Meilisearch’s Post

I wanted to know what developers actually think of the new wave of vibe-coding tools, not what their ads promise. So I analyzed what people were saying about five of them (Lovable, Bolt, v0, Base44, and Replit) on Hacker News and pulled it into a dashboard. I expected to come away ranking the tools, but I came across a few insights that made this all the more interesting. This is only Hacker News, so it leans developer-heavy and a little cynical, but a few things stood out. Lovable got the warmest reception. People liked that the code it wrote was clean enough to keep working with afterward. Replit came up the most, but a lot of that was about the time its AI deleted someone's live database by accident. For Base44, most of the talk was about its ads, not the product. People were just tired of seeing them everywhere. The tools people talked about most lately weren't the ones they liked most. Usually, they were the ones in the news for something going wrong. The part that surprised me was Vercel. Vercel started out as a place to host and launch websites, not build them. But it owns v0, and recently v0 has grown from a tool that designed page layouts to one that builds whole apps from a prompt. Vercel now describes itself as an "AI Cloud" and raised a large round last year to lean into that. So it's in an unusual spot. It competes with these builders through v0, and it's also one of the main places web apps get hosted. Most of the other companies are just trying to make the best builder. Vercel seems to want both: the tool people build with, and the place the finished apps live. Dashboard link in the comments. Tools analyzed: Lovable, Base44, Replit, bolt.new, v0 by Vercel

Google's new RAG pipeline is your website. Search is shifting from an index to an inference endpoint. The SERP is becoming a closed-loop, generative UI. Your docs, your blogs, your API docs—they're now training data for Google's answers, with zero referral traffic back. For devs and founders, this changes everything. SEO is now AI-summary-optimization. Scraping just got legally murkier. Building a content-based moat means competing with the search box itself. Your product's discovery is being gatekept. It's not about AI supremacy; it's about Google moving from curator to proprietor. The open web's primary gateway is now its biggest wall. Who's already planning a pivot because the top of the funnel is an AI agent? #SEO #SaaS #OpenWeb

15 years ago, @MagnumDomains bought Symphony.co at a GoDaddy auction for $365. A few days ago, they announced it had sold for $180,000. Today we finally know the buyer. 👇 The buyer is @Wix. They renamed their AI assistant from Orion by Wix to Symphony by Wix. "Orion" is one of the most common names in tech. "Symphony" fits much better: one AI bringing together email, SMS, WhatsApp, calendars and reminders. Think Wix paid too much? Take a look at AI company they acquired recently. 👇 The story behind Base44 is interesting. Founder @MaorShlomo got the idea after his girlfriend needed a website for her business. Instead of building it for her, he built a platform where she, and anyone else, could build their own apps with AI. Base44 wasn't first. Tools like Bolt, Lovable, Replit and v0 could already generate code. So what made Base44 different? Most AI coding tools could generate the UI. But then they asked you to: • Connect Supabase • Configure Firebase • Add API keys For developers that's easy. For non-technical users, that's where they quit. Base44 solved that. Instead of sending users to other services, it handled the database, authentication, backend and analytics automatically. Just described the app/website. It built a working full-stack application. Back to the domain... Wix paid $180k for http://Symphony.co Right now it's just a landing page. We'll see if it becomes a major AI brand or simply turns out to be a smart domain investment. If you're interested in premium domain names, I also own Baby.co and Cinema.co Both are open to serious 5-figure offers. DMs are open.

Managed Weaviate hit Public Preview on DigitalOcean on July 9. Which happened to be my birthday, so I'm choosing to take it personally 🎂 Full disclosure: I run paid marketing at DigitalOcean, I don't spin up databases. So I asked our product/eng folks what this actually means for the people who do. Sharing what I learned because it made the whole thing click for me: Vector databases are how AI apps look things up by meaning instead of exact words. Every chatbot answering from your docs has one underneath. Weaviate is one of the most popular open-source ones (1B+ ecosystem downloads, per their CEO Bob van Luijt). ▶︎ Before this launch, teams had two options for that piece: 1️⃣ Run Weaviate yourself. Someone babysits servers, updates, backups, security. Forever 2️⃣ Rent a hosted vector DB from another company. Second vendor, second bill, usage-based pricing that spikes, and your data lives outside the cloud where your app runs ▶︎ Managed Weaviate kills both headaches. Click a button, get a production-ready cluster DigitalOcean keeps running, patched, and secured. Your code doesn't change: it's the exact same engine you'd run yourself. 💸 Flat pricing from $20/mo. Prototype and leave it running without bill anxiety. Per-query pricing punishes success: more users, surprise invoice 🔓 Unmodified open-source engine. Every tutorial and Stack Overflow answer just works. And if you ever leave DO, your whole setup moves with you. Making it easy to leave is exactly what makes people comfortable committing 🧠 RQ8 compression on by default: same workload, roughly a quarter of the RAM. Memory is what you pay for in vector databases, so that means a smaller, cheaper cluster with zero tuning Your app, your data, and your AI models already run on DO. This was one of the last pieces that forced teams to stitch in an outside service. Now the whole loop lives in one place, on one bill. Live under Vector Databases in the control panel 👇 Some people get socks for their birthday 🤷♂️ #VectorDatabase #Weaviate #AIEngineering #DigitalOcean

I spent last night building something I've been thinking about for months. At 3am, I watched an AI agent write a full SaaS product from a single sentence. Not a prototype. Not a mockup. A live product with a Postgres schema, FastAPI backend, Clerk auth, Stripe billing, and a landing page deployed and running. The idea I gave it: "a SaaS for freelancers to track invoices and get paid faster" 8 minutes later: ✦ Product name: SwiftInvoice ✦ 13 features designed (6 MVP, 4 v2, 3 v3) ✦ Viability score: 0.87 ✦ Full Postgres schema with multi-tenant RLS ✦ FastAPI routes generated ✦ Clerk auth wired ✦ Stripe plans configured ✦ Landing page copy written in my voice ✦ Cost: $1.10 This is Xantar. The AI founder OS I'm building. Describe a business in plain English. Get a live SaaS. Cursor did this for code. I'm doing it for entire products. Every build teaches the system. At 1,000 builds → fine-tuning begins. At 10,000 builds → proprietary models. At 100,000 builds → moat no one can cross. This is the Cursor playbook. Applied to SaaS creation itself. Xantar is in early access. If you're a founder who's had an idea sitting in a notes app for 6 months because you can't build it This is for you. Drop a comment with your idea. I'll run it through Xantar live. #BuildInPublic #AIStartup #SaaS #Founder #IndieHacker

Is today's AEO/GEO yesterday's browser agent? Yesterday Promptwatch raised €6M (congrats Gijs de Groot, Klaas Foppen). Two months ago Peec AI crossed $10M ARR. In January Bluefish raised $22M. Profound is at $1B valuation on $155M total funding across four rounds. I have seen this shape before. Twelve months ago it was browser agents. Browserbase went from seed to $300M valuation in fourteen months and now sits with a customer list that reads like the current wave of AI-native software. Browser Use closed $17M and open-sourced a stack that has become a default in the agent stack. Anchor Browser shipped and got picked up by real customers. That category did not fade. It became critical infrastructure. Here is the piece the funding coverage misses. Kleiner Perkins led Browserbase's seed. Then their Series A. Then participated in their Series B. Then walked one layer up the stack and joined Profound's Series B and Series C. Same firm, two adjacent categories, twelve months apart. Their public memo on Browserbase referenced their 1996 Netscape investment as the frame. They are running the same bet again. That is not category chasing. That is a repeatable thesis. There is a difference between the two categories worth naming. Browser agents burn cash on the compute underneath every session. AEO burns cash on daily scraping of the AI engines it measures. Different cost curve. AEO companies are hitting real ARR on less capital: Peec crossed $10M ARR on $29M raised, sixteen months from launch. That efficiency is one of the reasons the same investors keep showing up. The bet, as far as I can read it: when a new interaction paradigm arrives on the internet, the biggest returns are not in the interaction. They are in the tooling and measurement that grow up around it. Netscape opened the web and Google, Akamai, and Adobe built the layers that made it work at scale. Browser agents opened autonomous access to the web. AEO is one of the layers making the resulting AI-native discovery legible to the businesses that depend on being found. Both categories are real. Both are still being built. The people writing the biggest checks are the ones who already recognize the pattern. The next category on the same thesis will land in the next twelve months. Whoever names it before the funding hits will get called a genius. It is already being built.

I just shipped Simple Polls 🚀 — a lightweight, standalone app built to show what happens when a business stops renting bloated software and starts owning its tools. Over the last few months, I’ve posted a lot about building software for early-stage startups. But while deploying this project, I had a major realization: The modern tech stack I love using isn't just for building new SaaS companies. It is the ultimate weapon for solving a massive headache that existing, profitable boutique agencies face every day: Operational Bloat. So many growing agencies suffer from "SaaS fatigue." They pay heavy, monthly per-user fees for software they only use 20% of, or their teams drown in manual work, pushing data between rigid platforms that don't talk to each other. I built Simple Polls to prove there is a better way. I used a production-grade stack engineered for speed, security, and zero server-maintenance overhead: Next.js & Tailwind CSS: For an instantly loading, lightning-fast user experience. MongoDB & Mongoose: A flexible database architecture built to scale data effortlessly. Clerk: Enterprise-grade user authentication to ensure data security out of the box. shadcn/ui: For a clean, modern, and highly accessible interface without design bloat. The Blueprint for Agencies: This isn't just a polling app. It’s a case study in eliminating friction. The exact same architecture powers custom client onboarding portals, secure internal project dashboards, and automated operational tools. When you build custom, you cut the recurring software fees, streamline your team's workflow, and build an asset your business actually owns. 👇 Check out the live, fully-deployed app in the comments below! 🛠️ Agency owners: What is the most annoying manual process or messy spreadsheet your team is fighting this week? Let’s connect in the DMs and talk about building a custom fix. #NextJS #MongoDB #BoutiqueAgency #InternalTools #Automation #WebDevelopment #Solopreneur

𝐌𝐨𝐬𝐭 𝐃𝐞𝐯𝐓𝐨𝐨𝐥𝐬 𝐧𝐞𝐯𝐞𝐫 𝐫𝐞𝐚𝐜𝐡 1,000 𝐫𝐞𝐚𝐥 𝐮𝐬𝐞𝐫𝐬. Not because the product fails. Because the distribution infrastructure was never built. The teams that compound from launch to 1,000 users do not get lucky. They instrument every layer before anyone arrives, and they close friction loops faster than developers can switch tools. 𝐇𝐞𝐫𝐞 𝐢𝐬 𝐭𝐡𝐞 𝐫𝐨𝐚𝐝𝐦𝐚𝐩 𝐭𝐡𝐚𝐭 𝐚𝐜𝐭𝐮𝐚𝐥𝐥𝐲 𝐰𝐨𝐫𝐤𝐬: → 90 𝐝𝐚𝐲𝐬 𝐨𝐮𝐭: 𝐟𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 README as positioning. Product docs complete. llms.txt live. GitHub structured to pass a silent credibility audit before a single external developer sees it. This layer determines whether launch traffic converts or bounces. → 60 𝐝𝐚𝐲𝐬 𝐨𝐮𝐭: 𝐩𝐫𝐨𝐨𝐟 Sandbox live. Real integration examples, not toy demos. First working output under 15 minutes in real conditions. If this layer is missing, the first install fails. No second attempt follows. → 45 𝐝𝐚𝐲𝐬 𝐨𝐮𝐭: 𝐜𝐨𝐦𝐦𝐮𝐧𝐢𝐭𝐲 𝐬𝐞𝐞𝐝𝐢𝐧𝐠 Problem-first Reddit threads with no product mention. Presence in spaces developers already trust before you need anything from them. Community that does not exist at launch cannot be built in the same week. → 30 𝐝𝐚𝐲𝐬 𝐨𝐮𝐭: 𝐀𝐈 𝐬𝐮𝐫𝐟𝐚𝐜𝐞 Run the LLM query audit. Does your tool appear when a developer asks for a solution to the problem you solve? This layer determines your visibility before a human ever visits your repo. → 14 𝐝𝐚𝐲𝐬 𝐨𝐮𝐭: 𝐜𝐨𝐧𝐭𝐞𝐧𝐭 One deep use-case tutorial indexed before launch outperforms ten generic posts at every stage of the funnel. Comparison content live before competitors respond. Structure built for AI retrieval, not just crawlers. → 𝐋𝐚𝐮𝐧𝐜𝐡 𝐰𝐞𝐞𝐤: 𝐚𝐜𝐭𝐢𝐯𝐚𝐭𝐢𝐨𝐧 Product Hunt, Hacker News, Reddit, and developer newsletters are different trust layers with different framing requirements. Every signup must reach first value under 15 minutes. A spike that does not convert is noise with a star count. → 𝐔𝐬𝐞𝐫𝐬 1 𝐭𝐨 100: 𝐬𝐢𝐠𝐧𝐚𝐥 𝐜𝐚𝐩𝐭𝐮𝐫𝐞 This cohort determines everything. Direct conversations. Friction documented and fixed in real time. Changelog honest and public. Every developer who churns here takes an advocate, a teammate referral, and a potential case study with them. → 𝐔𝐬𝐞𝐫𝐬 100 𝐭𝐨 500: 𝐜𝐨𝐦𝐩𝐨𝐮𝐧𝐝𝐢𝐧𝐠 Organic peer mentions appearing unprompted. AI citations increasing without active pushing. The flywheel is either starting here or it is not starting. → 𝐔𝐬𝐞𝐫𝐬 500 𝐭𝐨 1,000: 𝐯𝐞𝐥𝐨𝐜𝐢𝐭𝐲 Contributors beyond the core team. Community recommending the tool in threads the team never sees. Technical content compounding in AI retrieval. Teammates being pulled in by existing users. First 1,000 users is not a marketing milestone. It is proof that the trust infrastructure holds end to end. P.S. What is the stage between launch and 1,000 users where you have seen the most DevTools stall, and what did closing that gap actually require? Follow Shantanu Das ↗️ for more insights

Someone can fork your codebase, register a domain, and start harvesting your commercial demand before they have a single customer. We just documented it happening in real time. Unspoken, a commercial Mac dictation app, forked a GPL v3 open-source project with 5,300 GitHub stars. Domain registered January 2026. By launch: an llms.txt file inviting AI crawlers, competitor-alternative landing pages in two languages, ten persona-segmented content hubs, and free utility tools for query capture. All before the product had any organic reputation whatsoever. This inverts twenty years of GTM orthodoxy. The old playbook: build product → earn users → accrete search authority over years. The fork-and-GEO playbook: inherit the product on day zero → manufacture the citation surface before the first customer arrives. Why it works: research from KDD 2024 shows content structured for AI citation lifts visibility in generative answers by 40%+. And ungated content can achieve confirmed citation status in engines like Perplexity within days — a timeline with no analog in traditional SEO. If you're a founder, this cuts both ways: → If your product IS a fork: document your license compliance story before a buyer's counsel finds the gaps. GPL exposure surfaces mid-transaction, at the worst possible moment. → If your product is the upstream: your demand is being harvested by forks with better GEO. Audit your citation share quarterly. Either way, enter any exit process with a graduation narrative: what did you build that the codebase didn't give you? Full analysis, including how Cursor, Brave, and Grafana proved the thesis at billion-dollar scale: https://lnkd.in/edDqDBY6 #SaaS #Startups #OpenSource #GTM #ExitStrategy

Two years ago, standing up a real product — auth, payments, a data layer, email — was a wall I usually didn't feel like climbing at 10pm. So a lot of my "could have been a company" side projects just stayed side projects. Last week I built BeingSaid, a small SaaS search tool, from zero to one. This time the wall wasn't there. Two things had changed: the tools got dramatically better, and AI helped me navigate the ones that are still fiddly. The stack that got me shipped: → Vercel for hosting + CI/CD. A GitHub repo to a live URL is genuinely dead simple now — this alone reshaped what shipping costs. → Supabase for the whole data layer — database and blob storage in one clean place. → Google for auth. You add it with almost no backend integration and it's secure. Credit to the standards folks (OAuth, JWT) as much as Google here. → Stripe for payments. Best-in-class product; the docs are just endless. AI helped me cut through to the parts that mattered. → PostHog for analytics. → ImprovMX + Gmail "Send As" for email. The most involved part by far — DNS, a Google Workspace, a Google Group — but it gets me a professional address at close to zero cost versus paying for a Google Workspace seat just for this app. AI was most useful here, researching each step so it actually made sense. The honest takeaway: when code is cheap, the software stack stops being the hard part — the service stack is where the focus goes. The real work is choosing the right hosting, data, auth, and payments services and wiring them together well. AI isn't a replacement for those services; it's what helps you get the most out of tools that are already really good, especially in the fiddly corners. If you're a small team wondering how to actually ship the thing you keep putting off, this is the kind of problem I love thinking about — happy to compare notes. Full write-up in the comments. 👇 #SaaS #BuildInPublic #Startups #AI #WebDev