Tech https://lnkd.in/dJYV6t_E

20

How to use Claude Code better for designers: ⚡ I went to an event about using Claude Code and wrote down these tips. Get Claude to interview you about your project: A 10 min interview with Claude is better than starting from scratch. Ask Claude to interview you so it gets all the context before you begin. Using Plan Mode: Use Plan Mode to plan out your projects before starting. Most of engineering is logic, not syntax — Plan Mode forces you to think through the flow before building. Figma MCP: Use the Figma MCP to let Claude “inspect” your design files directly. It reads your Auto Layouts and design tokens to generate production-ready code that actually matches your mocks. It’s not a “one-click deploy,” but it eliminates the design-to-code translation error. Superpower Plugin: Use the Superpower plugin to brainstorm architectural approaches — great for when you’re stuck and want to explore a few different directions before committing. CLAUDE.md file: Use the CLAUDE.md file to set up your project — it acts as your source of truth. Claude will read this at the start of every conversation. Desktop App: A lot of people don’t know they can use the desktop app with Claude Code. Use it if you want to see previews of your design while using the terminal. #Claude #Claudecode #vibecoding #vibedesigning

721

23 Comments

Don't believe the BS that you can use Claude Code for free. Ollama recently made their API compatible with Claude Code. Many creators quickly jumped on the opportunity to farm engagement with the hook: "You can now use Claude Code for free!" My thought? Claude Code without Opus 4.5 is not Claude Code. Period. But this is exciting news. Not because I can use Claude Code for free, but because I see an opportunity to optimize costs by delegating easier tasks to local LLMs. The key question is: what tasks can local LLMs handle? I tested out 7 local LLMs. In this post, I will explain the BS and share my 1st experiment. -- Why the BS? Claude Code has been praised as one of the best AI tools by its users, not only for coding but for many other tasks. But the price feels steep to many. The $20/month plan is not enough for any serious work. You need to at least subscribe to Max 5x ($100/month). Many heavy users, including me, subscribe to Max 20x ($200/month). It’s a steal. But still, many were eager to try it but aren't ready to pay. Ollama's recent announcement means you can buy a Mac, a Strix Halo, or a GPU and use Claude Code for "free" with local LLMs. It’s appealing, as it is a one-time investment, and you can use the machine for other purposes. Creators are leveraging this opportunity to farm engagement. But the reality? Claude Code without Opus 4.5 is not the same Claude Code we praised. Local LLMs are far less intelligent. -- But for those who understand the difference, we see an opportunity to optimize costs by delegating some easier tasks to local LLMs. I'm interested in finding out what tasks local LLMs can handle. This is my typical flow when using Claude Code. This is for coding, but I have similar flows for marketing and content creation. 1. Research and planning 2. Create PRD and implementation plan 3. Break plan into bite-sized tasks 4. Implement + review with reflection pattern 5. Final review with agents 6. Final review and QA by me Based on my quick tests, we can forget about asking local LLMs to do research and planning. All of them failed at a simple instruction: "Visit https://learnparrot.ai/ and tell me about the website." So, I think the most viable use cases would only be (4) — implementation + review loops. While it looks like a very narrow use case, it is where we burn a lot of tokens. So I think it is worth a try. The main selection criteria for this will be instruction-following capability. One very common task is to refer to code samples or templates to code a new feature or page. This is a good test of instruction-following. So, I crafted my first test: - Used Opus to create HTML that I can screenshot as a LinkedIn carousel to display info for each model. - Turned one of the pages into a template and deleted the rest. - Asked each LLM to refer to the template to code its own page, given its specs. Swipe the carousel to see the results. Who would you hire? #LocalLLM #ClaudeCode #VibeCoding

20

10 Comments

iOS 26.2 is frustrating everyone I know. The new full-screen screenshot feature interrupts your workflow constantly. Quick fix: Settings → Search "screenshot" → Turn off "Full Screen Previews" Your screenshots return to the compact corner preview. If you screenshot frequently, this small change makes a big difference. Save this post, you'll need it next time someone asks you why their screenshots changed. #iOS #ProductivityTips #TechTips #WorkflowOptimization

4