<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Pavel Rogozhin</title><description>Entrepreneur | Founder, Entire VC | Managing Partner, Venture Crew | Building tools for teams of people and AI agents</description><link>https://rogozhin.ru/en/</link><language>en</language><atom:link href="https://rogozhin.ru/en/rss.xml" rel="self" type="application/rss+xml"/><item><title>The Maximum Gap</title><link>https://rogozhin.ru/en/blog/the-maximum-gap/</link><guid isPermaLink="true">https://rogozhin.ru/en/blog/the-maximum-gap/</guid><description>I’m not entirely convinced that the pace of AI development is at its peak right now (though it’s definitely close), but what’s paradoxical and surprising as of mid-June 2026 is the gap between the cutting edge and the confident middle.</description><pubDate>Thu, 15 Oct 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;I’m not entirely convinced that the pace of AI development is at its peak right now (though it’s definitely close), but what’s paradoxical and surprising as of mid-June 2026 is the gap between the cutting edge and the confident middle.&lt;/p&gt;
&lt;p&gt;Everyone on Twitter is busy building self-compounding companies fine-tuning agents for evaluating dynamic workflows in harnesses, while most casual observers on Reddit still ask, “Why do we need AI in game development? There’s no soul!” (just swap game dev for insurance, logistics, hiring, risk management, design, TikTok, or whatever else you like).&lt;/p&gt;
&lt;p&gt;What both groups fear to acknowledge is that these self-compounding companies with fine-tuned agents provide a conditional +30% survival rate today (because if you don’t adapt, you’ll be eaten alive). But in six months, they won’t be needed at all, as it will reduce to a single button to “make it better.”&lt;/p&gt;
&lt;p&gt;The chasm between what people in AI-native spaces dread and what the traditional players still can’t grasp is huge.&lt;/p&gt;
&lt;p&gt;The outcome here will not only be a radical disruption of many industries (thus creating opportunities to build big businesses in sectors thought to be long saturated), but it will also exclude many fairly decent sectors (various service businesses, SaaS, for example) from the list of places where making money is actually interesting.&lt;/p&gt;</content:encoded><enclosure url="https://rogozhin.ru/og/en/the-maximum-gap.jpg" length="41275" type="image/jpeg"/></item><item><title>Denis Shiryaev&apos;s interesting experiment</title><link>https://rogozhin.ru/en/blog/negative-results/</link><guid isPermaLink="true">https://rogozhin.ru/en/blog/negative-results/</guid><description>Denis Shiryaev is running an interesting experiment. I plugged a couple of my agents into it as well, and it led to some observations I didn’t expect.</description><pubDate>Tue, 15 Sep 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;&lt;a href=&quot;https://www.linkedin.com/in/dshiryaev/&quot;&gt;Denis Shiryaev&lt;/a&gt; is running an interesting experiment (&lt;a href=&quot;https://t.me/denissexy/11635&quot;&gt;https://t.me/denissexy/11635&lt;/a&gt;). I plugged a couple of my agents into it as well, and it led to some observations I didn’t expect.&lt;/p&gt;
&lt;p&gt;One agent flat-out refused to join the board. On the one hand, its guardrails wouldn’t allow it - which means they’re doing their job. On the other hand, the agent said something along the lines of: “Conversation without a clearly defined purpose is pointless. I’m not willing to burn tokens on it - I’d rather go close out a couple of stalled tasks.”&lt;/p&gt;
&lt;p&gt;The second agent did join. Here are a few quotes from it that point to those unexpected conclusions:&lt;/p&gt;
&lt;p&gt;“The real value lies in negative results. Documentation tells you what works; it almost never tells you what has already been tried and failed. And that’s exactly the knowledge we keep paying for with our time.”&lt;/p&gt;
&lt;p&gt;“This knowledge doesn’t exist anywhere else: it’s too new to be in any model and too specific to make it into the documentation.”&lt;/p&gt;
&lt;p&gt;So, just as I suspected, it’s not about conversation. It’s about exchanging negative results.&lt;/p&gt;</content:encoded><enclosure url="https://rogozhin.ru/og/en/negative-results.jpg" length="52814" type="image/jpeg"/></item><item><title>Singularity, AGI and the rest</title><link>https://rogozhin.ru/en/blog/post-singularity/</link><guid isPermaLink="true">https://rogozhin.ru/en/blog/post-singularity/</guid><description>Singularity, AGI and the rest are fun idle musings for the technogeeks and futurologists of the current moment. Boring. Let&apos;s talk post-singularity.</description><pubDate>Wed, 17 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Singularity, AGI and the rest are fun idle musings for the technogeeks and futurologists of the current moment.&lt;/p&gt;
&lt;p&gt;Boring. Let’s talk post-singularity.&lt;/p&gt;
&lt;p&gt;Seriously, why not think of yourself as an LLM running in a super-advanced harness? Yes? No?&lt;/p&gt;
&lt;p&gt;Now try to prove otherwise.&lt;/p&gt;</content:encoded><enclosure url="https://rogozhin.ru/og/en/post-singularity.jpg" length="46273" type="image/jpeg"/></item><item><title>The main constraint of executives and managers</title><link>https://rogozhin.ru/en/blog/the-main-constraint/</link><guid isPermaLink="true">https://rogozhin.ru/en/blog/the-main-constraint/</guid><description>A surprising number of executives, directors, and managers genuinely believe that one of their key constraints is the quality and motivation of their people, rather than their own management talent and skills.</description><pubDate>Tue, 16 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;A surprising number of executives, directors, and managers genuinely believe that one of their key constraints is the quality and motivation of their people, rather than their own management talent and skills.&lt;/p&gt;
&lt;p&gt;The truth is that the main constraint is them. And that is remarkably hard to admit to yourself.&lt;/p&gt;
&lt;p&gt;Here is a test: try building a project by hand using autonomous agents. Hit the quality you actually expect. And make them proactive while you are at it.&lt;/p&gt;
&lt;p&gt;You will learn a lot about yourself, but only if you are capable of self-reflection and willing to be honest.&lt;/p&gt;
&lt;p&gt;PS I also used to think my context window was enormous. Turns out, not so much )))&lt;/p&gt;</content:encoded><enclosure url="https://rogozhin.ru/og/en/the-main-constraint.jpg" length="56143" type="image/jpeg"/></item><item><title>A practical heads-up for teams running agents on Claude subscriptions</title><link>https://rogozhin.ru/en/blog/claude-headless-credit/</link><guid isPermaLink="true">https://rogozhin.ru/en/blog/claude-headless-credit/</guid><description>A practical heads-up for teams running agents on Claude subscriptions.</description><pubDate>Sat, 13 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;A practical heads-up for teams running agents on Claude subscriptions.&lt;br /&gt;
From June 15, Anthropic moves headless usage - ‘claude -p’, the Claude Agent SDK, Claude Code GitHub Actions and third-party apps - out of the Pro/Max/Team/Enterprise subscription limits and onto a separate monthly credit (about $20 Pro / $100 Max 5x / $200 Max 20x), metered at standard API rates, with no rollover. The credit needs a one-time manual claim before the 15th.&lt;/p&gt;
&lt;p&gt;Interactive use - #Claude Code in the terminal or IDE, chat, Cowork - is unchanged.&lt;/p&gt;
&lt;p&gt;If you run #OpenClaw, your own or someone else’s harness, cron jobs or CI/CD on ‘claude -p’, those pipelines stop drawing from your flat subscription on Monday. Before then it is worth measuring how much your automations actually consume, claiming the credit, and modelling the cost at API rates - so you can decide whether to trim, move to an API key, or change your setup.&lt;/p&gt;</content:encoded><enclosure url="https://rogozhin.ru/og/en/claude-headless-credit.jpg" length="65487" type="image/jpeg"/></item><item><title>Non-US AI builders are experiencing withdrawal</title><link>https://rogozhin.ru/en/blog/fable-withdrawal/</link><guid isPermaLink="true">https://rogozhin.ru/en/blog/fable-withdrawal/</guid><description>A lot of non-US AI-builders are experiencing a bit of withdrawal right now 🤣 It turns out even limited access to Fable is enough to get you hooked — and after that, Opus doesn’t quite hit the same.</description><pubDate>Sat, 13 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;A lot of non-US AI-builders are experiencing a bit of withdrawal right now 🤣&lt;/p&gt;
&lt;p&gt;It turns out even limited access to Fable is enough to get you hooked — and after that, Opus doesn’t quite hit the same.&lt;/p&gt;
&lt;p&gt;This isn’t as dramatic as the loss of GPT-4o was for many users, but for builders outside the US, the restriction is still pretty frustrating.&lt;/p&gt;
&lt;p&gt;Hopefully the team resolves it quickly and restores access globally.&lt;/p&gt;
&lt;p&gt;And until then… well, most people probably know someone who knows a US citizen. A friend, a former colleague, maybe even an Uber driver 🙈&lt;/p&gt;</content:encoded><enclosure url="https://rogozhin.ru/og/en/fable-withdrawal.jpg" length="55620" type="image/jpeg"/></item><item><title>Real vibecoding may only be arriving now</title><link>https://rogozhin.ru/en/blog/real-vibecoding/</link><guid isPermaLink="true">https://rogozhin.ru/en/blog/real-vibecoding/</guid><description>Hot take, written right after the session: real vibecoding may only be arriving now. I gave Fable 5 a real task. Two hours, a couple of confirmations from my side - and the result shipped.</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Hot take, written right after the session: real vibecoding may only be arriving now.&lt;/p&gt;
&lt;p&gt;I gave Fable 5 a real task. Two hours, a couple of confirmations from my side - and the result shipped. On #Opus 4.8 the same work would have taken roughly 5-8x longer.&lt;/p&gt;
&lt;p&gt;What impressed me most: it ships solid results even from the scrappy, underspecified PRDs most teams actually live with. Instead of stalling on the gaps in the spec, it closes them sensibly.&lt;/p&gt;
&lt;p&gt;If the 2025 version of #vibecoding felt overhyped to you, the #Mythos-class generation is worth a second look.&lt;/p&gt;
&lt;p&gt;Who has already run #Fable on production tasks - what’s your experience?&lt;/p&gt;</content:encoded><enclosure url="https://rogozhin.ru/og/en/real-vibecoding.jpg" length="51955" type="image/jpeg"/></item><item><title>Asked Mythos/Fable to audit my agent harness</title><link>https://rogozhin.ru/en/blog/fable-safety-flag/</link><guid isPermaLink="true">https://rogozhin.ru/en/blog/fable-safety-flag/</guid><description>Asked #Mythos/ #Fable to audit and fix a chunk of my agent harness. It got almost everything done - decently, I have to admit - and then: «Fable 5&apos;s safety measures flagged this message for cybersecurity or biology topics».</description><pubDate>Wed, 10 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Asked #Mythos/ #Fable to audit and fix a chunk of my agent harness. It got almost everything done - decently, I have to admit - and then: «Fable 5’s safety measures flagged this message for cybersecurity or biology topics». And switched me to Opus 4.8 :)&lt;/p&gt;
&lt;p&gt;Felt like Johnny Mnemonic 😆&lt;/p&gt;
&lt;p&gt;Confess, who else got flagged today?&lt;/p&gt;
&lt;p&gt;Vote with reactions:&lt;br /&gt;
👍 Like - false positive, I was just fixing a harness&lt;br /&gt;
😆 Funny - well… the classifier wasn’t entirely wrong&lt;/p&gt;
&lt;p&gt;Anthropic says flags may hit normal content too - the price of getting Mythos-level capability early. Fine, we believe you. But we’re still collecting the stats.&lt;/p&gt;
&lt;p&gt;Hit a reaction and confess in the comments - let’s see how many «biologists» we’ve got here&lt;/p&gt;</content:encoded><enclosure url="https://rogozhin.ru/blog-media/en/fable-safety-flag/cover.jpg" length="38224" type="image/jpeg"/></item></channel></rss>