Debunking AI myths and hype in product development

The top 5 websites in the world right now: Google. YouTube. Facebook. Instagram. ChatGPT. It’s kind of wild when you think about it: -Google helps us find literally everything—fast. -YouTube entertains, teaches, and sometimes makes us laugh at 2 a.m. -Facebook keeps people talking (and debating) in ways no one really expected. -Instagram makes scrolling a full-on visual experience. -ChatGPT… well, it’s like having a marketing, coding, and brainstorming assistant in your pocket. For marketers, it’s a reminder that attention is everywhere, formats are endless, and people are picky about what they engage with. Sometimes we overcomplicate campaigns. But these sites show that if it’s fast, useful, or entertaining—you’ve got a shot at making people stop scrolling.

The last time production and distribution costs both hit zero, we got TikTok. Now it is happening again—this time in software. Remember when watching videos meant worrying about data charges and editing required professional skills? Then 4G and CapCut arrived, costs dropped to near-zero, and Douyin exploded into a 400-billion-yuan ecosystem. The same mechanics are reshaping software right now. On the consumption side, AI has obliterated the installation barrier. When anyone can ask ChatGPT or DeepSeek to walk them through setup step-by-step, "download and install" stops being a technical challenge. GitHub's numbers tell the story: 36 million new developers joined in a single year, with 80% using Copilot in their first week. Many are not traditional programmers—they are marketers, designers, and entrepreneurs who simply learned to ask the right questions. On the production side, coding agents are doing to software what CapCut did to video. Building an MVP used to take months and a team. Now it is one person plus Cursor or Lovable, shipping in days. Cursor hit $2 billion in annualized revenue by early 2026. Forty-one percent of code already involves AI-generated contributions. Enter OpenClaw. Launched just months ago, it is already the fastest-growing project in GitHub history with over 250,000 stars. When non-programmers—students, retirees, office workers—are lining up at Tencent headquarters to get help installing open source software, you know the paradigm has shifted. This is not just about one viral project. It is platform economics 101. When creation and access both become frictionless, distribution explodes. Here is the strategic takeaway: Open source communities are becoming to software what short video platforms became to entertainment—the primary channel where production, distribution, and influence converge. With 121 million new repositories created in 2025 alone, the velocity is staggering. If you are in tech or product, this deserves more than casual observation. It deserves a core place in your strategy. Read the full article: https://lnkd.in/gYYPqWFz #OpenSource #AI #SoftwareDevelopment #FutureOfTech #Innovation

Comparing Claude Code to early TikTok is a great analogy - FOR CONSUMERS. In the Enterprise world they are not looking for the next TikTok. The want the next Railroad. Why? The Railroad became the connective tissue that the entire modern economy ran on during that time. Enterprise AI means companies that build correctly own the infrastructure that every other process is forced to route through. The LinkedIn echo chamber is obsessed with Claude. Most $50M+ companies I talk to are already sitting on a Google Cloud/Gemini backbone. They don't need to over spend to be disconnected... all they need AI-Native Infrastructure that talks to their proprietary data in real-time. For Enterprise Companies and Business Leaders I feel Claude is distraction... at best.

Sean Parker didn't accidentally build one of the most addictive products in human history. He did it on purpose. The co-founder of Facebook (Now Meta) said it himself in 2017: "How do we consume as much of your time and conscious attention as possible?" That was the design brief. If there are no connections, there is no community, hence no consumption. And it worked beyond anyone's wildest projections. The average person now spends 2 hours 23 minutes per day on social media. That's roughly 35 days a year; eyes down, thumb moving, mind on autopilot. But here's the part that doesn't make the headline: You didn't choose to spend 35 days scrolling; the algorithm chose for you. Every platform runs on a variable reward loop, the same psychological mechanism that powers slot machines. You don't know if the next post will be boring or brilliant, so you keep pulling the lever. Dopamine fires, the scroll continues, and the platform wins. And the echo chamber isn't a bug, it's a feature. The algorithm doesn't show you the world. It shows you a mirror, reflecting your existing beliefs back at you, amplified and validated, because outrage and agreement both drive engagement, and engagement drives revenue. A 2021 @MIT study found that false news spreads 6x faster than true news on Twitter (Now X). Not because people prefer lies, but because lies tend to be more emotionally provocative, and emotion is what the algorithm rewards. You weren't radicalised, you were optimised. The cigarette industry once hired doctors to appear in ads. They understood that the most dangerous product is one that feels harmless, even pleasurable, while it quietly rewires you. Social media hired behavioural psychologists. Same playbook, different product, but faster scale. The question isn't whether you're scrolling. It's whether you're aware of who designed the scroll, and why. Screengrab: Mint

OpenAI shutting down Sora is the most predictable outcome of misunderstanding consumer behavior. Everyone grossly overestimates how creative people want to be. 99% of humans simply want to scroll and zone out instead of spending energy conceptualizing, editing, and creating videos. No matter how much the barrier to creation is lowered by a single prompt. It still takes effort and a fair amount of creative thinking. Even Instagram has 2 billion users. But hardly ~10 million (or 0.5%) would be serious creators. So any AI consumer app betting on "with [new_app], everyone becomes a creator!" is being delusional and will go down almost the same way. After all the word "consumer" exists for a reason. They consume. They don't produce (or even want to lol)

Instagram has shown you thousands of Reels. It has almost never shown you the same one twice. Think about how strange that is. Millions of users. Billions of videos. Personalized for each person. Refreshed every few seconds. And somehow, it remembers what you've already seen. That's not a small problem. Let's walk through what's actually happening. The first challenge is recommendation. Instagram's ranking system scores videos based on your watch history, interactions, the accounts you follow, and signals from people similar to you. This is the part most engineers think about. Collaborative filtering, embeddings, two-tower neural networks. But ranking is only half the problem. The harder problem is deduplication at scale. Instagram doesn't store a list of every Reel you've ever watched in a simple database table. That would mean querying millions of rows every time your feed refreshes. Too slow. Too expensive. Instead, they use a probabilistic data structure called a Bloom filter. A Bloom filter is a compact bit array. When you watch a Reel, its ID gets hashed and a few bits flip to 1. Next time the system considers showing you that video, it checks those bits. If they're all 1, the video is filtered out. It uses almost no memory. The lookup is near-instant. The tradeoff is it can occasionally block a video you haven't seen. A false positive. Instagram accepts that. Showing you a slightly suboptimal video is better than the system slowing down. On top of this, a candidate generation layer pre-fetches your next batch of Reels before you finish the current one. You never wait. The pipeline is always a few steps ahead. The infinite scroll isn't magic. It's a Bloom filter, a ranking model, and a prefetch queue working together in milliseconds. Most engineers focus on building the recommendation engine. The unseen complexity is remembering what not to show. What other parts of your daily app experience do you think are solved by probabilistic data structures rather than exact ones?

Yesterday I sat with someone and scrolled through their Instagram. No AI content. Not one post. Not one reel. Nothing. I work in AI. I breathe this space daily. And in that moment I realized I had completely lost touch with where the average person actually is. That’s not a humble brag. That’s a warning. Because here’s what I had to confront: I didn’t fall into a bubble. I built one. Every piece of content I liked, paused on, rewatched, or searched for was an instruction I handed to an algorithm. And it executed perfectly. It showed me more of the same, every single day, until my feed became my reality. I genuinely thought AI was everywhere. It’s everywhere for me. Those are two completely different things. And this isn’t just about AI. Think about what your feed is full of right now. Politics. Fitness. Business. Spirituality. Crypto. Whatever it is, ask yourself honestly: is that a reflection of the world, or is it a reflection of the instructions you’ve been giving without realizing it? Because the algorithm isn’t trying to inform you. It was never designed to. It was designed to keep you on the app. And the most efficient way to do that is to wrap you so tightly in your own interests that every scroll feels like confirmation that you already see things clearly. That’s where it gets dangerous. You start making decisions, forming opinions, measuring progress, judging others, all through a lens the algorithm quietly built for you. And because it feels familiar, it feels true. So here is the responsibility nobody talks about: You are not a passive user. Every like is a vote. Every skip is a signal. Every time you let something you disagree with play without telling the app you’re not interested, you’re giving it permission to show you more. You are actively shaping what you see next, which means you are actively shaping how you think. Most people are doing this completely unconsciously. Every single day. The question I want to leave here is not “is your feed diverse enough.” That’s too easy to ignore. The real question is: if someone scrolled through your feed today, what would they learn about the world you’ve built for yourself? And is that the world you actually want to be living in?

87 open tabs. Bookmarked tweets. Saved Reels. YouTube Watch Later. Clipped to Notion. I now have 6 separate content graveyards with different aesthetics. We've all built them. The browser window you can't close because you'll "lose" something. The Instagram saved folder you haven't opened since 2023. The Twitter bookmarks tab that now needs its own search function. The Notion database with 300 articles, perfectly organized, never revisited. There's a reason you never go back. When you save something, your brain registers the task as handled. Done. Filed. The Zeigarnik Effect — your mind treats saved content as a closed loop. So the very act of saving is what stops you from learning. I noticed this in my own Notion content hub. Built the perfect capture system. AI summaries. Text Highlights. Clean tags. Felt productive. Remembered nothing. So I added friction back in deliberately. Before anything gets saved now, I write one sentence in my own words: what does this mean for what I'm building? No AI assist on that step. Just me, forced to think for 10 seconds. That one clunky manual moment does something no automation can — it makes your brain encode information you actually generated, not just consumed. The graveyard problem isn't a storage problem. It's an encoding problem. Better tools won't fix it. A little deliberate resistance will.

Instagram was the moment I realised something strange about software’s future. Two friends sitting next to each other can open Instagram and see completely different apps with varying features, layouts and ranking algorithms. This is because the product is constantly running experiments. Now imagine what happens when AI dramatically speeds up software shipping. My take is that AI might actually increase the number of engineers companies hire rather than reduce it. This is because cheaper shipping means companies ship much more, not less. Instead of a few big releases annually, products could become constantly evolving systems. These systems would involve thousands of experiments running, UI changes tailored to different user segments, continuous feature generation and testing, and algorithms adapting in real time. Apps might even lose their clear “versions” and behave more like living systems that are constantly changing. As teams experiment faster, they’ll run more experiments, which means more infrastructure, monitoring and complexity. AI won’t eliminate builders; it might create a world where software evolves so quickly that we need even more of them. I’m curious to know if others see this happening too. #AI #SoftwareEngineering #Startups #TechHotTake

I found a GitHub repo that lets you build your own AI agency 🤯 Not just one AI… A full team of AI employees. → Engineers   → Designers   → Marketers   → Product managers  All working together like a real company. And it got **10k+ stars in just 7 days.** Here’s how it’s structured: 1. Engineering (7 agents)   Frontend, backend, AI, DevOps, prototyping  2. Design (7 agents)   UI/UX, branding, research, visual storytelling  3. Marketing (8 agents)   Growth, content, Twitter, TikTok, Reddit  4. Product (3 agents)   Prioritization, trends, feedback  5. Project Management (5 agents)   Coordination, operations, execution  6. Testing (7 agents)   QA, performance, API testing  7. Support (6 agents)   Customer service, analytics, finance  8. Spatial Computing (6 agents)   XR, Vision Pro, WebXR  9. Specialized (6 agents)   Sales, analytics, orchestration  --- What’s interesting is the approach: Instead of one AI doing everything… They built it like a company: → Specialized roles   → Clear responsibilities   → Coordinated workflows  This is how businesses will run in the next few years. --- I’ve saved the full GitHub repo + will share more such AI systems. 👉 Comment your email ID and I’ll send you the repo  OR  👉 Join my WhatsApp community where I share such AI tools & systems regularly (link in comments)