<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Nese Ozler on Medium]]></title>
        <description><![CDATA[Stories by Nese Ozler on Medium]]></description>
        <link>https://medium.com/@neseozler?source=rss-20ba06d99a0f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*G0VqyilHBuF-8JWP</url>
            <title>Stories by Nese Ozler on Medium</title>
            <link>https://medium.com/@neseozler?source=rss-20ba06d99a0f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 16:07:06 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@neseozler/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Week 9/2026 in Silicon Valley]]></title>
            <link>https://medium.com/@neseozler/week-9-2026-in-silicon-valley-fb1fd266f2f5?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/fb1fd266f2f5</guid>
            <category><![CDATA[openclaw]]></category>
            <category><![CDATA[unstructuredsignal]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[consumer-ai]]></category>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Mon, 02 Mar 2026 23:19:10 GMT</pubDate>
            <atom:updated>2026-03-02T23:19:10.094Z</atom:updated>
            <content:encoded><![CDATA[<p>Let’s take a look back at my tweets (<a href="https://x.com/nesozl">@nesozl</a>) from last week to understand what dominated the conversation in Silicon Valley.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eHyHbQYYUTFVxRgJ2SinrQ.jpeg" /></figure><h3>The “SaaS-pocalypse”: AI Agents Erasing the Legacy Moat</h3><blockquote>Anthropic’s Claude Code also has triggered a “SaaS-pocalypse.” IBM stock plunged 13% on Monday (Feb 23) after Anthropic showed how AI could automate legacy COBOL modernization <a href="https://x.com/hashtag/UnstructuredSignal?src=hashtag_click">#UnstructuredSignal</a></blockquote><p><strong>From Copilot to Autopilot:</strong> AI is no longer just assisting developers; it is replacing entire legacy workflows. Anthropic’s Claude Code recently demonstrated the total automation of COBOL modernization, sending shockwaves through traditional IT and triggering a 13% plunge in IBM stock in a single day.</p><blockquote><a href="https://x.com/hashtag/Anthropic?src=hashtag_click">#Anthropic</a> Claude Code Security launch triggered a “<a href="https://x.com/hashtag/SaaSpocalypse?src=hashtag_click">#SaaSpocalypse</a>,” wiping billions in market value. Major stocks like <a href="https://x.com/hashtag/CrowdStrike?src=hashtag_click">#CrowdStrike</a>, <a href="https://x.com/hashtag/Datadog?src=hashtag_click">#Datadog</a>, and <a href="https://x.com/hashtag/Zscaler?src=hashtag_click">#Zscaler</a> plunged up to 11% in a single session as investors fear AI-driven disruption <a href="https://x.com/hashtag/UnstructuredSignal?src=hashtag_click">#UnstructuredSignal</a></blockquote><p><strong>Market Value Destruction:</strong> The immediate stock plunges of legacy giants like IBM, CrowdStrike, Datadog, and Zscaler highlight a massive shift in investor confidence. Capital is actively abandoning traditional SaaS models in fear of AI-driven disruption.</p><blockquote>The power law is accelerating as capital concentrates into a tiny set of companies. AI excitement meets liquidity anxiety while LPs rotate toward deep tech and Physical AI. The gap between AI-native winners and legacy software is widening fast. Adapt or be replaced <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/VC?src=hashtag_click">#VC</a></blockquote><p><strong>The Widening Gap:</strong> The “power law” in venture capital is accelerating because the gap in capabilities between AI-native startups and legacy software is expanding too fast for older companies to adapt.</p><h3>The Compute Divide: Multi-Cloud and the Hardware Hustle</h3><blockquote>#OpenAI latest funding -$110B raised + $10B more by March</blockquote><blockquote>-#Amazon, #Nvidia, #SoftBank</blockquote><blockquote>-$850B valuation</blockquote><blockquote>-$219B projected cash burn through 2029</blockquote><blockquote>-Profit: Projected $14B by 2029</blockquote><blockquote>-Cash flow + by 2030 Amazon’s $35B hinges on OpenAI hitting #AGI or an IPO #UnstructuredSignal</blockquote><p><strong>The High-Stakes Survival Math:</strong> OpenAI’s eye-watering $850B valuation is built on a mountain of capital and an unprecedented projected cash burn of $219B through 2029. With mega-backers like Amazon tying $35B specifically to OpenAI reaching AGI or an IPO, the company is racing against a ticking financial clock. The path to a projected $14B profit is paved with existential risk.</p><blockquote><a href="https://x.com/hashtag/Amazon?src=hashtag_click">#Amazon</a> is betting big on OpenAI with a massive $50B investment:</blockquote><blockquote>-From $38B to $100B expanded AWS server rental</blockquote><blockquote>-2GW of power via Amazon Trainium chips</blockquote><blockquote>-AWS is now the exclusive “third-party” cloud distributor for OpenAI Frontier <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> <a href="https://x.com/hashtag/AWS?src=hashtag_click">#AWS</a> <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/UnstructuredSignal?src=hashtag_click">#UnstructuredSignal</a></blockquote><p><strong>The Cost of Survival:</strong> With a projected $219 billion cash burn through 2029, securing compute power is existential. Mega-investments like Amazon’s $50 billion are funding dedicated power grids (2GW) and custom silicon (Trainium).</p><blockquote>OpenAI shifts its cloud strategy:</blockquote><blockquote>-<a href="https://x.com/hashtag/Amazon?src=hashtag_click">#Amazon</a>: $50B investment + exclusive “Stateful” runtime (memory/agents) on <a href="https://x.com/hashtag/AWS?src=hashtag_click">#AWS</a></blockquote><blockquote>-<a href="https://x.com/hashtag/Microsoft?src=hashtag_click">#Microsoft</a>: Exclusive home for “Stateless” API calls (one-off tasks) <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> Frontier goes multi-cloud, moving context-heavy loads to AWS <a href="https://x.com/hashtag/UnstructuredSignal?src=hashtag_click">#UnstructuredSignal</a></blockquote><p><strong>Strategic Decoupling:</strong> OpenAI is moving away from sole reliance on Microsoft by heavily integrating AWS. Splitting workloads into “Stateful” memory operations on AWS and “Stateless” API calls on Microsoft is a massive architectural shift to manage context-heavy agent tasks.</p><blockquote>To scale and survive massive inference costs, OpenAI is securing compute via deals with <a href="https://x.com/hashtag/Microsoft?src=hashtag_click">#Microsoft</a>, <a href="https://x.com/hashtag/AWS?src=hashtag_click">#AWS</a>, <a href="https://x.com/hashtag/AMD?src=hashtag_click">#AMD</a>, <a href="https://x.com/hashtag/Nvidia?src=hashtag_click">#Nvidia</a>, <a href="https://x.com/hashtag/Broadcom?src=hashtag_click">#Broadcom</a> &amp; <a href="https://x.com/hashtag/Cerebras?src=hashtag_click">#Cerebras</a>. Blending traditional GPUs with efficient SRAM tech from Cerebras allows them to lower costs and fuel future growth <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> <a href="https://x.com/hashtag/AIHardware?src=hashtag_click">#AIHardware</a></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/0*edKs0kDkBh0_ogE9" /></figure><p><strong>Hardware Diversification:</strong> To survive massive inference costs, AI leaders are blending traditional GPUs with specialized chips (like SRAM tech from Cerebras) to drive down the cost per query.</p><h3>The Market Share Mirage: First-Mover Advantage is Bleeding</h3><blockquote>The mobile AI race is heating up!</blockquote><blockquote>OpenAI’s ChatGPT share dropped from 69.1% to 45.3% as the market grew 152% Gemini is the clear winner with steady growth</blockquote><blockquote><a href="https://x.com/hashtag/DeepSeek?src=hashtag_click">#DeepSeek</a> peaked and then missed. Even Copilot is outperforming Claude and Perplexity! <a href="https://x.com/hashtag/Gemini?src=hashtag_click">#Gemini</a> <a href="https://x.com/hashtag/ChatGPT?src=hashtag_click">#ChatGPT</a> <a href="https://x.com/hashtag/Chatbot?src=hashtag_click">#Chatbot</a></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/0*uNGvSqFMLLlETVq1" /></figure><p><strong>Consumer Volatility:</strong> OpenAI is losing its dominant grip on mobile. Dropping from a 69% to a 45% share as Gemini captures new growth proves that users are fiercely loyal to capability, not brand.</p><blockquote><a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> missing its 1B user goal by 2025(hitting 910M instead) proves that ambitious 2030 targets of 2.75B (33% of the world) can’t mask the reality of a shifting market. Even with new device embedding 33% is nearly double the <a href="https://x.com/hashtag/iPhone?src=hashtag_click">#iPhone</a>’s 17% global penetration <a href="https://x.com/hashtag/UnstructuredSignal?src=hashtag_click">#UnstructuredSignal</a></blockquote><p><strong>The Apple Reality Check:</strong> OpenAI is targeting 33% global penetration by 2030, yet the iPhone only sits at 17%. Missing their 2025 user goals proves that without a dominant, physical device strategy, scaling pure software to a third of the planet is an unrealistic uphill battle.</p><blockquote><a href="https://x.com/hashtag/VoiceCommerce?src=hashtag_click">#VoiceCommerce</a> is peaking in India, where WhatsApp voice memos drive business. For <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> to win it must bridge the gap:India has 100M weekly users (11% of 910M) yet ChatGPT supports only 37 spoken vs 80 written languages missing most of India’s official 22 <a href="https://x.com/hashtag/UnstructuredSignal?src=hashtag_click">#UnstructuredSignal</a></blockquote><p><strong>The Localization Blind Spot:</strong> Raw AI power does not equal global dominance. India’s digital economy runs on voice commerce, yet ChatGPT’s spoken language capabilities severely lag behind its text features. By missing the majority of India’s 22 official languages, OpenAI is leaving a critical 100 million user demographic highly vulnerable to local competitors. This strategic gap was brought into sharp perspective just last week during the highly publicized travels of OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei to New Delhi, proving that while AI leaders are physically rushing to court the Indian market, their core products still have a long way to go to truly speak the language.</p><blockquote><a href="https://x.com/hashtag/Anthropic?src=hashtag_click">#Anthropic</a> Claude is seeing a massive consumer surge -Daily signups 3x since Nov -Paid subscribers 2x+ since Oct -Free users up 60% in the last month -2025 revenue: $4.5B ($600M from chatbot) Growth is fueled by Claude Code &amp; promise of an ad-free experience <a href="https://x.com/hashtag/UnstructuredSignal?src=hashtag_click">#UnstructuredSignal</a></blockquote><p><strong>Anthropic’s Consumer Surge:</strong> While OpenAI fights battles on multiple fronts, Claude is quietly capturing a massive consumer surge, tripling signups and doubling paid subscribers fueled by the promise of ad-free, high-quality coding experiences.</p><h3>Enterprise Proxies &amp; Walled Gardens</h3><blockquote>OpenAI’s “Frontier Alliances” with McKinsey, Accenture, BCG, and Capgemini is an aggressive attempt to grab enterprise share. By turning top consultants into a certified sales force for its agent platform, OpenAI aims to push its enterprise rev from 40% toward 50% by year-end</blockquote><p><strong>Consultants as Sales Reps:</strong> OpenAI’s “Frontier Alliances” with McKinsey, BCG, and others is a brilliant, aggressive move. Instead of building a massive internal sales force, they are weaponizing the world’s top consultants to push their enterprise revenue toward 50%.</p><blockquote>The AI “Data War” is heating up as platforms block automated agents: — <a href="https://x.com/hashtag/Google?src=hashtag_click">#Google</a> restricts <a href="https://x.com/hashtag/OpenClaw?src=hashtag_click">#OpenClaw</a> from Antigravity -<a href="https://x.com/hashtag/Anthropic?src=hashtag_click">#Anthropic</a> blocks Claude Code via OpenClaw -Salesforce/Glean style: Guarding the moat The wall is going up against AI-on-AI automation <a href="https://x.com/hashtag/UnstructuredSignal?src=hashtag_click">#UnstructuredSignal</a></blockquote><p><strong>The Data Blockade:</strong> As AI agents become autonomous, the “Data War” is escalating. Platforms are building walls to protect their proprietary data from being scraped or automated by competitor AIs, evidenced by Google and Anthropic blocking OpenClaw.</p><h3>Summary</h3><p><strong>The Old Moat is Gone: Adapt or Die</strong> The era of legacy SaaS is being crushed by autonomous agents, while the consumer AI landscape fragments under the surge of Gemini and Claude. Behind the scenes, the survival of AI giants now hinges on managing massive cash burns by splitting cloud workloads, hoarding compute, and building impenetrable data walls to capture the enterprise market.</p><p>#agents #ConsumerAI #OpenAI #Anthropic #Cloud #Compute #Enterprise #OpenClaw #UnstructuredSignal</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fb1fd266f2f5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 8/2026 in Silicon Valley]]></title>
            <link>https://medium.com/@neseozler/week-8-2026-in-silicon-valley-30db6f3fdb3d?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/30db6f3fdb3d</guid>
            <category><![CDATA[agent-wars]]></category>
            <category><![CDATA[continual-learning]]></category>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[private-equity]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Sat, 21 Feb 2026 00:23:05 GMT</pubDate>
            <atom:updated>2026-02-21T18:02:00.606Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aGn_RcgX5wpsh_y6ITfCJg.jpeg" /><figcaption>Created by Gemini</figcaption></figure><p>Let’s take a look back at my tweets (<a href="https://x.com/nesozl">@nesozl</a>) from last week to understand what dominated the conversation in Silicon Valley.</p><h3>Compute is the Ultimate Moat: Why Amazon and Microsoft Always Win</h3><blockquote>Amazon reportedly takes ~50% of Anthropic’s gross profits (winning on compute + split). Google? ~20–30% of net revenue. Microsoft: 75% of <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> profits until repaid, then 49% until capped. Different metrics, same goal: capturing the AI value stream <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/Cloud?src=hashtag_click">#Cloud</a> <a href="https://x.com/hashtag/TechNews?src=hashtag_click">#TechNews</a></blockquote><p>Amazon’s reported 50% cut of Anthropic’s gross profits is a brutal infrastructure tax that exposes a universal industry reality. Even as Microsoft renegotiated its aggressive 75% profit recovery into a 20% revenue cut and equity stake, they still locked OpenAI into a staggering $250 billion Azure commitment. While contract metrics fluctuate, the dynamic is identical. Foundation model companies do the heavy lifting on innovation, but their total dependence on hyperscalers for compute turns cloud providers into inescapable toll booths. No matter which AI startup wins the intelligence race, the infrastructure house always captures the ultimate value.</p><h3>Agent Wars: How OpenAI and Anthropic are Fighting for Dominance</h3><blockquote><a href="https://x.com/hashtag/Anthropic?src=hashtag_click">#Anthropic</a> is setting the IPO standard. <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> with 85% consumer revenue, risks looking like the next <a href="https://x.com/hashtag/Facebook?src=hashtag_click">#Facebook</a>. Hiring Peter is the antidote: a signal that OpenAI builds tools, not feeds, and is still a developer-first platform. <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> <a href="https://x.com/hashtag/Anthropic?src=hashtag_click">#Anthropic</a> <a href="https://x.com/hashtag/IPO?src=hashtag_click">#IPO</a> <a href="https://x.com/hashtag/ConsumervsEnterprise?src=hashtag_click">#ConsumervsEnterprise</a></blockquote><blockquote><a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> lands <a href="https://x.com/hashtag/OpenClaw?src=hashtag_click">#OpenClaw</a> to block <a href="https://x.com/hashtag/Meta?src=hashtag_click">#Meta</a> and revive <a href="https://x.com/hashtag/Codex?src=hashtag_click">#Codex</a>, but the signal is clear: OpenClaw chose the home where consumer comes first. Now Anthropic needs its own equivalent to lock down the enterprise. The agent wars are just getting started. <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> <a href="https://x.com/hashtag/OpenClaw?src=hashtag_click">#OpenClaw</a> <a href="https://x.com/hashtag/Anthropic?src=hashtag_click">#Anthropic</a> <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a></blockquote><blockquote><a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> acquiring <a href="https://x.com/hashtag/OpenClaw?src=hashtag_click">#OpenClaw</a> is a strategic masterstroke: blocking <a href="https://x.com/hashtag/Meta?src=hashtag_click">#Meta</a>, fueling the autonomous agent narrative, and reviving <a href="https://x.com/hashtag/Codex?src=hashtag_click">#Codex</a>. With Anthropic’s Claude Code dominating dev workflows, this is the jolt OpenAI needs to reclaim the lead. The agent wars are on. <a href="https://x.com/hashtag/AgentWars?src=hashtag_click">#AgentWars</a> <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a></blockquote><p>OpenAI’s acquisition of OpenClaw is a direct counterpunch to the rise of Claude Code and Anthropic’s aggressive enterprise land grab. On the surface, OpenAI’s recent momentum makes them look like a pure consumer company, but Sam Altman clearly wants it all: total dominance across both consumer and enterprise markets. By locking down a massive, developer-focused tool like OpenClaw, OpenAI is building the vital bridge it needs to reach enterprise customers, deploying the exact same playbook that made Claude Code such a massive success for Anthropic.</p><h3>The Non-SaaS-calypse: Private Equity’s New AI Playbook</h3><blockquote>Is <a href="https://x.com/search?q=%24TRIP&amp;src=cashtag_click">$TRIP</a> the first target of the “Non-SaaS-calypse”? Starboard Value (9% stake) is overhauling the board, claiming TripAdvisor is “massively late” on GenAI Stock down 40% in 6mo = vulnerability. The new PE roadmap: Legacy consumer tech must integrate AI or die <a href="https://x.com/hashtag/PrivateEquity?src=hashtag_click">#PrivateEquity</a></blockquote><p>Legacy consumer tech companies are waking up to a brutal new reality: integrate generative AI immediately or lose control of your company. As seen with Starboard Value’s aggressive proxy fight at TripAdvisor, failing to adapt to the AI era leads to plummeting valuations, which instantly attracts activist investors looking to overhaul the board or force a sale. Being “late to AI” is no longer just a product disadvantage; it is the primary trigger for corporate restructuring.</p><h3>Beyond the Scratchpad: The Architectural Reality of Continual Learning</h3><blockquote>“Continual learning” is the new AI holy grail. Startups are pitching workarounds like scratchpads &amp; test-time training as sols, but investors warn that true real-time learning requires fundamental architectural innovation 🧠🤖 #ContinualLearning #TestTimeTraining #DeepLearning</blockquote><p>Startups are currently applying clever patches like test time training and extended scratchpads to mimic real time adaptation. However, the investment community recognizes that reaching AI’s ultimate goal of continual learning will require entirely new foundational architectures, not just temporary fixes on top of existing models.</p><h3>Takeaway</h3><p>The AI ecosystem is maturing faster than anyone predicted. The foundational models are locked in a battle for distribution, legacy companies are fighting for their lives, and the hyperscalers are quietly collecting the rent on the underlying infrastructure. But the true paradigm shift is yet to come. The companies that solve the architectural puzzle of true continual learning will not just win the next round of the Agent Wars; they will fundamentally rewrite the economics of the entire stack. Until then, keep an eye on the infrastructure!</p><p>#AgentWars #OpenAI #Anthropic #Cloud #PrivateEquity #ContinualLearning</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=30db6f3fdb3d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 4/2026 in Silicon Valley]]></title>
            <link>https://medium.com/@neseozler/week-4-2026-in-silicon-valley-9432d6add1c6?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/9432d6add1c6</guid>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[google-cloud]]></category>
            <category><![CDATA[apple]]></category>
            <category><![CDATA[futureof-work]]></category>
            <category><![CDATA[gemini]]></category>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Mon, 26 Jan 2026 00:20:46 GMT</pubDate>
            <atom:updated>2026-01-26T00:20:46.840Z</atom:updated>
            <content:encoded><![CDATA[<p>Let’s take a look back at my tweets (<a href="https://x.com/nesozl">@nesozl</a>) from last week to understand what dominated the conversation in Silicon Valley.</p><h3>Monetizing the Flywheel: OpenAI’s Push from Viral Success to Industrial Scale</h3><blockquote><a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> is now chasing $100M+ multi-year contracts to lock in the biggest enterprise players. Anthropic’s “business-first” focus has carved out a massive trust advantage that is hard to replicate. Can <a href="https://x.com/hashtag/Codex?src=hashtag_click">#Codex</a> really catch up to the momentum of <a href="https://x.com/hashtag/ClaudeCode?src=hashtag_click">#ClaudeCode</a>? <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/Enterprise?src=hashtag_click">#Enterprise</a></blockquote><p>OpenAI started 2026 strong by proving they aren’t just a research lab but they are an ecosystem. Much like Google successfully paired its B2C roots in search and ads with the B2B powerhouse of Google Cloud, OpenAI is now executing a massive dual-front offensive.</p><p>To sustain its massive capital requirements, OpenAI is pivoting toward “mega-deals.” However, they are facing a “Trust Gap” where Anthropic’s safety-first reputation is forcing OpenAI to use aggressive pricing and multi-year lock-ins to compete for the Fortune 500.</p><blockquote><a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a>’s growth is on a massive trajectory!</blockquote><blockquote>- Compute Power:⚡️ 2023: 0.2 GW ⚡️ 2024: 0.6 GW ⚡️ 2025: 1.9 GW<br>- Revenue (ARR):💰 2023: $2B 💰 2024: $6B 💰 2025: $20B+</blockquote><blockquote>Both compute and ARR have tripled YoY, showing no signs of slowing down.</blockquote><p>OpenAI is successfully scaling at a rate rarely seen in tech history, maintaining a perfect 1:1 correlation between massive compute investment and revenue returns. They aren’t just spending money; they are effectively <strong>“monetizing silicon”</strong></p><blockquote>OpenAI’s “next phase” is officially here, per CFO Sarah Friar! 🚀<br>What’s coming: <br>🤖 Agents &amp; Automation: Systems running continuously with persistent context and take action across tools<br>📂 Knowledge Operating Layer: AI managing projects &amp; coordinating plans for organizations</blockquote><p>The focus is shifting from <strong>Chat</strong> (reactive) to <strong>Operating Systems</strong> (proactive). By moving toward “persistent context” and “agents,” OpenAI aims to become the invisible glue that manages an organization’s projects rather than just a window for queries.</p><blockquote>Tried to cancel my #ChatGPT Plus sub after jumping back to #GoogleGemini, but they lured me stay with a free month! 🎁 The battle for our loyalty is heating up. Are the GPT games officially on? 🍿</blockquote><blockquote>#AI #TechNews #OpenAI #ChatGPTPreset #ARR</blockquote><p>Despite the massive enterprise push, the consumer war is becoming a “street fight.” As rivals like Gemini reach parity, OpenAI is increasingly relying on aggressive retention tactics (the “churn defense”) to keep their 900M+ weekly user base from migrating.</p><h3>API Velocity: The New Pulse of Enterprise AI</h3><blockquote>#GoogleGemini 🚀<br>- API Demand: Doubled in 5 months ($35B$ to $85B$ requests) 1.42x +<br>- #GeminiEnterprise: Scaled to 8M subscribers</blockquote><blockquote>With OpenAI’s business user base recently topping 1M orgs, wonder what the total seat count for #ChatGPTEnterprise looks like right now? 🧐</blockquote><p>Gemini API demand just doubled in 5 months (85B requests!). Are we witnessing the end of the ‘Chatbot Era’ and the beginning of the ‘Infrastructure Era’? Is OpenAI’s platform lock-in strategy already too late?</p><p>The <strong>143% surge</strong> in Gemini API requests (jumping from 35B to 85B in just five months) signals that businesses are moving past the “chatting with a bot” phase and are now <strong>hard-coding Gemini into their own internal products.</strong> While OpenAI’s “1M organizations” metric sounds impressive, many of those are small-scale adoptions. Google’s <strong>8M Enterprise subscribers</strong> (a 300% YoY increase) suggests they are winning the “seat war” by simply flipping the switch for existing corporate Workspace users.</p><h3>The “AI-first” hardware race is accelerating</h3><blockquote><a href="https://x.com/hashtag/Apple?src=hashtag_click">#Apple</a> is reportedly developing an AI pin for 2027, aiming to succeed where <a href="https://x.com/hashtag/Humane?src=hashtag_click">#Humane</a> failed</blockquote><blockquote>The dual-camera, 3-mic disc, a thicker, glass-aluminum AirTag lookalike helps Apple to take on <a href="https://x.com/hashtag/Meta?src=hashtag_click">#Meta</a>’s smart glasses and <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a>’s hardware? Will Apple’s ecosystem be the winning edge?</blockquote><p>Apple is playing the long game, using its hardware mastery to pivot the AI Pin from a “solution in search of a problem” into a seamless extension of the iPhone. By integrating dual cameras and a 3-mic array into an AirTag-sized form factor, Apple aims to win on <strong>ambient awareness; </strong>making Siri the first assistant that actually sees what you see.</p><blockquote>OpenAI is officially on track to launch its first hardware device in H2 2026! Developed in collaboration with Jony Ive’s LoveFrom, this move signals a major shift beyond software for the company’s future ARR. 🚀</blockquote><blockquote>Any guesses on the form factor? 🧐</blockquote><blockquote>#OpenAI #AI #TechNews</blockquote><p>Partnering with Jony Ive isn’t just about design; it’s a strategic move to define the <strong>post-smartphone era</strong>. By targeting a 2026 launch (likely a screenless, audio-first wearable codenamed “Sweetpea”), OpenAI is betting that a “peaceful” hardware interface will unlock new levels of high-margin <strong>Hardware ARR</strong>, turning ChatGPT from an app into a lifelong companion.</p><h3>The End of Isolation: Apple’s Pragmatic Pivot to the Cloud</h3><blockquote>#Apple is answering the cry to fix #Siri by going full vertical integration with #GoogleCloud</blockquote><blockquote>This marks a major reversal from the original plan of only using on-device or Private Cloud Compute. Apple is now tapping Google TPUs for the 1.2T parameter brain it needs 🚀 #AI</blockquote><p>While the OpenAI deal was a tactical “plugin” to buy time, the Google partnership is a fundamental architectural shift. By moving the <strong>1.2T parameter brain</strong> to Google TPUs, Apple has officially ended its “on-device only” era, admitting that true World Knowledge requires the massive, vertical infrastructure that only Google Cloud can provide.</p><h3>The Enterprise Orbit: Satellite Internet Race</h3><blockquote>#BlueOrigin enters the #SatelliteInternet race with TeraWave, aiming for 6 Tbps speeds and 5,000+ satellites by Q4 2027. Unlike #Starlink (9M+ users) or #AmazonLeo (180 satellites), this service targets only high-end enterprise/gov clients. Competition is heating up. #SpaceX</blockquote><p>Blue Origin’s <strong>TeraWave</strong> marks a fundamental shift in the satellite race: the move from “Broadband for All” to <strong>“Infrastructure for the Few.”</strong> While <strong>Starlink</strong> (9M+ users) and <strong>Amazon Kuiper</strong> focus on the mass consumer and retail markets, Jeff Bezos is positioning TeraWave as a high-end, sovereign utility. By aiming for <strong>6 Tbps speeds</strong> and focusing exclusively on governments, data centers, and enterprise giants, Blue Origin is building a “private cloud in the sky.” This isn’t about connecting the rural world — it’s about providing the high-speed, low-latency backbone for the <strong>agentic enterprise infrastructure</strong> we’ve been tracking all week.</p><h3>Autonomous Driving: From Liability to Asset</h3><blockquote>Switching from Root to Lemonade, just activated car insurance! Surprised to see #Tesla did the same. Their new product cuts premiums by 50% for #FSD miles. #Robotaxis are top of mind; why own when you can rent to the fleet in future? Need a Tesla by V-Day! 🚀 #Tesla #Lemonade</blockquote><p>The launch of <strong>Lemonade’s “Autonomous Car” insurance, </strong>offering a <strong>50% premium discount</strong> for miles driven in FSD, is the first concrete economic proof that the car is evolving from a depreciating liability into a high-yield asset. By leveraging real-time Tesla data, insurers are finally pricing the “safety premium” of autonomy. This isn’t just about saving money; it’s about the shift to the <strong>Robotaxi fleet</strong> model. If your FSD miles are 50% cheaper to insure, the financial incentive moves from “personal driving” to “renting to the fleet,” turning the vehicle into a productive node in the “Hobby Economy.”</p><h3>The Post-Scarcity Pivot in other words Future of Work</h3><blockquote>Well written article about the facts about AI &amp; Future! <a href="https://x.com/hashtag/TechNews?src=hashtag_click">#TechNews</a> <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/FutureOfWork?src=hashtag_click">#FutureOfWork</a> <a href="https://x.com/hashtag/ARR?src=hashtag_click">#ARR</a> <a href="https://x.com/hashtag/AmericanDream?src=hashtag_click">#AmericanDream</a> Silicon Valley fears this is the last chance to amass generational wealth before AI makes money worthless, writes the WSJ’s Tim Higgins <a href="https://t.co/OQM9lyrQod">https://wsj.com/tech/ai/why-the-tech-world-thinks-the-american-dream-is-dying-daf793dc?st=tG7Pp8</a> via <a href="https://x.com/WSJ">@WSJ</a></blockquote><p>We are entering a “Post-Scarcity Scramble” where the traditional rules of wealth and work are being fundamentally rewritten. As Elon Musk recently emphasized at Davos 2026, the shift from Universal Basic Income to <strong>Universal High Income (UHI)</strong> means that in an AI-driven “benign scenario,” labor costs hit zero and money loses its significance as a constraint.</p><p>In this new reality:</p><ul><li><strong>The “Asset Vault” Strategy</strong>: San Francisco is no longer just a housing market; it is a desperate grab for physical scarcity in the world’s most concentrated agentic hub. Buying a home here isn’t an “investment” in the traditional sense, it’s securing a permanent seat at the table before the physical window closes.</li><li><strong>From Deals to Pipes:</strong> Leverage no longer comes from bank accounts, but from owning the “Hard Infrastructure” like the satellite networks, power grids, and AI agent layers that provide the “universal abundance”</li><li><strong>The Optional Economy:</strong> With work becoming a hobby (like backyard gardening), the only strategic move left is to shift from “working for a salary” to orchestrating the systems that deliver the world’s goods and services.</li></ul><h3>The Non-China Battery Corridor</h3><blockquote>🇨🇦 Canada x 🇸🇦 Saudi: A #battery powerhouse in the making!</blockquote><blockquote>#NorthernGraphite is teaming up with #Obeikan to build a $200M plant in #Yanbu, turning it into a hub for the non-China battery supply chain. 🔋🚀</blockquote><blockquote>Projected revenue 8x+ to $250M ARR! 📈</blockquote><blockquote>#BatteryTech #GraphiteRefining</blockquote><p>Northern Graphite’s partnership with Obeikan to build a <strong>$200M plant in Yanbu</strong> is a strategic masterpiece in geopolitical diversification. By creating a direct, traceable supply chain from mines in Namibia to refining in Saudi Arabia, they are building the first major <strong>non-China “power pipe”</strong> for battery anode materials. With projected revenue jumping <strong>8x to $250M ARR</strong>, this move positions Saudi Arabia as the critical bridge for Western OEMs who are desperate to de-risk their EV supply chains from Chinese dominance.</p><p>Week 4 of 2026 confirms that we have moved from the <strong>“Viral Chat”</strong> phase into the era of <strong>Hard Infrastructure</strong>, where winning depends on controlling the <strong>physical and digital “pipes”</strong>, from Google’s 1.2T parameter cloud brain to the graphite refineries in Saudi Arabia. As we pivot toward a <strong>“Post-Scarcity” economy of abundance</strong>, the only remaining strategic leverage is owning a seat in “asset vaults” like San Francisco or orchestrating the autonomous systems that will soon make traditional labor optional.</p><p>Is this the end of white collar jobs era?</p><p>#OpenAI #Gemini #GoogleCloud #Apple #FutureOfWork</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9432d6add1c6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 51 in Silicon Valley]]></title>
            <link>https://medium.com/@neseozler/week-51-in-silicon-valley-4adb07ea5226?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/4adb07ea5226</guid>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Wed, 24 Dec 2025 07:13:41 GMT</pubDate>
            <atom:updated>2025-12-24T07:13:41.854Z</atom:updated>
            <content:encoded><![CDATA[<p>Let’s take a look back at my tweets (<a href="https://x.com/nesozl">@nesozl</a>) from last week to understand what dominated the conversation in Silicon Valley.</p><h3>Founders Betting Big on Moonshot AI</h3><blockquote>🧪🤖 #Neolabs are having a moment.</blockquote><blockquote>New AI labs chasing ideas Big AI may have missed:</blockquote><blockquote>• Hark: Brett Adcock (Figure AI), $100M personal bet<br>• Richard Socher: <a href="http://you.com/">http://You.com</a> founder, raising $1B</blockquote><blockquote>Breakthroughs or moonshots?</blockquote><blockquote>#AI #DeepTech</blockquote><p>A new tier of AI labs is emerging to tackle problems <strong>“Big Tech”</strong> might be ignoring. High-net-worth founders are using personal capital to bypass traditional VC constraints and pursue high-risk, high-reward AI breakthroughs.</p><h3>Specialized vs. General Intelligence: The Rise of the Bespoke “KPI-Driven” Model</h3><blockquote>🧠🚀 #ThinkingMachinesLab plans to launch its own in-house AI models next year, betting big on customized models tied to company KPIs</blockquote><blockquote>Compelling idea, but still unclear whether most companies need bespoke AI, or simply strong general models + better workflows</blockquote><blockquote>#Neolabs #GenAI</blockquote><p>Is the <strong>“one model fits all”</strong> era is ending? Is the next phase is specialized AI built for specific business outcomes? This is what Thinking Machines Lab is betting on.</p><h3>From Chatbots to “Agentic” Automation</h3><blockquote>2025 Service Agent upgrades: ✅ Refined topic structures ✅ Stronger guardrails &amp; retrieval ✅ Rapid feedback loops</blockquote><blockquote>Goal: Move from broad to targeted same! It’s a side step, not a leap! LLM randomness hurts trust and stalls the path to full automation</blockquote><blockquote>#Salesforce #AI #LLM #CX</blockquote><p>The goal is moving from <strong>“AI as a co-worker”</strong> to <strong>“AI as a worker”</strong> that can be trusted to handle complex customer service tasks without human supervision. But LLM randomness keeps creating the trust issues (The <strong>“Stochastic Trust” </strong>Problem) and remains the primary barrier. Companies are now moving toward <strong>“Deterministic Orchestration/Guardrails</strong>” using strict code-based guardrails to “trap” the AI’s randomness before it reaches the customer. <strong>AgentOps</strong>, a new trend focused on monitoring “reasoning coherence” and “decision accuracy” to prove that an agent is safe for production became the hit in 2025. Let’s see how the agents will evolve in 2026.</p><h3>AI Infrastructure vs. Consumer Growth IPOs</h3><blockquote>The 2025 IPO Report Card is volatile!</blockquote><blockquote>Non-AI:<br>StubHub (Sep ’25, $23.50): Down 40%<br>Klarna (Sep ’25, $40): Down 22%<br>Navan (Oct ’25, $25): Down 42%<br>Chime, down 3%</blockquote><blockquote>AI Stocks:<br>CoreWeave (Mar ’25, $40): Closed at $78.59 (Peak $183).<br>Fermi (Sep ’25, $21): Down 57%</blockquote><blockquote>#2025IPOs #AI</blockquote><p>Investors are prioritizing <strong>“AI-enablement”</strong> (hardware and compute) over traditional high-growth software and fintech.</p><h3>Markets Become the New Search Engines for Probabilities</h3><blockquote>How many prediction markets? #kalshi, #Polymarket, #Manifold</blockquote><blockquote>Odds: <br>🎥 AI movie by 2028: 28% <br>🤖 Chores robot by 2029: 48% <br>🚀 #OpenAI IPO (end ‘26): 34% <br>🦄 #Anthropic IPO (mid ‘26): 43% (Polymarket) vs 65% (Kalshi)</blockquote><blockquote>Markets are the new info layer. #PredictionMarkets #kalshi</blockquote><p>Prediction markets like <strong>Polymarket</strong> and <strong>Kalshi</strong> are being used as real-time APIs for <strong>“ground truth”</strong>. As AI agents become the primary traders, these markets provide the structured data that helps AI models separate hype from reality.</p><h3>The Orbital Data Boom:SpaceX and the Launch of Space-Based Data Centers</h3><blockquote>We started to see these graphs about decreasing payload costs in 2020 due to the #connectivity trend 🛰️ now the conversation has shifted, and it is for data centers in space!<br>The launch cost per kilogram down from over $20K to near $1K (Falcon Heavy). <br>#SpaceDataCenters #SpaceX</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ALv6UKeWFQdi8h2l" /></figure><p>Falling payload costs have made it economically viable to move <strong>“the cloud”</strong> into the stars.</p><p>#SpaceDataCenters #SpaceX #PredictionMarkets #kalshi #2025IPOs #AI #Neolabs #GenAI</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4adb07ea5226" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 49 in Silicon Valley]]></title>
            <link>https://medium.com/@neseozler/week-49-in-silicon-valley-d8e2b8544631?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/d8e2b8544631</guid>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Tue, 09 Dec 2025 10:38:01 GMT</pubDate>
            <atom:updated>2025-12-09T10:38:01.241Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/765/0*ep5cw67r-kQ0kYir" /></figure><p>Let’s take a look back at my tweets (<a href="https://x.com/nesozl">@nesozl</a>) from last week to understand what dominated the conversation in Silicon Valley.</p><h3>OpenAI’s New Playbook: Full-Stack M&amp;A + AI Rollups = Total Market Control</h3><blockquote>OpenAI’s $500B valuation powers an M&amp;A spree!</blockquote><blockquote>Recent buys using stock:</blockquote><blockquote>— May’25: <strong>io</strong> (Jony Ive AI HW): &gt; $6B</blockquote><blockquote>— Sept’25: <strong>Statsig</strong> (Product Dev): $1.1B</blockquote><blockquote>— Oct’25: <strong>Software Apps Inc.</strong></blockquote><blockquote>— Dec’25: <strong>Neptune</strong> (AI Model Monitoring): &lt; $400M</blockquote><blockquote>They’re consolidating AI talent &amp; tooling fast</blockquote><p>This isn’t random dealmaking. It’s <strong>ecosystem consolidation</strong>: infrastructure + tooling + talent, all pulled inside OpenAI’s orbit. The M&amp;A activity is not for revenue; it’s a strategic move to secure <strong>top-tier AI talent </strong>and <strong>critical tooling</strong> (like model monitoring and product development infrastructure). The goal is clear: <strong>eliminate dependencies and accelerate the path to AGI</strong> by owning the <strong>full stack of AI development and deployment</strong>.</p><blockquote>Rollups are evolving! Thrive, like General Catalyst, is also buying service firms and infusing them with AI from the ground up. OpenAI’s stake in Thrive highlights the growing momentum behind AI-native rollups in service industries <a href="https://x.com/hashtag/rollups?src=hashtag_click">#rollups</a> <a href="https://x.com/hashtag/ThriveCapital?src=hashtag_click">#ThriveCapital</a> <a href="https://x.com/hashtag/ThriveHoldings?src=hashtag_click">#ThriveHoldings</a></blockquote><p>While the <strong>M&amp;A spree</strong> secures the internal technical stack, infrastructure, tooling, and top talent, the <strong>stake in Thrive Holdings</strong> signifies a pivotal, external shift in OpenAI’s <strong>GTM</strong> strategy. It moves OpenAI beyond simply selling AI via API to actively embedding its models in the field, a natural extension of its <strong>forward-deployed engineering approach</strong>. This “AI Rollup” model, where cash-flow-positive service businesses (like accounting and IT) are acquired and rebuilt around their frontier models, allows OpenAI to <strong>vertically integrate its technology directly into industry workflows</strong>. The reasoning: the Thrive partnership serves as a <strong>proof-of-concept</strong> for enterprise transformation, demonstrating massive efficiency gains and <strong>establishing a template for reliable, ground-up corporate distribution that bypasses slow sales cycles</strong>, effectively completing the ecosystem control from core talent to end-user deployment.</p><p><strong><em>OpenAI Dominance in the Market = M&amp;A (buying infrastructure/talent) + Thrive Investment (buying distribution/application, in other words new, high-leverage GTM playbook)</em></strong></p><blockquote>Seriously considering canceling Chatgpt plus subscription and starting to use Gemini!!!</blockquote><p>And yet, user sentiment is shifting in real time: <strong>the consumer AI market is still up for grabs</strong>. Loyalty is not guaranteed, even as the company grows its industrial footprint.</p><h3>AI’s Reality Check: Scaling Breaks Down, Alignment Costs Spike, and Benchmarks Lose Credibility</h3><p>This year’s NeurIPS and research conversations reveal a striking shift: the community is increasingly skeptical that <strong>pure scaling</strong> will produce AGI.</p><blockquote>Sutskever says no one knows how to build the kind of machine intelligence that leads to <a href="https://x.com/hashtag/AGI?src=hashtag_click">#AGI</a>. Hence the shift to an “age of research,” rather than “age of scaling”, not just scaling compute. Yet researchers want new methods while business insists AI is scale-ready <a href="https://x.com/hashtag/AgeOfResearch?src=hashtag_click">#AgeOfResearch</a></blockquote><p><strong>The “Age of Research” vs “Age of Scaling”?</strong></p><p>Former OpenAI Chief Scientist <strong>Ilya Sutskever</strong>’s blunt admission that <strong>“no one knows how to build AGI”</strong> marks a turning point for the field. The industry is recognizing that brute-force scaling is hitting diminishing returns, pushing research toward new architectures, cognitive modeling, and fundamentally different training science. Yet this shift clashes with companies still insisting that today’s LLMs are scale-ready and enterprise-proven. The result is a growing disconnect between <strong>research reality and the massive infrastructure bets by tech giants </strong>driving the business side of AI.</p><blockquote>Hearing about the “split-brain problem” a lot in SV. It shows how nuanced model training is, forcing devs to pay experts (math, law, coding) Bs for the right mix. They need to ensure models don’t make simple errors when experts use them. The <a href="https://x.com/hashtag/alignment?src=hashtag_click">#alignment</a> struggle is real! <a href="https://x.com/hashtag/LLMs?src=hashtag_click">#LLMs</a></blockquote><p>The <strong>“split-brain problem” </strong>highlights how AI models can excel in one domain yet fail at simple tasks in another, exposing the difficulty of achieving true <strong>multi-domain coherence</strong>. To close these gaps, developers increasingly rely on <strong>expensive domain experts</strong>, lawyers, mathematicians, engineers to refine training data and provide continuous human feedback, turning alignment into an economic requirement rather than just a safety goal. This has fueled a surge in demand for <strong>expert-in-the-loop training</strong>, creating a new class of winners such as <strong>Scale AI</strong>, <strong>Surge AI</strong>, <strong>Gantry</strong>, and <strong>Humanloop</strong>, which supply the specialized expertise modern models require.</p><blockquote>RL can help models ace benchmarks, but it doesn’t teach them to truly generalize. It’s a reminder that “passing the test” ≠ “understanding the world.” <a href="https://x.com/hashtag/RL?src=hashtag_click">#RL</a> <a href="https://x.com/hashtag/ReinforcementLearningShortcomings?src=hashtag_click">#ReinforcementLearningShortcomings</a> <a href="https://x.com/hashtag/GeneralAI?src=hashtag_click">#GeneralAI</a></blockquote><p>RL has a core technical flaw: it optimizes models for <strong>benchmarks, not for intelligence</strong>. Reinforcement learning can reliably boost scores on structured tests, but it fails to teach <strong>true generalization or the common-sense reasoning</strong> needed in real-world environments. This is why a model can score 90% on a coding exam yet still introduce subtle, <strong>catastrophic bugs in production</strong>, a gap now fueling backlash against evaluation results for models like Gemini and the latest LLaMA. And while this plays out, we continue to see <strong>high-grossing tweets and viral claims</strong> about every new model release, pushing people to form their own judgments and run their own tests because benchmark scores alone no longer inspire trust.</p><h3>AI Growth Is Up, Stocks Are Down: Inside the Strange New Economics of the AI Boom</h3><blockquote>Snowflake stock 8%!</blockquote><blockquote>CEO says growth is driven by AI apps, hitting $100M+ ARR from new AI products. Market seems spooked by the $200M Anthropic spend, ignoring the steady gross profit margin projection. Confusing reaction! <a href="https://x.com/hashtag/SNOW?src=hashtag_click">#SNOW</a></blockquote><p>Investors appear spooked by Snowflake’s <strong>$200M commitment to Anthropic</strong>, which they interpret as risky, high-burn AI capex, even if revenues suggest the opposite. This reflects a new phenomenon: <strong>AI optimism + capex fatigue.</strong></p><blockquote>Oracle’s data center debt is making lenders cautious. Now Morgan Stanley is using SRTs (significant risk transfer) to hedge the risk of the AI data center funding boom. Banks are getting cautious. <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/Oracle?src=hashtag_click">#Oracle</a> <a href="https://x.com/hashtag/MorganStanley?src=hashtag_click">#MorganStanley</a> <a href="https://x.com/hashtag/SRT?src=hashtag_click">#SRT</a></blockquote><p>Banks are beginning to realize that funding the AI boom carries its own form of systemic risk:</p><ul><li>huge upfront capital</li><li>uncertain long-term returns</li><li>concentration in a few hyperscalers</li></ul><h3>AI Isn’t Magic: It’s Just Digital Transformation with Higher Stakes</h3><blockquote>AI Transformation Basics: 1. Understand your company’s processes, data structures, and workflows 2. Identify the key use cases 3. Study the model landscape and market trends 4. Adapt and train the teams <a href="https://x.com/hashtag/AITransformation?src=hashtag_click">#AITransformation</a> <a href="https://x.com/hashtag/LLMs?src=hashtag_click">#LLMs</a></blockquote><p>AI transformation isn’t fundamentally different from digital transformation: Successful AI transformation is not about chasing the newest model; it’s about <strong>organizational readiness</strong>. Companies must first understand their own processes, data structures, and workflows before identifying the few high-impact use cases that truly warrant AI. <strong>With vendors and models changing monthly</strong>, teams also need the fluency to <strong>evaluate tools and avoid lock-in</strong>. Without this internal readiness and cultural adaptation, even the best LLM subscription will lead to failed pilots and stalled initiatives. The winners in AI will be those who prioritize <strong>strategy over shiny new tech</strong>.</p><h3>#openai #AITransformation #Strategy #Neurips #AGI #ThriveHoldings</h3><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d8e2b8544631" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 47 in Silicon Valley]]></title>
            <link>https://medium.com/@neseozler/week-47-in-silicon-valley-4073dc1dca7f?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/4073dc1dca7f</guid>
            <category><![CDATA[stablecoin-development]]></category>
            <category><![CDATA[gemini]]></category>
            <category><![CDATA[crew-ai]]></category>
            <category><![CDATA[google-gemini]]></category>
            <category><![CDATA[open-source-ai]]></category>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Tue, 25 Nov 2025 02:58:50 GMT</pubDate>
            <atom:updated>2025-11-25T02:59:48.347Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/765/0*S_Y6E90KX-WZLBMl" /></figure><p>Let’s take a look back at my tweets (<a href="https://x.com/nesozl">@nesozl</a>) from last week to understand what dominated the conversation in Silicon Valley.</p><h3>What Happened Around the Gemini 3 Launch: A Story Told Through My Tweets</h3><p>From there, I broke down Google’s position in the AI landscape. One tweet summed it up:</p><blockquote>Google has:</blockquote><blockquote>— Gemini, top-notch LLM</blockquote><blockquote>— Google Cloud, cloud firm</blockquote><blockquote>— TPU, AI chip of its own design</blockquote><blockquote>— Chrome, dominant web browser and search engine</blockquote><blockquote>— devices, Pixel smartphone and (maybe soon) smart glasses but also have 190k employees marred by slow decision-making &amp; fiefdoms <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a></blockquote><p>Another tweet highlighted a metric everyone struggled to wrap their heads around:</p><blockquote><em>Hard to benchmark, but Gemini app has 650M #</em><strong><em>monthly</em></strong><em> active users vs. ChatGPT’s 800M #</em><strong><em>weekly</em></strong><em> users!”</em></blockquote><p>A reminder of the massive scale AI products now operate at and how tricky comparisons have become.</p><p>Then came the competitive landscape:</p><blockquote>Tier 1 LLM pricing vibes: • Anthropic Claude 4.5 Sonnet • Google Gemini 3 • OpenAI GPT-5.1 Google feels perfectly positioned in the middle, an underdog-to-contender story similar to early GCP. Big waves ahead in price/performance. <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/LLM?src=hashtag_click">#LLM</a> <a href="https://x.com/hashtag/GenAI?src=hashtag_click">#GenAI</a> <a href="https://x.com/hashtag/Anthropic?src=hashtag_click">#Anthropic</a> <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> <a href="https://x.com/hashtag/Gemini?src=hashtag_click">#Gemini</a></blockquote><p>But I also asked <strong>why Google might actually be winning</strong> at the deepest technical layer. As I tweeted:</p><blockquote>Why is Google winning? TPUs are architecturally well-suited for large-scale pre-training &amp; improving pre-training is key to getting a model to better generalize! Pre-training = general world model Post-training = domain-specific refinement Is this an edge for <a href="https://x.com/hashtag/Google?src=hashtag_click">#Google</a> for future?</blockquote><p>Still, technology isn’t Google’s main issue. I reminded everyone:</p><blockquote>Google’s challenge isn’t building great tech: it’s turning it into products people actually love (and remember). Just look at Google Home → Nest Now the question:can <a href="https://x.com/hashtag/Google?src=hashtag_click">#Google</a> turn impressive AI models into products? And seriously, what’s going on with LLM trip planning use case?</blockquote><p>Because honestly, even keeping track of the top models is hard right now. As I put it:</p><blockquote>It’s hard to follow the leading models. Right now: GPT-5 and Claude Opus 4 and Gemini 3?</blockquote><p>And Google didn’t slow down. I flagged yet another launch. Then I reminded everyone of the flurry of Google dev-tools this year.</p><blockquote>Google again launched an AI product for developing software → Gemini Antigravity! What were the other AI Coding tools from Google this year:</blockquote><blockquote>— Jules (May 2025)</blockquote><blockquote>— Stitch (May 2025)</blockquote><blockquote>— Gemini CLI (June 2025)</blockquote><blockquote>— AlphaEvolve (May 2025)</blockquote><h3>And here’s the bigger question I added in the commentary:</h3><p>Why does Google need seven different coding products in a single year?</p><p>This is the elephant in the room.</p><p>Google’s frantic shipping cadence isn’t just momentum; it’s a symptom of internal fragmentation. Each team has its own mandate, its own leadership chain, its own roadmap, and its own incentives. Instead of consolidating around one flagship AI coding assistant, Google ends up with multiple overlapping tools:</p><ul><li>different interfaces,</li><li>different philosophies,</li><li>different intended users,</li><li>often built by completely different orgs.</li></ul><p>It’s the classic “fiefdom problem” I mentioned earlier.</p><p>Inside Google, this isn’t viewed as duplication; it’s viewed as <em>parallel bets.</em> But outside Google, it looks <strong>chaotic</strong>: seven separate developer AI tools in one year speaks to ambition, but also to a lack of product cohesion. And the risk is obvious instead of becoming the go-to AI coding platform, Google dilutes its own momentum.</p><p>This is the flip side of their strength: they can build almost anything, but struggle to build <strong>one thing</strong> everyone uses!!!</p><h3>Market Shifts, Chip Wars, and Cloud Alliances: What Else Moved in AI This Week</h3><p>A few of my tweets last week highlight how quickly the broader AI landscape is shifting.</p><blockquote>Meanwhile, In late July Meta share were up 32% for the year, while Google stock was up 1.5%. Those positions have now reversed: Meta shares are up just 3% on the year, while Google stock is up 50% <a href="https://x.com/hashtag/meta?src=hashtag_click">#meta</a> vs <a href="https://x.com/hashtag/google?src=hashtag_click">#google</a> in <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a></blockquote><p>This reversal shows how investor conviction has rotated: Meta’s early-year momentum cooled, while Google boosted by Gemini 3 and renewed confidence in its AI stack; suddenly became one of the strongest AI trades of the year. Everybody around me wants to go big on google investment all of a sudden!</p><blockquote>Nvidia’s projected October-quarter revenues would reach $54 billion (54% higher than 2024) vs AMD’s $9.2 billion revenue reported last month! <a href="https://x.com/hashtag/amd?src=hashtag_click">#amd</a> <a href="https://x.com/hashtag/nvidia?src=hashtag_click">#nvidia</a> <a href="https://x.com/hashtag/chipwars?src=hashtag_click">#chipwars</a></blockquote><p>The compute story is even more extreme. Nvidia is accelerating at a historic pace, widening the gap with AMD as AI infrastructure spending continues to consolidate around a single dominant supplier.</p><blockquote>Now all three major cloud providers are invested in <a href="https://x.com/hashtag/Anthropic?src=hashtag_click">#Anthropic</a> , an <a href="https://x.com/hashtag/OpenAI?src=hashtag_click">#OpenAI</a> rival. AI, cloud, and compute are all intertwined and everyone is placing bets on the same hot startups. Feels like we’re heading back into circular deals again. <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/BigTech?src=hashtag_click">#BigTech</a> <a href="https://x.com/hashtag/CloudComputing?src=hashtag_click">#CloudComputing</a> <a href="https://x.com/hashtag/TechTrends?src=hashtag_click">#TechTrends</a></blockquote><p>With Amazon, Google, and Microsoft all backing Anthropic, the cloud giants are making overlapping bets again, a sign that the AI model layer and the cloud layer are converging, and that no one wants to risk being left out of the next big breakthrough.</p><h3>Open vs. Closed Coding Assistants</h3><blockquote><a href="https://x.com/hashtag/Cursor?src=hashtag_click">#Cursor</a> as a closed-source coding assistant vs <a href="https://x.com/hashtag/Cline?src=hashtag_click">#Cline</a> as an open-source coding assistant <a href="https://x.com/hashtag/Devin?src=hashtag_click">#Devin</a> as a closed-source coding agent vs <a href="https://x.com/hashtag/OpenHands?src=hashtag_click">#OpenHands</a> is open-source coding agent</blockquote><p>The coding-assistant ecosystem is now splitting into two clear camps. On one side are the <strong>closed-source assistants</strong> (Cursor, Devin) that focus on polish, reliability, and tight integration. On the other side are the <strong>open-source agents</strong> (Cline, OpenHands) that trade some refinement for transparency, community contributions, and rapid experimentation.</p><p>This mirrors a pattern we’re seeing across all of AI: closed systems push for performance and productization, while open systems push for flexibility, trust, and ownership. The speed at which both sides are evolving signals that the developer tooling wars are just beginning and we may end up with two parallel ecosystems rather than one dominant winner.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/758/0*bI-warDcKPp7zHZI" /></figure><h3>CrewAI Signal Summit (San Francisco)</h3><p>At the CrewAI Signal Summit in SF, several speakers highlighted how fast the agent ecosystem is evolving and how dramatically it could reshape organizations.</p><p>Replit’s CEO set the tone with a provocative question:</p><blockquote>“Today 20% of the organization has to develop sw for the rest of the company! What of everybody develops sw in the company?” Says Replit CEO. Everybody building sw means massive org structure changes! <a href="https://x.com/hashtag/replit?src=hashtag_click">#replit</a></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ODZYydQ_kOFxtnG5" /></figure><p>If agents enable <strong>everyone</strong> to build software, company structures will have to be redesigned from the ground up.</p><p>Andrew Ng weighed in on the research discourse:</p><blockquote>“MIT study is flawed as success factors are narrowly scoped!” Said Andrew Ng at Crewai Signal!</blockquote><p>A reminder that much of today’s agent research is still early, inconsistent, and hard to generalize even from respected institutions.</p><blockquote>Deterministic systems and agents … saas categories that doesn’t have any incumbents (?) and it’s knowledge management for unstructured data. Eventually, services firms will become your agents’ customers. Thx Box CEO</blockquote><p>The message: agents aren’t replacing SaaS, they’re creating entirely new layers of software where incumbents don’t exist yet.</p><p>And one of the most forward-looking ideas was about self-improving systems:</p><blockquote>Agentic context engineering — ACE: result is self improving agents! Stochastic engines every now and then fails! Memory is important! Self-evolving LLMs is the result!</blockquote><p>This captures a key insight: the next leap in agent performance will come from <strong>better memory + self-correction loops</strong>, not just larger models.</p><h3>The New Payments Race: Stablecoins Go Mainstream</h3><blockquote><a href="https://x.com/hashtag/Mastercard?src=hashtag_click">#Mastercard</a> is eyeing <a href="https://x.com/hashtag/Zerohash?src=hashtag_click">#Zerohash</a> for ~$1.5B as <a href="https://x.com/hashtag/stablecoin?src=hashtag_click">#stablecoin</a> rails heat up. Amazon &amp; Walmart are exploring stablecoins, <a href="https://x.com/hashtag/Stripe?src=hashtag_click">#Stripe</a> is beefing up with its Bridge acquisition, and <a href="https://x.com/hashtag/Visa?src=hashtag_click">#Visa</a> is hunting stablecoin infra too. The crypto payments race is on <a href="https://x.com/hashtag/CryptoPayments?src=hashtag_click">#CryptoPayments</a> <a href="https://x.com/hashtag/stablecoins?src=hashtag_click">#stablecoins</a></blockquote><p>The financial system is quietly entering a new phase: <strong>stablecoins</strong> are moving from crypto circles into the core of global commerce. When companies like <strong>Mastercard, Visa, Amazon, Walmart, and Stripe</strong> all push into the same infrastructure layer, it signals something bigger than experimentation. It’s a shift in <strong>settlement architecture</strong>.</p><p>Stablecoins offer instant settlement, lower fees, programmable money, and cross-border simplicity; features traditional rails can’t match. What we’re seeing now is every major payments player racing to own a piece of this emerging backbone before it goes mainstream.</p><p>This isn’t about crypto hype; it’s about <strong>upgrading the pipes of the financial system</strong>. The next decade of payments may be defined not by banks or card networks alone, but by whoever controls the <strong>stablecoin</strong> rails beneath them.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4073dc1dca7f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 46 in Silicon Valley]]></title>
            <link>https://medium.com/@neseozler/week-46-in-silicon-valley-b4be8bc04b12?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/b4be8bc04b12</guid>
            <category><![CDATA[wordlab]]></category>
            <category><![CDATA[search-engines]]></category>
            <category><![CDATA[solar-energy]]></category>
            <category><![CDATA[enterprise]]></category>
            <category><![CDATA[neolab]]></category>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Tue, 18 Nov 2025 22:41:33 GMT</pubDate>
            <atom:updated>2025-11-18T22:47:59.688Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/765/0*BkruYXqrDWSCZVLS" /></figure><p>Let’s take a look back at my tweets (<a href="https://x.com/nesozl">@nesozl</a>) from last week to understand what dominated the conversation in Silicon Valley.</p><blockquote>Next AI search war: who owns the index + browser. Perplexity offered Brave $1B in stock this summer to grab its independent index &amp; 80M users. Brave said no. Vertical integration is coming. Google has it. Brave is building it. Everyone else is renting. Who wins? <a href="https://x.com/hashtag/Brave?src=hashtag_click">#Brave</a> <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a></blockquote><p>The next AI search war will be defined by who controls both the <strong>search index</strong> and the <strong>browser</strong> layer. Owning the index means not relying on Google or Bing, which gives strategic independence and better AI retrieval quality. Vertical integration matters because Google already controls the full stack, while Brave is actively building its own. Everyone else is renting access from incumbents, and in the long run, renters rarely win!</p><p>— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —</p><p>World Labs launch … World Labs is introducing Marble as a platform/tool that enables you (creators, designers, developers) to build your own 3D worlds easily, turning imagination into interactive spatial environments. It’s positioned not just as a tool, but as a <strong>foundation</strong> for a future where AI understands and generates space something broader than just image or text generation. Marble opens the door to <strong>world-building as a service</strong>: you give prompts or rough layout, it gives you an explorable 3D world, you can share, edit, export. Check it out: <a href="https://marble.worldlabs.ai/">https://marble.worldlabs.ai/</a></p><p>— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —</p><blockquote>Seeing a lot of takes that agents are converging, totally agree. In 2023 we thought the battle would be incumbents’ embedded agents vs new AI startups. Today real war is AI Labs vs AI startups (custom built agents) <a href="https://x.com/hashtag/AI?src=hashtag_click">#AI</a> <a href="https://x.com/hashtag/AIAgents?src=hashtag_click">#AIAgents</a> <a href="https://x.com/hashtag/AIStartups?src=hashtag_click">#AIStartups</a> <a href="https://x.com/hashtag/AILabs?src=hashtag_click">#AILabs</a> <a href="https://x.com/hashtag/AIWrappers?src=hashtag_click">#AIWrappers</a></blockquote><p>Although incumbents currently offer agents across a broader range of verticals, the long-term advantage appears to favor startups and emerging AI labs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*AwCHaamv6CxlGo6f" /></figure><p>— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —</p><blockquote>Summary of 2025: “geriatric startups” → over a decade old, have solid finances but lack the explosive growth story. It’s no longer enough to be a “tech” company. Cos going public must now prove they are either “AI advantaged, AI insulated or… have additional AI risk exposure”</blockquote><p>The phrase <strong>“geriatric startups”</strong> refers to companies that are more than ten years old: mature, financially stable, but no longer growing fast enough to fit the classic “startup” narrative. In 2025, simply being a tech company isn’t impressive anymore; investors expect a compelling <strong>AI-related story</strong>. For a company to go public, it now has to show one of three things: that it is <strong>AI advantaged</strong> (AI will supercharge its growth), <strong>AI insulated</strong> (AI won’t disrupt its core business), or that it has <strong>AI risk exposure</strong> that investors must understand. In other words, AI has become the defining lens through which public markets evaluate future potential and vulnerability.</p><p>— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —</p><blockquote>Solar power and battery storage are on track to account for 50% of U.S. electric power capacity additions this year <a href="https://t.co/TEpTOHSD8q">https://eia.gov/todayinenergy/detail.php?id=65964</a></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*WKoTD_yrZfg-iw2T" /></figure><p>This means that in the U.S. energy sector, <strong>half of all new power-generating capacity added this year will come from solar panels and large-scale battery storage systems</strong>. In other words, when utilities and developers build new power plants in 2025, <strong>1 out of every 2 megawatts</strong> of new capacity will be either solar or batteries. It signals a major shift in the electricity mix: clean energy is now the dominant source of new infrastructure. Batteries, in particular, are becoming essential because they store excess solar power and help stabilize the grid. Overall, it’s evidence that the U.S. is rapidly moving toward a more renewable, storage-supported energy system.</p><p>— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —</p><p>Neolabs, a new wave of research-driven AI startups, are securing seed-stage valuations <strong>above $1B</strong> despite generating <strong>less than $10M in revenue</strong>. Many are founded by former frontier-model researchers with personal earnings in the <strong>$10–100M+ range</strong>:</p><p>— <a href="https://humansand.ai/">Humans&amp;</a>($4B | 2025): <a href="https://www.linkedin.com/in/ericzelikman/">Eric Zelikman</a>, Stanford PhD and prev. leading AI researcher at xAI</p><p>— Isara ($1B | 2025): Eddie Zhang, prev. safety researcher at OpenAI</p><p>— Richard Socher’s Lab ($1B raised, likely at ~$5B | 2025), prev. CEO of you.com, chief scientist at Salesforce</p><p>— <a href="https://periodic.com/">Periodic Labs</a> ($1B | 2025): <a href="https://www.linkedin.com/in/william-liam-fedus-26547811/">William Fedus</a>, prev. VP of Post-Training at OpenAI and <a href="https://www.linkedin.com/in/ekin-dogus-cubuk-9148b8114/">Ekin Dogus Cubuk</a>, prev. lead materials &amp; chemistry team at DeepMind</p><p>— <a href="https://thinkingmachines.ai/">Thinking Machines Labs </a>($8B | 2025): <a href="https://www.linkedin.com/in/mishalaskin/">Misha Laskin</a> and <a href="https://www.linkedin.com/in/ioannis-alexandros-antonoglou-45393253/">Ioannis Antonoglou</a>, both prev. research scientist at DeepMind</p><p>— <a href="https://www.inceptionlabs.ai/">Inception Labs</a> (2024): <a href="https://www.linkedin.com/in/ermon/">Stefano Ermon</a>, <a href="https://www.linkedin.com/in/aditya-grover/">Aditya Grover</a> and <a href="https://www.linkedin.com/in/volodymyr-kuleshov-6aa83294/">Volodymyr Kuleshov</a>; professors at Stanford, UCLA and Cornell</p><p>— <a href="https://ssi.inc/">Safe Superintelligence</a> ($32B | 2024): Ilya Sutskever, prev. Chief Scientist at OpenAI</p><p>— <a href="https://reflection.ai/">Reflection AI</a> ($8B | 2024): <a href="https://www.linkedin.com/in/mishalaskin/">Misha Laskin</a> and <a href="https://www.linkedin.com/in/ioannis-alexandros-antonoglou-45393253/">Ioannis Antonoglou</a>, both prev. research scientist at DeepMind</p><p>— <a href="https://poolside.ai/">Poolside</a> ($3B | 2023): <a href="https://www.linkedin.com/in/jcw148/">Jason Warner</a>, prev. CTO at Github</p><p>— <a href="https://www.generalintuition.com/">General Intuition</a> (2025): <a href="https://www.linkedin.com/in/pimdw/">Pim de Witte</a>, prev. co-founder of Medal (OpenAI tried to acquire Medal for $500m)</p><p><strong>Verdict on Neolabs</strong>: A new wave of “neolabs”; small, elite research-driven AI startups; is raising billions as investors bet they can out-innovate giants like OpenAI and Anthropic. Labs like Isara, Humans&amp;, and Periodic aim to push into overlooked areas such as massive multi-agent systems, emotional-intelligence models, long-term reinforcement learning, and automated scientific discovery.</p><p>The appeal is clear: scarce top-tier talent and huge potential upside. But the risks are steep; the incumbents have a massive lead, enormous compute budgets, and can copy any successful idea.</p><p><strong>Bottom line:</strong> Neolabs are a bold, high-risk bet that small teams with fresh ideas can break through in a market dominated by trillion-dollar players.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b4be8bc04b12" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[TechCrunch Disrupt 2025: The Post-Training Revolution and the New Startup Playbook]]></title>
            <link>https://medium.com/@neseozler/techcrunch-disrupt-2025-the-post-training-revolution-and-the-new-startup-playbook-9f61ff09a4ee?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/9f61ff09a4ee</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[techcrunchdisrupt]]></category>
            <category><![CDATA[robotics]]></category>
            <category><![CDATA[vc]]></category>
            <category><![CDATA[technology-trends]]></category>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Wed, 29 Oct 2025 19:28:21 GMT</pubDate>
            <atom:updated>2025-10-29T21:19:30.627Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/765/0*DxKW-C3SynijV64h" /></figure><p><a href="https://techcrunch.com/events/tc-disrupt-2025/">TechCrunch Disrupt 2025</a> captured the feeling of an industry mid-shift, where founders, investors, and AI infrastructure players are collectively re-writing the rules of innovation. Below are my distilled takeaways from multiple sessions: from the “Post-Training Revolution” and “Intelligence in Motion,” to the “VCs Placing Bets in 2026” panel.</p><h3>🧬 The AI Stack Is Changing: Post-Training Takes Center Stage</h3><p>One of the most technically fascinating discussions revolved around how <strong>post-training</strong> is overtaking pre-training as the dominant compute workload.</p><ul><li>Traditional training → inference boundaries are collapsing</li><li><strong>Post-training = inference</strong> in the new regime: models continuously learn, sample, and refine as they run</li><li>Tooling (monitoring and evals) is emerging as the next gold rush layer</li></ul><p>Peter Hoeschele from OpenAI framed it perfectly:</p><blockquote>“Let’s stop talking about training versus inference. Models are now constantly running, sampling, training, and getting better all the time.”</blockquote><p>The compute charts shared showed that while GPT-5 used less training FLOPs than its predecessors, <strong>total post-training workloads have exploded</strong>, driven by reinforcement learning, fine-tuning, and real-world usage loops.</p><p>Traditionally, training was seen as a <strong>one-time process</strong>: you’d train the AI model, finish, and then “deploy” it for use.</p><p>Now, training is becoming an <strong>ongoing environment, </strong>meaning the system keeps learning continuously, even after it’s deployed. It adapts to new data, new conditions, or feedback from users.</p><p>We used to train AI like giving it a one-time lesson. Now, it’s more like the AI is going to school every day; continuously learning, improving, and evolving.</p><p>This marks a shift in how we think about infrastructure: <strong>training is no longer an event; it’s an environment.</strong></p><h3>🧠 The Rise of Agents</h3><p>Another theme was the rapid specialization of AI agents, for <strong>consumers, code, sales, marketing, and law, </strong>each a building block in a future where “AI employees” handle entire functions.</p><p>OpenAI’s move “up the stack” into the <strong>app layer</strong> reflects this: it’s no longer about APIs, but about orchestration, how agents talk to each other and to humans.</p><h3>💰 Raising Series A in 2025</h3><p>In the past, pre-AI boom, roughly 2010s–early 2020s, for most SaaS startups, the <strong>classic PMF threshold</strong> was around <strong>$1–2M ARR. </strong>Benchmarks may look simple: strong retention, believable TAM, but the context has changed dramatically. The challenge after that was scaling efficiently, not proving PMF. Right now:</p><ul><li><strong>Distribution has become the moat</strong>, not technology</li><li>Going from $0→$1.5 M ARR is doable; scaling from $1.5→$5 M is the new filter for true product-market fit</li><li>The bar has shifted upward. Because AI products can grow faster and attract <em>try-curious</em> customers, <strong>early revenue is less reliable</strong> as proof of PMF</li><li>The best GTM beats the best product almost every time</li><li>AI startups can reach early revenue faster than ever. Some hit <strong>$10M–$50M ARR in a few months</strong>, because companies are eager to experiment with new AI tools. But that early growth can be misleading.</li><li>Many customers: Buy <em>pilot licenses</em> or small trials just to “see what it can do”; Then <em>drop off or shrink usage</em> once the novelty fades, or when real workflows or ROI don’t justify it.</li></ul><p>That’s why some AI companies see revenue spikes followed by sudden declines. The question becomes: <strong>can they retain and expand customers once the excitement wears off?</strong></p><p>Early revenue in AI isn’t always a signal of lasting success anymore. The true measure is <strong>defensibility; </strong>things like:</p><ul><li>proprietary data or models,</li><li>network effects,</li><li>deep workflow integration,</li><li>or strong customer dependence.</li></ul><p>Don’t forget about the incumbents! Incumbents are no longer asleep, they’re fast, integrating AI into every workflow and squeezing time windows for early entrants.</p><h3>📈 VCs Placing Bets in 2026</h3><p>A highlight panel with <strong>Index Ventures, Greylock </strong>and <strong>Felicis Ventures</strong> revealed how investors are redrawing their maps for the AI era.</p><h4>1. Funding Cycles and Integration Loops</h4><p>Index framed it as a full-circle moment:</p><ul><li><strong>First</strong>, foundational model labs got funded.</li><li><strong>Then</strong>, app-layer startups built on top.</li><li><strong>Now</strong>, we’re seeing <em>vertical integration back down the stack, </em>“OpenAI, Anthropic and rest of the world” blending infra and product.</li></ul><h4>2. Founders, PMF, and Honesty</h4><p>Both firms stressed that <strong>false positives in enterprise PMF</strong> are rampant, pilots without real ROI.<br> “Try honestly,” one partner said, “and you’ll see if it’s truly solving a workflow problem or just an exec’s curiosity.”</p><h4>3. Talent and Velocity</h4><p>The <strong>talent war is brutal, </strong>every OpenAI Dev Day births ideas and kills 1,000 startups. The bar for speed, clarity, and execution keeps climbing.</p><h4>4. Data Flywheels and Differentiation</h4><p>One of the VCs hammered on <strong>unique data flywheels, </strong>how to identify and own data that models can’t easily replicate. Enterprise AI success will hinge on unlocking proprietary workflows and data, not just wrapping LLMs.</p><h4>5. Verticalization and Workflow AI</h4><p>Both investors see deep verticals (insurance, healthcare, manufacturing) as the next frontier.<br>These sectors demand <strong>process-aware AI, </strong>systems that skip “2–3 steps in between” and insert intelligence directly into workflows.</p><h4>6. Business Models and Pricing Shifts</h4><ul><li><strong>Seat-based pricing is dying; outcome-based pricing is rising</strong></li><li>The <strong>unit of compute (GPU / token)</strong> becomes the new cost driver.</li><li>SaaS in 2026 looks more like <em>software-wrapped hardware</em> or <em>robotics SaaS</em>, with private clouds and bundled workstations returning in fashion.</li></ul><h4>7. Market Gaps and “Hot Takes”</h4><ul><li><strong>Consumer AI</strong> still isn’t working.</li><li><strong>Enterprise enablement</strong> and <strong>AI-powered marketplaces</strong> are fertile ground.</li><li>“AI as a teammate” startups are proliferating, especially in China, where 200+ teams are subdividing roles like PM and analyst into agent systems.</li></ul><h3>⚙️ The Post-Training Flywheel for Founders</h3><p>VCs agreed on one unifying thesis:<br><strong>the defensible moat isn’t model quality; it’s data, workflow depth, and iteration speed.</strong></p><p>The “white space” lies in workflow defense; startups that integrate tightly with how work gets done will outlast those chasing novelty.</p><h3>🤖 Intelligence in Motion: The Future of Physical AI</h3><p>The “Intelligence in Motion” session explored <strong>embodied AI, </strong>where intelligence meets physical action.<br>Key themes included:</p><ul><li>Robots as data generators: the <strong>best but hardest-to-scale data source</strong>.</li><li><strong>Three horizons</strong>:</li></ul><ol><li>Manufacturing &amp; logistics (labor shortages)</li><li>Healthcare &amp; retail (safety, process efficiency)</li><li>Home assistance (affordable, reliable personal robotics)</li></ol><p>Partnerships like <strong>Google Gemini × Waabi</strong> suggest a coming “ChatGPT moment” for robotics. We might be near a turning point where robotics, powered by advanced AI like Gemini, becomes as suddenly capable and visible as ChatGPT made language AI.</p><h3>🧭 Closing Thoughts</h3><p>Disrupt 2025 felt like standing on shifting ground; not chaotic, but alive.<br> The line between <em>training</em> and <em>doing</em> is blurring; AI agents are evolving into teammates; and investors are chasing vertical depth over horizontal hype.</p><p>As one VC put it:</p><blockquote><em>“Every six months, the goalposts move. But that’s the fun part: in AI, velocity </em>is<em> strategy.”</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9f61ff09a4ee" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The End of Apps: Where Software Fades and Intent Becomes the Interface]]></title>
            <link>https://medium.com/@neseozler/the-end-of-apps-where-software-fades-and-intent-becomes-the-interface-e8331d29cd47?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/e8331d29cd47</guid>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Fri, 01 Aug 2025 18:00:04 GMT</pubDate>
            <atom:updated>2025-08-01T18:00:04.865Z</atom:updated>
            <content:encoded><![CDATA[<p>Microsoft’s leadership has signaled a paradigm shift in how we interact with software. Satya Nadella, Microsoft’s CEO, predicts that <strong>many traditional and custom applications will fade away,</strong> replaced by generative AI interfaces. In his vision, <strong>“traditional tools like Excel might become less essential as AI agents take over tasks such as data analysis and business logic execution”. </strong>In other words, instead of navigating menus and forms in multiple apps, we’ll simply tell an intelligent agent what we need, and it will execute the workflow behind the scenes.</p><p>Let’s dissect this prediction.</p><p><strong>So, what happens to Microsoft’s flagship tools like Excel, PowerPoint, and Word?</strong><br>Excel might evolve from a daily-use tool to a behind-the-scenes engine, used by the AI but rarely seen by the user. PowerPoint could shift from being a slide design tool to a generative storytelling assistant, crafting narratives on command. Word’s familiar editing interface might fade, as composition moves toward voice and chat-based interaction.</p><p><em>Microsoft likely won’t eliminate these tools, but may increasingly abstract them. </em>The software may not disappear, but for many users, it could become <strong>invisible</strong>: orchestrated behind the scenes by AI agents that focus on delivering <strong>outcomes</strong>, not interfaces.</p><p>This marks a shift from:<br> <strong>“How do I use this tool?”</strong><br> to<br> <strong>“What outcome do I want, and how do I express it?”</strong></p><blockquote>Just like markets used to run smoothly because of the “invisible hand”, where people’s individual choices added up to a bigger result. AI is becoming the invisible hand of software. It quietly pulls together the right tools, data, and steps to help us get what we want, without us needing to manage the process ourselves. The interface fades, but the outcome is still delivered.</blockquote><h4>Tools Become Infrastructure</h4><p>Just as the command line gave way to the GUI, today’s tools are becoming <strong>invisible, </strong>not obsolete, but <strong>foundational</strong>. <strong>Tools become infrastructure.</strong> The new interface is shaped by <strong>intent</strong>. This means, tools haven’t disappeared. They’ve become <strong>invisible performers, </strong>summoned and coordinated by your <strong>intent</strong>.</p><p>Using traditional software is like plugging your laptop into an Ethernet cable. You need to find the port, carry the cable, stay physically tethered, and manage the connection manually.</p><p>An interface shaped by intent is like using <strong>Wi-Fi</strong>. You just open your device and say: <strong>“Connect me to the internet.”</strong> You don’t think about IP addresses, routing tables, or DHCP handshakes. The system automatically interprets your intent, <strong>“get me online”, </strong>and handles the complexity behind the scenes. The infrastructure (routers, protocols, cables) is still there. But it’s <strong>transparent</strong> to the user. You’ve moved from managing connections to <strong>assuming connectivity, </strong>because the system understands what you want and handles how to get it done.</p><p>This shift mirrors the <strong>cloud transformation of the 2000s</strong>, when users stopped worrying about servers and started expecting services to “just work.”<strong> </strong>Now, AI is doing the same for interaction: removing friction, hiding complexity, and making <strong>intent the new default interface, </strong>because it’s the <em>starting point</em> for all interaction, and the system is smart enough to do the rest.</p><blockquote>In the age of AI, we no longer pull the levers of software, we describe <strong>outcomes</strong>, and an <strong>invisible hand</strong> sets the system in motion. <strong>Tools become infrastructure</strong>. <strong>Interfaces fade</strong>. I<strong>ntent leads</strong>.</blockquote><p>Early signs of this trend, AI transformation of specific categories of custom or labor-intensive applications, are already here:</p><ul><li><strong>Presentation and Document Creation:</strong> Tired of crafting slide decks or reports manually? New AI tools can do it for you. For example, <strong>Gamma</strong> is a startup offering an AI-powered alternative to PowerPoint. <em>Rather than designing slides slide-by-slide, users can paste in raw content or even just a one-line prompt, and Gamma automatically generates a polished, engaging presentation</em>. <a href="https://www.fastcompany.com/91328009/gamma-ai-presentations-powerpoint-alternative">Fast Company</a> dubs this approach “a way to pass all the design work off to AI,” freeing users from <strong>‘death by PowerPoint’</strong>. In practice, you focus on your ideas, and the generative tool builds the visual narrative. This content-first, design-later workflow means apps like PowerPoint could become less central,<strong> </strong>you might not open PowerPoint at all, instead using an AI service that outputs a ready-to-show presentation.</li><li><strong>Spreadsheets and Data Analysis:</strong> Spreadsheets are shifting from manual tools to AI-driven engines. Instead of writing complex formulas, users will increasingly rely on AI agents that understand data requests and deliver instant insights. Tools like Excel Copilot and startups like <strong>Fundamental Research Labs</strong> are leading this change, their agent, <a href="https://twitter.com/funcresearch/status/1793679336857827774"><strong>Shortcut</strong></a>, reportedly solved complex Excel challenges 10× faster than human experts. As this evolves, custom scripts and dashboards may be replaced by simple conversations with a data-savvy AI.</li></ul><p><strong>As knowledge workers, what will our day-to-day work look like in a world without Microsoft Office or Google Workspace as we know them?</strong></p><p>In 3 to 5 years, knowledge work may no longer revolve around apps like Word, Excel, or Docs. Instead, what you’ll have on your MacBook (or any device) is a single, intelligent interface, likely voice- or chat-based, where you describe what you need, and an AI agent does the rest. You won’t open Excel to build a model; you’ll ask the agent to “analyze sales trends and highlight anomalies,” and it will handle everything, using underlying tools you never see.</p><p>The future of productivity won’t be defined by <em>which</em> app you open, but by <em>what</em> outcome you want, and how efficiently an AI can deliver it. Apps won’t be gone, but they’ll be background infrastructure, invoked and orchestrated by agents, not humans.</p><h4>Impact on Enterprises: Owning the Interface and the Moat in an AI-Driven World</h4><p>To stay competitive in the AI era, companies must <strong>own the final user interaction, in other words last mile of user experience, even if AI does the work behind the scenes</strong>. Whether it’s a smart home app or an internal employee portal, delivering answers through a branded, context-aware interface builds trust, consistency, and differentiation. Letting generic assistants handle that interaction layer risks losing user loyalty and control.</p><p>Internally, owning the UX could mean having a <strong>single company AI portal for employees</strong> e.g., a chat interface branded and configured for the company, through which all internal requests go. A competitor might use similar AI tech, but if your interface is more seamless, trustworthy, and context-aware, users (or employees) will prefer it, creating a <strong>moat of experience</strong>.</p><h3>Conclusion: From Apps to Agents and the Rise of Context</h3><p>We’re entering a new era of computing, one where software isn’t a tool you use, but a system that works on your behalf. Traditional apps like Excel, PowerPoint, and Word won’t vanish overnight, but their role is clearly shifting. They’re moving into the background, becoming <strong>infrastructure</strong> that AI agents call upon as needed. What comes to the foreground is <strong>intent </strong>and the intelligent <strong>interfaces</strong> that can understand and act on it.</p><p>In this world, the competitive edge won’t come from who has the best spreadsheet or dashboard, but from who owns the <strong>interface to the user </strong>and how well that interface understands the user’s <strong>context</strong>.</p><p>That’s where <strong>context engineering</strong> comes in. The next wave of innovation isn’t just about building smarter models, it’s about building smarter <strong>interfaces, systems, and experiences</strong> around those models. Ones that understand your data, your voice, your workflow, and your goals. Whether you’re serving consumers or enabling employees, the winners will be those who don’t just adopt AI, but embed it deeply, intentionally, and invisibly into how work gets done.</p><p>The future of productivity isn’t defined by which app you open, but by how easily you can achieve your goals. And it starts with systems that can understand your <strong>intent</strong> and act on it to produce the intended <strong>outcome</strong>.</p><p><strong>In this future:</strong></p><ul><li>There’s no need to navigate GUIs, we’ll express <em>intent</em></li><li><em>Intent</em> becomes the new input, driving actions across systems</li><li>The <em>interface</em> is everywhere (across devices)</li><li>The <em>logic</em> lives in the agent, not the app</li><li>And <em>context engineering</em> is how we make that experience personal, trusted, and seamless</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e8331d29cd47" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Charlie Munger’s Investing Principles Checklist]]></title>
            <link>https://medium.com/@neseozler/charlie-mungers-investing-principles-checklist-581c96cbc870?source=rss-20ba06d99a0f------2</link>
            <guid isPermaLink="false">https://medium.com/p/581c96cbc870</guid>
            <dc:creator><![CDATA[Nese Ozler]]></dc:creator>
            <pubDate>Thu, 31 Jul 2025 01:35:52 GMT</pubDate>
            <atom:updated>2025-07-31T01:40:36.615Z</atom:updated>
            <content:encoded><![CDATA[<p>I wanted to note down the investing checklist while reading <a href="https://www.amazon.com/Poor-Charlies-Almanack-Essential-Charles/dp/1953953239/ref=sr_1_1?dib=eyJ2IjoiMSJ9.vg5pUX_jT-eNweisJvvTBwk2PYGLpHbJOiv4_tnaeW1zlUPTcPZm1biw-1RS7BU2xgUN73UYmlXeGjZCBkck2oxMox8e1YvukuMZwvjzAfxChnOCfvqP2CgnZG7tTUrrxsJUbPdAoH0ygBV8fXSd8kUKO62SdUaZNYLBg9fzcbfj8gAQhKB4fdnoVUvD3XGinoXQjFBTKgGaT_BiQQKFjg_194-mgb_k-CPpeHsvxJU.eFWGX6uXgoCqY4MwXr46V1fmMZSXuUDSa6M-Jz1py3g&amp;dib_tag=se&amp;hvadid=695177619720&amp;hvdev=c&amp;hvexpln=67&amp;hvlocphy=1014221&amp;hvnetw=g&amp;hvocijid=3164079090725570665--&amp;hvqmt=e&amp;hvrand=3164079090725570665&amp;hvtargid=kwd-297927973344&amp;hydadcr=24631_13611696&amp;keywords=poor+charlie%27s+almanack&amp;mcid=98599a228cc336cfbc6ae845968cacdf&amp;qid=1753904209&amp;sr=8-1">Poor Charlie’s Almanack</a>. It’s a recap of his approach by using the checklist methodology. It’s an essential list for the investment analysis process.</p><p>These 10 principles aren’t meant to function like a <strong>linear to‑do list</strong>; they form an interconnected <em>gestalt</em> — each supports and reinforces the others. Munger stressed that it’s the entire system, not isolated steps, that builds wise investing mindset.</p><p>His overarching philosophy boiled down to <strong>four foundational values</strong>:<br><strong>Preparation</strong>, <strong>Discipline</strong>, <strong>Patience</strong>, and <strong>Decisiveness</strong>. Together, they drive what he called the “<em>lollapalooza effect”</em> — multiple mental models feeding powerful, aligned outcomes.</p><p>How to use the checklist:</p><ol><li><strong>Review all 10 principles</strong> before evaluating an investment or business opportunity.</li><li>Approach each principle thoughtfully — don’t rush.</li><li>Use it as a framework for thinking — not as rigid rules to follow blindly.</li><li>Remember that <strong>winning is the result of preparing</strong>, not just hoping; the checklist trains your mind over time for better judgment.</li></ol><h3>Risk — Always begin by evaluating risk, especially reputational risk</h3><p><em>All investment evaluations should begin by measuring risk, especially </em><strong><em>reputational</em></strong></p><ul><li>Include a <strong>margin of safety</strong></li><li>Avoid investments tied to people of questionable character</li><li>Ensure proper compensation for risk assumed</li><li>Be mindful of inflation and interest-rate exposures</li><li>Avoid big mistakes, Prioritize avoiding large permanent capital losses</li></ul><blockquote><strong><em>“It’s not worth doing something if you’re going to lose your reputation over it. You can’t buy a good reputation — you have to earn it, and you should guard it carefully.”</em></strong><em><br> — </em>Charlie Munger, in multiple talks and writings including Poor Charlie’s Almanack</blockquote><p>And more specifically on evaluating risk in investments:</p><blockquote><strong><em>“The wise ones bet heavily when the world offers them that opportunity. They bet big when they have the odds. And the rest of the time, they don’t. It’s just that simple.”</em></strong><em><br> — </em>Charlie Munger, Wesco Annual Meeting (2000)</blockquote><h3>Independence — Think for yourself instead of following the herd</h3><p><em>Only in fairy tales are emperors told they are naked</em></p><ul><li>Prioritize objectivity and rational judgment</li><li>Consensus doesn’t equal correctness; rely on analysis, not popularity</li><li>Herd behavior tends toward regression to the mean (merely average performance)</li></ul><blockquote><strong><em>“The right way to think is the way Zeckhauser plays bridge. It’s just that simple. And your brain doesn’t naturally know how to think the way Zeckhauser knows how to play bridge. Now you notice I put in that ‘simple’ idea — ‘think the way Zeckhauser plays bridge.’ It’s just not very simple. You have to think about odds and odds and odds.”</em></strong><em><br> — </em>Charlie Munger, “A Lesson on Elementary, Worldly Wisdom” (1994)</blockquote><h3>Preparation — Be relentlessly curious and disciplined</h3><p><em>The only way to win is to work, work, work, work, and hope to have a few insights</em></p><ul><li>Develop into a lifelong self-learner through voracious reading; cultivate curiosity and strive to become a little wiser every day</li><li>More important than the will to win is the will to prepare</li><li>Master mental models from major academic disciplines</li><li>Keep asking “Why?” to deepen understanding</li><li>Recognize that preparation far outweighs ambition; Opportunity comes to the prepared mind.</li></ul><p>I think my highlight from this list is “If you want to get smart, the question you have to keep asking is “why, why, why?”</p><blockquote><strong>“I constantly see people rise in life who are not the smartest… but they are learning machines. They go to bed every night a little wiser than they were when they got up.”</strong><br> — <em>Charlie Munger</em></blockquote><h3>Intellectual humility — Know and respect your limits</h3><p><em>Acknowledging what you don’t know is the dawning of wisdom</em></p><ul><li>Operate within your <strong>circle of competence</strong></li><li>Seek out disconfirming information</li><li>Resist false precision and self-deception</li><li>Above all, never fool yourself, and remember that you are the easiest person to fool</li></ul><blockquote><strong>“Knowing what you don’t know is more useful than being brilliant.”</strong></blockquote><blockquote>“You have to figure out what your own aptitudes are. If you play games where other people have the aptitudes and you don’t, you are going to lose.”<br> — <em>Charlie Munger</em></blockquote><h3>Analytic rigor — Apply scientific discipline to decision‑making</h3><p><em>Use of the scientific method and effective checklists minimizes errors and omissions</em></p><ul><li>Determine value apart from price; progress apart from activity; wealth apart from size</li><li>It is better to remember the obvious than to grasp the esoteric</li><li>Be a business analyst, not a market, macroeconomic, or security analyst</li><li>Consider totality of risk and effect; look always at potential second order and higher level impacts</li><li>Think forwards and backwards — <strong>Invert, always invert</strong></li></ul><blockquote><strong>“It is better to remember the obvious than to grasp the esoteric.”</strong><br> — Charlie Munger, emphasizing clarity over unnecessary complexity when evaluating opportunities</blockquote><h3>Allocation — Where to deploy capital is paramount</h3><p><em>Proper allocation of capital is an investor’s number one job</em></p><ul><li>Compare any investment to the next best alternative (opportunity cost)</li><li>Good ideas are rare—when odds heavily favor you, bet (allocate) big</li><li>Don’t get emotionally attached; allocate opportunistically</li></ul><blockquote><strong>“Experience tends to confirm a long‑held notion that being prepared, on a few occasions in a lifetime, to act promptly in scale, in doing some simple and logical thing, will often dramatically improve the financial results of that lifetime… A few major opportunities, clearly recognizable as such, will usually come to one who continuously searches and waits… And then all that is required is a willingness to bet heavily when the odds are extremely favorable…”</strong><br> — <em>Charlie Munger, Wesco Annual Meeting, 1996</em></blockquote><h3>Patience — Wait, don’t act hastily</h3><p><em>Resist the natural human bias to act</em></p><ul><li>“Compound interest is the eighth wonder of the world” (Einstein); never interrupt it unnecessarily</li><li>Avoid unnecessary transactions, taxes, or friction; never take action for its own sake</li><li>Be ready when <strong>luck or timing</strong> favors you</li><li>Enjoy the process along with the proceeds, because the process is where you live</li></ul><blockquote><strong>“The big money is not in the buying or the selling, but in the waiting.”</strong><br> — <em>Charlie Munger</em>, repeated in multiple venues, including Berkshire Hathaway commentaries</blockquote><blockquote><strong>“Most people are too fretful, they worry too much. Success means being very patient, but aggressive when it’s time.”</strong><br> — <em>Charlie Munger</em></blockquote><blockquote><strong>“A few major opportunities, clearly recognizable as such, will usually come to one who continuously searches and waits … And then all that is required is a willingness to bet heavily when the odds are extremely favorable, using resources available as a result of prudence and patience in the past.”</strong><br> — <em>Charlie Munger, from Poor Charlie’s Almanack</em></blockquote><h3>Decisiveness — Act boldly when the time is right</h3><p><em>When proper circumstances present themselves, act with decisiveness and conviction</em></p><ul><li>Be fearful when others are greedy, and greedy when others are fearful</li><li>Opportunities are infrequent—seize them with conviction</li><li>Opportunity meeting the prepared mind; that’s the game</li></ul><blockquote><strong>“Quickly eliminate the big universe of what not to do, follow up with a fluent, multidisciplinary attack on what remains, then act decisively when, and only when, the right circumstances appear.”</strong><br> — <em>Charlie Munger, </em>In other words: cut through the noise, identify the few high-quality opportunities, and commit confidently — but selectively — when they truly warrant action.</blockquote><h3>Change — Embrace complexity and adapt</h3><p><em>Live with change and accept unremovable complexity</em></p><ul><li>Accept that the world is complex and ever-changing; don’t expect the world to adapt to you</li><li>Challenge and revise long-held beliefs when warranted</li><li>Confront reality even when it’s uncomfortable</li></ul><blockquote><strong>“Those who will not face improvements because they are changes, will face changes that are not improvements.”</strong><br> — <em>Charlie Munger</em></blockquote><h3>Focus — Keep it simple and deliberate</h3><p><em>Keep things simple and remember what you set out to do</em></p><ul><li>Remember that reputation and integrity are your most valuable assets — and can be lost in a heartbeat</li><li>Guard against the effects of hubris and boredom</li><li>Don’t overlook the obvious by drowning in minutiae</li><li>Be careful to exclude unneeded information or slop: “A small leak can sink a great ship”</li><li>Face your big troubles; don’t sweep them under the rug</li></ul><blockquote><strong>“You need to have a passionate interest in why things are happening.That cast of mind, kept over long periods, gradually improves your ability to focus on reality. If you don’t have that cast of mind, you’re destined for failure even if you have a high I.Q.”</strong></blockquote><blockquote><strong>“When I want to read something I can tune everything else out… I think people who multi-task pay a huge price. … Concentrate hard on something that’s important. I did not succeed in life by intelligence. I did because I have a long attention span.”</strong><br> — <em>Charlie Munger</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=581c96cbc870" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>