John Merrells

2026 Stop thinking in “AI features,” “AI tools,” and “AI optimization.” That’s small-game thinking. It keeps you trapped in the same founder/owner identity as everyone else. Start behaving like a capital allocator. Engineer an ecosystem that makes AI Startups and competitors your partner. The exact same game the big tech boys are playing right now. Take one small founder: we snapped three fragmented companies together under a single platform. Within 90 days, revenues doubled and value 10x'd all without adding a new client. The business scaled because their identity scaled first. If your current path feels slow, uncertain, or capped, it is. Your 2026 shouldn't be about working harder. It’s about stepping into a new identity that thinks bigger, moves faster, and creates outcomes with less risk that feel inevitable. #Founder #AI #2026 #Marketing #Growth #Partnerships #CEO #Tech #CFO

9

1 Comment

Keith Teare breaks AI into three categories — and the distinction matters more than most people realize. 1. Additive AI: supplements existing human tasks. Lawyers using Harvey, accountants using copilots. This is where most enterprise adoption lives today. 2. AI-first platforms: starts with AI, plugs into legacy tools. ChatGPT, Claude Code, Codex. The current state of the art. 3. AI-only: fully emergent systems where AI operates autonomously. Moltbook (the AI agent social network Meta just acquired) is an early example. Most of the public only experiences category 1 — and judges all of AI by it. But category 3 is where the transformation actually lives. The gap between "AI as a tool" and "AI as an agent" is the gap between a calculator and a colleague. 🎬 Full issue: https://lnkd.in/gttscQFp

4

1 Comment

While everyone was debating whether AI would replace consultants, OpenAI just became one. Their new consulting arm charges a minimum of $10 million per engagement, embedding Forward-Deployed Engineers directly into client organizations to build custom GPT-4o implementations. This isn't just another enterprise sales play. It's vertical integration disguised as consulting. Despite OpenAI reaching $10 billion in annual revenue and serving over 500 million weekly users, they're still burning through approximately $5 billion annually. API subscriptions alone aren't cutting it. If OpenAI, which started the ChatGPT revolution, needs to move into services, what does that mean for smaller AI startups? Google, Anthropic, Meta - they're all building comparable models. The differentiation isn't in the core technology anymore. It's in implementation, integration, and delivering measurable results. So what's the play for everyone else? 𝗕𝘂𝗶𝗹𝗱 𝘁𝗵𝗲 𝘁𝗲𝗰𝗵𝗻𝗼𝗹𝗼𝗴𝘆 → 𝗢𝘄𝗻 𝘁𝗵𝗲 𝗶𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 → 𝗗𝗲𝗹𝗶𝘃𝗲𝗿 𝘁𝗵𝗲 𝗼𝘂𝘁𝗰𝗼𝗺𝗲. OpenAI is shifting from the first to the second. The real money is in the third. The future belongs to vertical solutions that understand specific industries deeply enough to deliver turnkey outcomes. Not APIs. Not platforms. 𝗥𝗲𝘀𝘂𝗹𝘁𝘀. This is why I'm betting on micro-SaaS in specialized verticals. When AI becomes commoditized infrastructure, domain expertise and implementation capability become crucial differentiators. Pure technology plays are getting harder and harder to defend. APIs are becoming commodities. Implementation is the new IP. I share more musings on my substack >> https://t2m.io/4aDyNCpu For the love of the game 🏴‍☠️⚡️ #ai #futurism #chatgpt #openai 𝘐𝘮𝘢𝘨𝘦 𝘊𝘳𝘦𝘥𝘪𝘵: 𝘚𝘪𝘭𝘪𝘤𝘰𝘯 𝘝𝘢𝘭𝘦𝘳𝘪𝘦 𝘉𝘦𝘳𝘵𝘦𝘭𝘦 

1,393

156 Comments

The consensus in Silicon Valley right now is that "Vertical AI" is the only defensible moat left for startups. Zach from Adam (YC W25) disagrees, arguing that general agents are actually the ones winning the long-game. By stripping away domain-specific parameters and focusing on general programmatic execution, they unlocked a higher level of performance. It’s a counter-intuitive strategy that challenges the "Vertical AI" thesis many VCs are currently betting on. Watch why the most successful AI startups are pivoting away from niche specialization before the general models catch up. 🔗 Link in the comment section

4

1 Comment

🤖The Truth Spectrum ⛓️ Generative AI is redefining truth. Truth is no longer black and white. Blockchain is validating how truth should be considered. For years we’ve talked about “deepfake vs real” as if that’s the core spectrum. It isn’t. That framing mixes up three different questions: - Where did this come from? - How much has it been manipulated? - Is it trying to deceive? A better framework: Spectrum for Truth AI-composed <—> Reality-captured On one end, AI-composed content: text, images, video that models assemble from patterns, not from a specific real-world event. On the other end, reality-captured content: cameras, microphones, sensors, human eyewitness accounts. Once you consider a spectrum for truth, you can layer the attributes we actually care about: - Authentic vs Counterfeit – is it representing reality faithfully, or impersonating / fabricating? - Licensed vs Unlicensed – do the creator and subjects have rights and consent, or not? - Verified vs Unverified – can we prove origin and edit history, or are we guessing? - Inspired vs Remixed – is it a new work influenced by sources, or a transformation of specific originals? That gives you a richer spectrum: - from counterfeit fabrications and unlicensed clones at the far AI-composed end, - through synthetic simulations and labeled reconstructions, - into AI-assisted hybrids (summaries, translations, restorations), - and finally into provenance-verified, rights-cleared, reality-captured media. In this world, “deepfake” becomes just one corner case: AI-composed + counterfeit + often unlicensed + unverified. And “real” is no longer “I saw it on video,” but reality-captured + verified + authentic + appropriately licensed. This is the potential enabled by blockchain. Platforms, regulators, brands, and creators will need to: - Design policies on the full spectrum, not just ban “deepfakes.” - Label content by origin (AI-composed vs reality-captured) and attributes (authentic, counterfeit, licensed, verified, inspired, remixed). - Build incentives and infrastructure for provenance and authenticity, not nostalgia for a pre-AI world that’s already gone. We’re not going back to black and white in truth. There is a spectrum for truth, in today’s AI and blockchain world. The opportunity now is to define a new, shared language for truth in an AI-native internet and build the blockchain systems that validate that truth spectrum visible to everyone.

9

1 Comment

Sometimes the most elegant solutions are also the simplest. Early in Taiv's journey, they faced a critical product challenge: their AI sometimes made mistakes during live sports games. They needed a fail-safe that customers could use instantly. Their Y Combinator partner, Eric Migicovsky (founder of Pebble SmartWatch), gave them unconventional but brilliant advice: "Why don't you just take a big red button right on the bar and they can just smash it when they're angry?" They built it. They shipped custom hardware with literal big red buttons to every customer. For years, this simple solution: - Gave customers immediate control - Built trust during the product's imperfect early days - Provided invaluable feedback data - Ultimately saved their business The lesson? Don't over-engineer. Sometimes the most human-centered solution—even if it involves physical hardware—is exactly what your product and customers need. What's the simplest solution that solved a complex problem in your business? Watch the full story on Breakout: https://lnkd.in/gtWRY8vP

1

🚨🚨🚨 App Launch Strategy + Prosperity AI System Launching an app isn’t just about building. It’s about orchestrating clarity, alignment, and compounding momentum. Here’s how Prosperity AI structures it from day zero 👇 1️⃣ 5 Core Dimensions - Clarity Phase Before launch, master the fundamentals: • Market → Who exactly are you serving? Early adopters or mainstream users? • Customer Pain → What problem do you solve in their words, not yours? • Offer → Launch with a minimum wow product, not a maximum feature list. • Value → Quantify real benefits: time saved, money gained, risks reduced, joy created. • Differentiation → Why you vs. five others in your category? Find the positioning edge. 🧭 Prosperity AI’s Pre-AI Review flags early blind spots like fuzzy ICPs or overly broad messaging. 2️⃣ Pre-AI Review - Diagnosis Phase Before scaling spend, stress-test your strategy: • ✅ Consistency Check: Does the product truly match the stated problem? • 🔁 Feedback Loop: Run a 50-user pilot → measure activation, retention, referral. • ⚠️ Red Flags: Misaligned pricing, wrong channels, or unclear offer–audience fit. 🧩 This phase prevents the classic “launching into the void” mistake. 3️⃣ 22 Advanced Dimensions - Expansion Phase Turn clarity into a repeatable launch engine: GTM Strategy → Channel mix: App Store, Product Hunt, LinkedIn, X, and niche communities. → Plan for the first 1,000 users: incentives, waitlists, ambassador loop. Pricing & Monetization → Test models: freemium, subscription, or hybrid. → Track conversion and churn from day one. Onboarding & Retention → Simplify signup (3 steps max). → Add AI-powered in-app hints to boost activation. Referral & Growth Loops → Reward users for inviting 3–5 others. → Optimize viral mechanics before ad spend. Ops & Finance → Split budget: 40% product, 40% GTM, 20% experiments. → Match runway to measurable growth milestones. 4️⃣ Full AI Review - Optimization Phase Prosperity AI runs a personalized strategic audit: • 📊 Insights: Which channels convert vs. burn. • 🧠 Risks: Feature creep, ICP drift, cost inefficiency. • 🎯 Recommendations: Double down where ROI compounds. • 💹 Metrics: CAC, retention, LTV, referral %, benchmarked vs. peers. This becomes your dynamic execution playbook. 5️⃣ Execution Roadmap - Living Strategy A real launch is a loop, not a single event: Phase Weeks 0–2 Build anticipation: waitlist, content drip, ambassadors. Weeks 2–6 Launch with high-impact PR + community seeding. Months 2–3 Iterate weekly from user feedback. Months 3–6 Activate referral loops, micro-influencers, partnerships. Quarterly Run Full AI Review → eliminate waste → scale what compounds. ✅ Core Principle: Never scale ads or sales before achieving clarity → alignment → validation. With Prosperity AI, every launch step is backed by data, updated dynamically, and designed to evolve with you. 🌐 www.prosperityai.ai 🎥 YouTube: https://lnkd.in/dDTh-kcA ~~||~~~

4

Oren Zeev is truly one of a kind in venture capital. The man manages $2.7BN AUM as a solo capitalist. He has 30% carry. (wow) He is the largest LP in every single one of his funds. (13% per fund = over $20M per fund, wow again!!!) He does not adhere to traditional rules around deployment period or cross-fund investing. The man is the most authentic and awesome GP in venture. Spotify 👉 https://lnkd.in/eJdyCU7z Youtube 👉 https://lnkd.in/egHy2Y7Z Apple Podcasts 👉 https://lnkd.in/eMs8RUm4 My 6 takeaways with oren zeev 👇 1. The Best Companies Look Weird and That Way They Have a Head Start - If the company looks weird or wrong, there probably won’t be 15 other startups doing it. - You will have 2-3 years without real competition & can build a real moat. - If you happen to be right, these will be some of the greatest outcomes. 2. Investing Lessons From the Boom of 2021 - In 2021, I probably paid 3-4x more than I should have for almost every deal. - Some of them are still going to be winners despite overpaying. - In any given market, you want to do the best deals you can, and some vintages are going to be better than others. 3. 50% of Funds Will Go Out of Business - If you are not a platform, then it is much harder for you to raise. - At least 50% of funds today cannot raise or aren't sure if they’re able to. - Many of them are not going to be able to raise. 4. My Advice to Founders on Pre-Emptive Rounds - Take the money, but continue to behave as if you didn’t. - Companies can be overfunded & lead to a loss of focus. - Spend it based on the signals from the market, not the pressure from the boardroom. 5. What I Want to See When Leading a Series A: - I want to see real risk reduced, not better-looking milestones. - Logos without renewal are just noise. - You want to see real product-market fit. 6. Why You Shouldn't Judge a Decision by the Outcome    - Like poker, you can have the odds in your favor, make the right decisions, and still lose. - Judge the process, not one result, or you’ll learn the wrong lesson. - If you make the right decisions, you will win eventually. #founder #funding #business #investing #vc #entrepreneur #startup #investment #20VC

167

40 Comments

DeepSeek-V3.2 is also available on Atlas Cloud now. We worked with SGLang to fix tool-use in non-streaming mode. The model now supports tool_call, tool_choice, response_format and reasoning. Price: $0.28/M Input & $0.4/M Output. Try it on Atlas Cloud: https://lnkd.in/gUb9i_B8

7

The virtual machine is dead. Long live the virtual machine. Cloud was supposed to kill the single VM but with what's happening in AI right now you should expect to see a resurgence of virtual machines, folks just spinning up single VMs, no orchestration, just running individual Clawdbots/Moltbots/Claude Code setups. Why? - no need to root your home machine - isolate it in a VM, kill it easily if you need to - LLMs can coach you (yes, even if you are not technical) through managing it quite easily

6

1 Comment

At Sequoia, how you pass matters as much as how you win. In this clip, Sequoia partners explain that a “no” is still a chance to build trust, especially when it’s done with care. Mike Moritz is the classic example. He passed on Zappos multiple times before investing. Each time, he did real diligence, gave clear and thoughtful feedback, and framed it as “maybe not now”, not a flat rejection. The standard is simple, but hard to fake: ▫️ Show real respect for the problem ▫️ Be specific about what’s missing ▫️ Help the founder get better, even without a check That’s how relationships compound. And why great companies often come back. I broke this clip and the full Sequoia Playbook, including 10 systems most firms never had the patience to build, here: https://lnkd.in/e9ubxeyM How do you usually say no when there’s a real founder on the other side of the table?

13

3 Comments

Can AI invent new materials? I brought together three deep-tech investors (Voyager Ventures, APEX Ventures, Contrarian Ventures) and asked them the hard questions. Matthew, Robina, Vera. Welcome to the show! In this episode, we talked about: 👉 Why AlphaFold worked, and why materials are fundamentally harder. 👉 Why synthesis and characterization, not models, are the real moat, using altrove as an example. 👉 Vertical vs generalist: who actually wins the race to new materials? Listen on Spotify: https://lnkd.in/e4y2NQkv (or wherever you get your podcasts) Let's go! --- Thanks to EUVC for supporting the podcast. 🙌

60

11 Comments

Step step step… “This strategic acquisition enhances the realism and diversity of Applied Intuition’s digital human library — virtual representations used to safely and effectively test autonomous systems in simulation. Reblika’s technology enables the creation of high-fidelity avatars that vary in gender, age, body type, ethnicity, clothing and accessories, and are fully animatable to perform complex, realistic interactions with vehicles and environments.”

10

💥 Some startups build apps. Others build the infrastructure behind engineering, finance, and AI workflows. This US lineup just raised and teams are expanding. 🇺🇸 Axiomatic_AI (Cambridge, MA) #AI Engineering systems require provable reasoning. Axiomatic AI raised $18M Seed led by Engine Ventures, with Kleiner Perkins, Big Sur Ventures, Propagator Ventures, and others, bringing total funding to $25M. Founded by Marin Soljacic, Frank Koppens, Joyce Poon, led by CEO Jake Taylor, the company builds verification infrastructure combining AI with physics-based validation for semiconductors and advanced manufacturing. ❗ Hiring: 14 openings - https://bit.ly/4bqMxeu 🇺🇸 KAST (New York, NY) #FinTech Cross-border money movement remains slow and fragmented. KAST raised $80M Series A co-led by QED Investors and Left Lane Capital, with Peak XV Partners, and others. Founded by Raagulan Pathy, James Butland, and Sam Kerrins, the platform lets users hold, send, and spend digital dollars globally through stablecoin infrastructure. ❗ Hiring: 18 openings - https://bit.ly/4cP9gTU 🇺🇸 DiligenceSquared (YC F25) (New York, NY) #FinTech Commercial due diligence remains time-consuming for investment teams. DiligenceSquared raised $5M Seed led by Relentless, with Y Combinator. Founded by Frederik Kofoed Hansen, Søren Biltoft-Knudsen, and Harshil Rastogi, the platform uses AI to automate expert interviews, analysis, and reporting for private equity firms. ❗ Hiring: 4 openings - https://bit.ly/4uDlKV5 🇺🇸 Mega (New York, NY) #MarTech Marketing execution is shifting to autonomous systems. Mega raised $11.5M Series A led by Goodwater Capital, with Andreessen Horowitz, Atreides, SignalFire, and Kearny Jackson. Founded by Robbie Schneidman and Lucas Pellan, the platform deploys AI agents that continuously optimize search rankings, ads, and websites for SMBs. ❗ Hiring: 6 openings - https://bit.ly/4uLic38 🇺🇸 Dify (Menlo Park, CA) #DevTools Agentic workflows are becoming a core layer of enterprise AI infrastructure. Dify raised $30M Series Pre-A led by HSG, with Alt-Alpha Capital, 5Y Capital, and others. Founded by Luyu Zhang, John Wang, and Junchen Yan, the open-source platform lets teams build and run AI workflows with a visual builder and production infrastructure. ❗ Hiring: 16 openings - https://bit.ly/46YyF9Y 🇺🇸 Yourco (Chicago, IL) #HRTech Connecting frontline workers remains a blind spot for many organizations. Yourco raised $6M Series A led by High Alpha. Founded by Brodie Meyer and Benjamin Meyer, the SMS platform helps companies communicate with deskless employees across 135+ languages without apps or company email. ❗ Hiring: 6 openings - https://bit.ly/4dmBF3B Funding is the milestone. Execution starts the day after. If you’re facing post-raise people decisions around hiring, people planning, and execution risk, I open limited office hours for founders and People leaders on what comes next after a raise. DM me or comment below for a 30-min slot.

3

What happens when boomers exit the market? | A clip from my Substack post, EP028: The Architect Who Became the Grey Rabbit of Finance. Watch it here: https://lnkd.in/eiT_7Dse

1

Sustained ~252–291 GiB/s of HBM memory throughput on H100s under decode load — essentially hitting the hardware roofline. This matters because HBM throughput, not FLOPs, is the real bottleneck in LLM inference. By keeping memory nearly fully saturated, I’ve unlocked far higher efficiency and throughput than standard engines. The result: 0.9–1.36M tokens/sec with ~0.1 ms first-token latency  #AI #LLM #GPURouter #H100 #Inference #CostEfficiency #Innovation #DeepLearning #AIInfrastructure #HighPerformanceComputing NVIDIA AMD OpenAI Google Shilpa Kolhatkar Keith Strier a16z speedrun

5

The "Fake It Till You Make It" Era is Over: Why Deep Tech Requires Real Science. Watching the film about Elizabeth Holmes and the Theranos collapse is a stark reminder of Silicon Valley's most dangerous myth: the idolization of the "dropout founder." In consumer software, "move fast and break things" built unicorns. But we are entering the deep tech decade, and the rules of engagement have fundamentally changed. In frontier space infrastructure and breakthrough deep tech, breaking things doesn't mean a software patch. It means losing a $500M payload or setting critical scientific progress back by a decade. The laws of physics do not care about a charismatic pitch deck. You cannot "fake it till you make it" with orbital mechanics, quantum computing, or biotech. That is why the era of the dropout is making way for the era of the Doctorate. At Beyond Earth Ventures, we don't rely on hype—we rely on peer-reviewed reality to finance technologies shaping the next trillion-dollar economy. That is exactly why our core team includes three PhDs, and why our scientific diligence is guided by a Nobel Prize laureate as an advisor. If your diligence team doesn't have the scientific literacy to read the underlying academic papers, you aren't investing in deep tech. You're just gambling.

43

9 Comments

If you’re using GPT to generate business ideas, watch this video. It will help you understand that LLMs don’t have opinions - they simply follow your prompts. If you ask them to “support” an idea, they will. If you ask them to “critique” an idea, they will. If you ask them to “kill” an idea, they will. The way you prompt determines the outcome - even on the exact same subject. ♻️ Subscribe to ‘Siliconnector’ Telegram channel for insights, news and fun from Silicon Valley and the Tech world: https://t.me/siliconnector

15

1 Comment