<?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 Will Cheung on Medium]]></title>
        <description><![CDATA[Stories by Will Cheung on Medium]]></description>
        <link>https://medium.com/@willcheung?source=rss-a582df0e11ec------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*9HwORCPSRnusGX3dy3xmTg.jpeg</url>
            <title>Stories by Will Cheung on Medium</title>
            <link>https://medium.com/@willcheung?source=rss-a582df0e11ec------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 03:19:53 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@willcheung/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[AI Is Rewriting Tech Jobs. After 20 Years in Tech, This Is the Only Skill I’d Bet On.]]></title>
            <link>https://medium.com/vibe-coding-dad/ai-is-rewriting-tech-jobs-after-20-years-in-tech-this-is-the-only-skill-id-bet-on-ae21bf8bc9b5?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/ae21bf8bc9b5</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[careers]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[tech-jobs]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Thu, 29 Oct 2026 17:32:31 GMT</pubDate>
            <atom:updated>2026-10-29T17:32:31.149Z</atom:updated>
            <content:encoded><![CDATA[<h4>Every 10 years there’s a new Lego brick for builders. This time it’s AI.</h4><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FO1QnCnZZKZo%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DO1QnCnZZKZo&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FO1QnCnZZKZo%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/300fe6656d68c3a590e957aa35004cc3/href">https://medium.com/media/300fe6656d68c3a590e957aa35004cc3/href</a></iframe><p>This week Microsoft offered voluntary buyouts to 60,000 employees, about 27% of its workforce, as AI leads to an efficiency push. A few days earlier, Barclays said it would cut 5,000 jobs in an AI efficiency push, and American Express said it’s cutting 5% of its workforce to “prepare for AI.”</p><p>And that’s just one week. Every day for the past few years has felt uncertain. There are layoffs every week, and there’s a new model killing startups every week.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_nyTG1Vd00fQ9GDY.png" /></figure><p>I’ve been working in tech for 20 years. First as a software engineer, then as a founder, and now as a full-time product manager. I’ve never felt this uncertain about where my career is going, or what my kids will be doing 10 years from now. So I made a video about it. This post is the short version.</p><h3>Every 10 years, there’s a new Lego brick</h3><p>I grew up in Taiwan, and I started building PCs in middle school. You could buy CPUs, motherboards and CD-ROM drives at the market, snap them together and get a working computer. I didn’t have to know anything about the circuits.</p><p>Years later, Ruby on Rails did the same thing for apps. Big components, no plumbing. You just had to know the framework and you could build a whole app. I tell the full story <a href="https://youtu.be/O1QnCnZZKZo?t=90">in the video, from Taiwan to my first startup</a>.</p><p>Looking back, every 10 years or so there’s a new Lego brick:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/0*4RP-KhQcsruy1UTP.jpeg" /></figure><ul><li><strong>1990s, PCs:</strong> hardware engineers were super popular.</li><li><strong>2000s, Rails:</strong> software engineers were the crown jewels of careers.</li><li><strong>2010s, Mobile</strong>: continuation of software engineering bleeding into mobile tech.</li><li><strong>Now, AI:</strong> the winners are people who know how to use AI to build things and get more done.</li></ul><p>AI is just the next brick. It hides the circuits, the plumbing and the code, and gets you to the outcome faster.</p><h3>The one skill that doesn’t change</h3><p>The only skill that matters through all of this is critical thinking and problem solving.</p><p>Technology keeps changing. Hardware, software, mobile, and now AI. What stays the same is figuring out which problem is worth solving, then putting the pieces together to solve it.</p><h3>I got a front-row seat this year</h3><p>I built an AI app last year, Cal Autobot. You forward it any email and it puts the dates on your calendar. I shut it down the same weekend Meta launched Muse, its new AI agent. (I wrote about <a href="https://blog.vibecodingdad.com/p/what-meta-muse-means-for-builders">what Muse means for builders</a> here.)</p><p>What actually killed my app surprised me, and it wasn’t people vibe coding their own version. <a href="https://youtu.be/O1QnCnZZKZo?t=352">This is the part of the video where it clicked for me</a>, and I think it’s where a lot of jobs are heading too.</p><h3>What I’d do this week</h3><p>Start playing with AI and uplevel. You don’t have to chase every new tool. Pick one thing you do every week that you hate and hand it to AI. It could be research, it could be coding, or in my case, managing the kids’ calendar.</p><p>If you’re thinking about building something with AI, I go through the questions I ask myself first <a href="https://youtu.be/O1QnCnZZKZo?t=556">in the video at 9:16</a>. The first one is whether an AI agent could do it, if not today, then in two weeks or a month. The second one is about who you’re really building for.</p><h3>Where I land</h3><p>I don’t know what we as professionals are going to do in the near future. I’m taking it one step at a time, building stuff on the side and playing with AI. But I’m optimistic, because every 10 years or so there’s a new brick, and it lets us build something better.</p><p><a href="https://www.youtube.com/watch?v=O1QnCnZZKZo">Watch the full video here</a>. What’s one thing AI changed about your job this year? I’d love to hear from you in the comments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qthtzmJbYNvcHVD9ySfdbA.jpeg" /></figure><p>Sources: <a href="https://www.cnbc.com/2026/10/26/microsoft-msft-voluntary-buyouts.html">Microsoft buyouts (CNBC, Oct 26, 2026)</a> · <a href="https://www.cnbc.com/2026/10/22/barclays-ai-jobs-cuts.html">Barclays job cuts (CNBC, Oct 22, 2026)</a> · <a href="https://www.cnbc.com/2026/10/23/american-express-job-cuts.html">American Express job cuts (CNBC, Oct 23, 2026)</a> · <a href="https://www.trueup.io/layoffs">Layoff tracker</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ae21bf8bc9b5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vibe-coding-dad/ai-is-rewriting-tech-jobs-after-20-years-in-tech-this-is-the-only-skill-id-bet-on-ae21bf8bc9b5">AI Is Rewriting Tech Jobs. After 20 Years in Tech, This Is the Only Skill I’d Bet On.</a> was originally published in <a href="https://medium.com/vibe-coding-dad">Vibe Coding Dad</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Coding by Hand is for Losers]]></title>
            <link>https://medium.com/vibe-coding-dad/coding-by-hand-is-for-losers-4511cc9893e4?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/4511cc9893e4</guid>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[ai-coding]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ruby-on-rails]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Sun, 11 Oct 2026 00:11:49 GMT</pubDate>
            <atom:updated>2026-10-19T19:13:11.319Z</atom:updated>
            <content:encoded><![CDATA[<h4>From building PCs, to building apps with Ruby on Rails, to building with AI.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*byqJ-i9pvGbqW8L0.png" /></figure><p>The guy who got me building software just told a room full of programmers he’s done writing code by hand. And if you disagree? “The black pill is for fucking losers.”</p><p>He’s David Heinemeier Hansson (DHH), creator of Ruby on Rails. Twenty years ago, Rails is what made me super excited about building software.</p><p>I’ve been building things since I was in middle school. Growing up in Taiwan, I used to build my own custom PCs. Taiwan was mecca of computer parts like motherboards, CPUs, RAM, CD-ROMs, and sound cards (remember those?).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*R_fb6NgMFzSGHzML.jpeg" /><figcaption>One of hundreds of stores that sells graphics cards, CPUs, and other parts (2015)</figcaption></figure><p>Building a PC was pretty easy. Once you know the specs you want, it comes down to buying the right parts and assembly. It works just like lego pieces. That was 1995.</p><p>Software is different. Ten years later, even after receiving a Computer Science degree from Carnegie Mellon, it felt like I was just scratching the surface. I could tinker with code, but building an entire application still felt like something even “real programmers” can’t do by themselves.</p><p>Then I found Ruby on Rails.</p><p>Rails was the first framework that made software feel buildable to me in the same way a computer was buildable. MVC, routes, database schema migration. A lot of complexity suddenly had conventions and defaults.</p><p>I started learning Ruby. I still broke things constantly. But Rails dramatically shortened the distance between “I have an idea” and “holy crap, this actually works.”</p><p><strong>Rails made software feel like lego pieces</strong>.</p><h3>The guy who got me building software now says: stop coding</h3><blockquote><em>“We had a wonderful run chiseling that code by hand, and that it’s over.”</em></blockquote><p>DHH (David Heinemeier Hansson), who built Ruby on Rails, isn’t saying he stopped making software. It’s actually the opposite. He’s still building, but increasingly AI agents are doing the implementation instead of him typing every line himself.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FvDjW_dRyKXY%3Fstart%3D2361%26feature%3Doembed%26start%3D2361&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DvDjW_dRyKXY&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FvDjW_dRyKXY%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/2c8700e17f27f151941e8d35636f4cf9/href">https://medium.com/media/2c8700e17f27f151941e8d35636f4cf9/href</a></iframe><p>What I appreciated about his talk during <a href="https://rubyonrails.org/world/2026/">2026 Rails World conference</a> was how optimistic he was. Instead of “AI is taking your jobs,” he said now’s the best time to be alive to witness such transformation. Only “losers” write code by hand.</p><blockquote><strong><em>“At 37signals, a couple of weeks ago, we made the decision that it clearly means we’re done writing code by hand.</em></strong><em> We have gone pencils down on the idea that we were gonna write code by hand, as a normal course of business creating things.”</em></blockquote><p>A real, measurable example he cited was his coding agent that rewrote his flagship software, Basecamp, from Ruby to Rust → 99% less CPU and 95% less memory!</p><p>And listening to him, I realized that AI is just another abstraction layer like Rails was 20 years ago. It didn’t eliminate programming. It abstracted away a bunch of plumbing and made programming easy.</p><p>AI agents are pushing that abstraction much further.</p><h3>From vibe coding to agentic coding</h3><p>In 2025 when I start vibe coding, it felt like the breakthrough. I used Replit, describe something, and get working software back ridiculously fast. I remember asking AI to refactor my Rails 2 project to Rails 8.</p><p>It told me 6–8 months with dedicated team (without AI).</p><iframe src="https://cdn.embedly.com/widgets/media.html?type=text%2Fhtml&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;schema=twitter&amp;url=https%3A//x.com/willcheung/status/1956559760410792243&amp;image=https%3A//i.embed.ly/1/image%3Furl%3Dhttps%253A%252F%252Fpbs.twimg.com%252Fmedia%252FGycYRrHbUAAgkyy%253Fformat%253Dwebp%2526name%253Dlarge%26key%3Da19fcc184b9711e1b4764040d3dc5c07" width="500" height="281" frameborder="0" scrolling="no"><a href="https://medium.com/media/a8a04300ef5c12bf596a2772df82124e/href">https://medium.com/media/a8a04300ef5c12bf596a2772df82124e/href</a></iframe><p>After months of ferocious prompting and hundred of dollars, which is still cheaper than team of devs for 8 months, I was proud to announce a working prototype of the Rails 8 app.</p><iframe src="https://cdn.embedly.com/widgets/media.html?type=text%2Fhtml&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;schema=twitter&amp;url=https%3A//x.com/willcheung/status/1976020800420577671&amp;image=https%3A//i.embed.ly/1/image%3Furl%3Dhttps%253A%252F%252Fpbs.twimg.com%252Fprofile_images%252F2032622602502721536%252FtWEZ_dKK_400x400.jpg%26key%3Da19fcc184b9711e1b4764040d3dc5c07" width="500" height="281" frameborder="0" scrolling="no"><a href="https://medium.com/media/bf2c3d5205228065bcc8988635a3397e/href">https://medium.com/media/bf2c3d5205228065bcc8988635a3397e/href</a></iframe><p>Today, one year later, I can rewrite this app in three hours with a single /goal prompt and <strong>spend $20.</strong></p><p>Are we living in science fiction or what?</p><h3>What does it mean for my teens?</h3><p>While my kids are skeptical of AI, I’m not sure “learning to code” is the right path for them either. I wanted them to hear this legend speak, but all I think they’d react to is “his hair is weird” during the first few seconds before moving on. So I turned DHH’s speech (and added a little of my own flair) into a 60s video for them — as well as satisfying my builder curiosity around Opus 5.5.</p><p>So I gave the clip to Claude Code running Opus 5.5 and prompted it:</p><blockquote><em>Go through this DHH video. Rebuild it into an engaging explainer for developers and anyone who doesn’t care about developers, CS students, non-CS folks. You have complete creative freedom.</em></blockquote><blockquote><em>Keep tension. Compose the soundtrack yourself, fully synthesized in code. No samples, no audio files, no VSTs. Every cut must land on the beat. 1080p, 60fps.</em></blockquote><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FiBVMKmPPdEk%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DiBVMKmPPdEk&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FiBVMKmPPdEk%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/384dc2ab67e5620fb63ce98c625bac44/href">https://medium.com/media/384dc2ab67e5620fb63ce98c625bac44/href</a></iframe><p>The result is a demonstration of DHH’s point.</p><p>Opus transcribed the talk, analyzed the footage, came up with a visual concept, wrote the renderer, animated the scenes, synthesized the soundtrack, and synchronized everything to the beat — autonomously in 30 minutes. Everything was written in Javascript, no other frameworks like the ones I used in other <a href="https://blog.vibecodingdad.com/p/i-turned-a-multi-day-video-shoot-into-a-3-minute-ai-prompt-b101ae706d12">videos</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*52xhQmokMk9F3vDu.png" /></figure><p>As for my kids? They’re still goofing off.</p><h3>The AI could execute. I still had to judge.</h3><p>AI can’t read your mind. If you have a specific creative direction in mind, you’ll have to be <em>very</em> descriptive. But if you’re like me, who has never created an animation video in my life, the first version got me 80% of the way.</p><p>A few places that I had to tweak, for example, one animation showed programmers going dark, but it implied programmers were disappearing. That wasn’t DHH’s point, so we changed it to show programmers moving from keyboards to steering.</p><p>Another line said, “Typing just stopped paying,” but it sounded like programmers stopped getting paid. Wrong message. We changed that too.</p><p>So my contribution to a fairly complicated 2,000-line JavaScript video was mostly:</p><p><em>Too long. Wrong message. This part is off.</em></p><p>Which sounds hilariously unimpressive until you realize those were the decisions that determined whether the final thing was actually good.</p><p><strong>Those weren’t coding problems. They were judgment problems.</strong></p><h3>I’m writing less code and building more</h3><p>I still like understanding how things works. I’m enough of a nerd that I’ll happily prompt my agent at midnight trying to figure out why it did something stupid.</p><p>But increasingly, I’m not the person physically producing every piece. I didn’t even need to steer much anymore. More of my time goes toward deciding what should be built, what good looks like, what the agent misunderstood, and whether the result is actually correct.</p><p>Taste. Judgment. Context. Direction. Verification.</p><p>None of those are new skills. They’re just becoming a much bigger percentage of the job. If your craft is the “doing”, then you have to level up to steering.</p><h3>Full circle</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aOTjKetzBjunBW4U.png" /></figure><p>My builder journey has mostly been a series of abstraction layers — my lego pieces.</p><p>As a kid in Taiwan, I assembled the computer from ready-made components. Rails gave me enough structure that I could assemble the application. Now agents increasingly let me define the intent, context, constraints, and desired outcome while they handle all of the implementation.</p><p>Each time, some part of the previous craft became less important.</p><p>But the reason I’m still building never changed. I like making things. I’m writing less code than I used to. And I’m building more than ever.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4511cc9893e4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vibe-coding-dad/coding-by-hand-is-for-losers-4511cc9893e4">Coding by Hand is for Losers</a> was originally published in <a href="https://medium.com/vibe-coding-dad">Vibe Coding Dad</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Meta Muse: I’ve Never Been So Excited and So Depressed to Be a Builder]]></title>
            <link>https://generativeai.pub/meta-muse-ive-never-been-so-excited-and-so-depressed-to-be-a-builder-7136ad2d7cc7?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/7136ad2d7cc7</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[muse]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[meta-muse]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Tue, 22 Sep 2026 01:38:30 GMT</pubDate>
            <atom:updated>2026-09-22T16:17:08.370Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>Saaspocalypse is destroying one-trick ponies. How Meta Muse is swallowing personal productivity apps and what I’d do if I were to start a company today.</em></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HEII-c3Y9UvvLzmtwBua1Q.png" /></figure><p>Last June, I built an AI scheduling tool that you can “cc” on an email. It’s like your secretary: it knows the availability in your calendar, negotiates a time with the other person, and books the meeting for you.</p><p>Nothing to install. Just cc our email and “agent Cal” takes care of everything.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/0*9BCrP1CPnd1HjCbj.gif" /></figure><p>Coding agents weren’t nearly as good back then, but as a builder, it still wasn’t that hard to build a full-stack SaaS app that automagically reads incoming email context and uses LLMs to figure out what to do next.</p><p>If you read the <a href="https://blog.vibecodingdad.com/p/i-built-and-deployed-a-micro-saas-in-3-hours-here-s-what-i-learned">article I wrote back then</a>, you’ll know there are actually a lot of complexities beyond just having agents generate code.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ffdwgdjLFvkhcZn5.png" /></figure><h4>What I built</h4><p>Long story short, I had a ton of fun building it. I deployed the app and eventually added a few Chrome extensions as distribution channels.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NfsCR9oyMogh9TL4.png" /><figcaption>The main app — calautobot.com. It has Calendly features like booking link and AI scheduling.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*RKBINJXKYMfiUe_t.png" /><figcaption>Forward any emails or pictures, and we’ll create calendar event for you</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1P-H-vt73VD_DlC7.png" /><figcaption>Copy-cat of all the email trackers out there. Out of fashion but still super useful for power email users (like sales &amp; biz dev)</figcaption></figure><p>Building the main app alone would have taken me at least three months full-time before AI, but I built a prototype in a day, basically by feeding AI screenshots of Calendly and then figuring out the AI parts myself.</p><p>But the problem isn’t that anyone can use AI agents to code SaaS. The problem is <strong>who needs a single-purpose SaaS anymore, when everyone has capable AI personal assistants.</strong></p><h4>Why SaaSpocalypse for SMBs and prosumers is real</h4><p>Unless a SaaS has a vibrant ecosystem, proprietary data, or deeply embedded workflows in an organization, there is not really a moat against AI. Building SaaS for prosumers or SMBs used to be one of the best ways to start a software company: find a painful workflow, build a focused product, slap PLG on it, and grow from there.</p><p>That playbook breaks today.</p><p>This weekend, I tried <a href="https://muse.ai/">Meta Muse</a>, one of the best consumer AI agents I’ve tried so far. It’s an “OpenClaw moment” for consumers. I’d say Grok Bot came close, but it’s not free, and it’s more like a coworker than a personal assistant.</p><p>The onboarding is ridiculously easy with Muse as well. In less than 30 seconds, I got started <em>for free</em> with intuitive user interactions and super generous usage. Since Meta already has my personal info anyway, what’s the harm in connecting my email and calendar?</p><p>The big revelation is the breadth of things I can ask Muse to do without feeling like a chatbot or “work mode” in ChatGPT. A general-purpose agent can now do pieces of what hundreds of point-solution startups have built entire companies around: scheduling, research, filling out forms, dealing with customer service, negotiating bills, organizing email, managing calendars, and increasingly taking actions across the web.</p><p>Just last month, AI was mainly for nerds and developers, but now that Meta lowered the technical bar for everyone, I can actually see a plethora of mass-market, consumer use cases.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r3mzufR6vzANmDT4oTkCUg.png" /><figcaption><a href="https://x.com/ChrisUniverse/status/2101499530763972804">https://x.com/ChrisUniverse/status/2101499530763972804</a></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1022/1*Ko6ZnF_hBI6WQnkxgz5RDw.png" /><figcaption><a href="https://x.com/petergyang/status/2101033599319613533">https://x.com/petergyang/status/2101033599319613533</a></figcaption></figure><h4>What I’d build next</h4><p>This is the part that depresses me.</p><p><a href="https://calautobot.com/">CalAutobot.com</a> works. It solves a real problem. I spent a ton of time polishing it. A few years ago, it might have been the beginning of a perfectly reasonable SaaS business.</p><p>Today, I decided to finally end-of-life CalAutobot. General-purpose AI like Muse and Instinct can do hundreds of things so well, and scheduling meetings is no longer a product by itself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lqYj6TaL2LD2OGsP.jpeg" /></figure><p>Build something agents want.</p><p>If Muse, ChatGPT, Claude, or whatever agent wins can act on behalf of millions of users, their connector and tool ecosystems become the new distribution channels.</p><p>Forget UI (user interface), think Agent Interface (coining AI x AI). UI is for user-in-the-loop governance and visualizing dashboards, not for interacting with your app.</p><p>The cost of building is basically zero. It’s also why I’ve never been more excited to be a builder. But the hard part is figuring out <strong>what is still worth building.</strong></p><p>For nerds who want to play with it, here’s the <a href="https://github.com/willcheung/CalAutobot">repo</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*TKS_m8T80AKaUvIc.png" /></figure><p>This story is published under the <a href="https://generativeai.pub/">Generative AI</a> publication. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*gnbPduC85L4ohAKz.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7136ad2d7cc7" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/meta-muse-ive-never-been-so-excited-and-so-depressed-to-be-a-builder-7136ad2d7cc7">Meta Muse: I’ve Never Been So Excited and So Depressed to Be a Builder</a> was originally published in <a href="https://generativeai.pub/">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Gave GPT-6 Astra a 3-Hour Coding Job. You Probably Don’t Need It.]]></title>
            <link>https://medium.com/vibe-coding-dad/i-gave-gpt-6-astra-a-3-hour-coding-job-you-probably-dont-need-it-a5c52a5197f8?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/a5c52a5197f8</guid>
            <category><![CDATA[hermes-agent]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[openai-codex]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[gpt-6-astra]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Tue, 08 Sep 2026 02:14:07 GMT</pubDate>
            <atom:updated>2026-09-10T06:02:19.447Z</atom:updated>
            <content:encoded><![CDATA[<p>What a 3-hour refactor taught me about where Astra shines, where it’s overkill, and the future of autonomous coding.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*omIWcCisnpdsFeSm.png" /></figure><p>OpenAI released <a href="https://openai.com/index/gpt-6-astra/">GPT-6 Astra</a> this week, describing it as its most capable model for complex, end-to-end work. According head of product:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O6QSo1k27DX7O8Zqr6Fecg.png" /></figure><p>And the usual X mafia went crazy about “Astra is AGI” posts, I also have to jump on the bandwagon and see it for myself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C_GSHwRErmuByTcgLxCJvA.png" /><figcaption><a href="https://x.com/JensenHuang/status/2096700264569090384?s=20">https://x.com/JensenHuang/status/2096700264569090384</a></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rTaZJgkXPV8GDIy6aoCHBA.png" /><figcaption><a href="https://x.com/AlexFinn/status/2096065384290128372">https://x.com/AlexFinn/status/2096065384290128372</a></figcaption></figure><h3>My refactoring project: AI Hedge Fund Analyst</h3><p>I’ve been building an <a href="https://blog.vibecodingdad.com/p/using-karpathys-wiki-how-i-turned-an-ai-agent-into-a-stock-analyst-2261baa54795">AI stock research system</a> that continuously researches companies and markets and generates investment analysis. It started with a wiki to help me remember every AI bottleneck stock I was reading about. It will scan X, Reddit, news sites and various sources regularly, and compound knowledge to the wiki. Then I can easily recall them:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*OemyTZsHXcfbxJGj.png" /></figure><p>And now it has evolved into a self-learning, decision system and dashboard. My favorite is actually the daily market brief and its “conviction list”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*fy6mIQtPuRBHL2z8.png" /></figure><p>Two goals I wanted to accomplish during the refactor:</p><ol><li>Untangle the wiki, automation, and dashboard code.</li></ol><ul><li><strong>Goal:</strong> Create a repo for the dashboard so I can tweak look &amp; feel without changing underlying data.</li></ul><p>2. Redesign the dashboard for easier reading and create a design system</p><ul><li><strong>Goal:</strong> Consistent look &amp; feel when new data is published</li></ul><blockquote><a href="https://blog.vibecodingdad.com/subscribe">Subscribe to Vibe Coding Dad for more AI projects and builds</a></blockquote><h3>I cranked it up to High Effort and gave the agent the goals</h3><p>I used to go back-and-forth with my agent to fix things piecemeal. With this refactor, I used /goal so I didn’t have to babysit the process. Tibo also gave us two weekly resets to play with Astra, so I cranked it up to High Effort.</p><p>After meta-prompting and brainstorming with AI on the goal, it gave me a .md file with specific design system, color schemes, layouts, and how to present text information, including tickers and green/red indicators. High-level, the spec defined:</p><ul><li>what should change</li><li>what should be preserved</li><li>how the result should be verified</li><li>what counted as finished</li><li>where human approval was required</li><li>pro-tip: always ask AI what else you didn’t think of</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cwdj4nf-NHITW9sR.png" /></figure><h3>The results were disappointing</h3><p>After working for 3.5 hours of working on High effort, I eagerly awaited my brand new dashboard.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gBhSCVpK9D5oYsCi.png" /></figure><p>Maybe because UI isn’t Astra’s strong suite, or my goals weren’t specific enough (or too specific), or the fact I should have been using Codex instead of Hermes as harness, but the results require another few hours of tweaking. My first question was — what happened to my Market Brief (my favorite feature of the dashboard).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*k83HbCDuHEd1SUkH.png" /></figure><p>After another few hours of tweaking and using <strong>three weekly resets</strong>, it finally came together. The entire specification → implementation → review → revision → production verification window was roughly <strong>12 hours and 40 minutes elapsed</strong>, including my reviews, interruptions, and analytics setup.</p><p>A separate source-control and portability goal took <strong>another hour </strong>on actually enforcing the design system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LEFi6ySXNycWLrzb.png" /></figure><p>Total time to refactor my UI:<strong> 14+ hours</strong></p><h3>Before vs After</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NxlrIAwj9QS0K90F.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*QUqvIvqahukdGJXu.png" /></figure><h3>This feels different from vibe coding</h3><p>Similar to Fable 5.1, Astra is designed for long-running tasks. <strong>Instead of telling Michael Jordan how to shoot a basketball, you should tell the model more like “I want to win this game, figure it out.”</strong></p><p>Out of the 13+ hours I spent with AI, I probably only actively worked on this for a few hours. Other times I spent driving kids around and playing Civ 7 (if you’re curious, it’s my first time playing Civ, and learning curve is too steep).</p><p>It handled hours of implementation, debugging, testing, and verification while I moved up a level and spent more of my time defining outcomes, reviewing the product, finding failure modes, and deciding what was actually good enough to ship.</p><p>Tokenmaxxing is also at another level. I have never hit my Pro plan limit before, but Astra burned through my weekly usage limit <strong>three times</strong>. Thankfully, OpenAI handed out free resets to try Astra — great business idea to unlock builders.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ICKtdnglZcW2Ila_k-7u3A.png" /></figure><h3>Pro Tips</h3><p>My take — don’t use Astra unless you have a lot of money to spend, a lot of patience to wait around, or solving gnarly complex problems. Everyone’s tempted to use the latest and greatest, but in this case, the latest isn’t the greatest at addressing most use cases.</p><p><strong>Tip #1:</strong> Always use Codex, which is specifically tuned for Astra.</p><p><strong>Tip #2:</strong>/goal command can work wonders for a long-running “AGI” model like Astra, but make sure you have the budget and very, very clear goals. Otherwise you&#39;ll spend more tokens fixing the results.</p><p><strong>Tip #3:</strong> Use Astra Low for everything.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QwbWAtfO-zrOltIqEPpkkQ.png" /></figure><p>Tip for long-running jobs:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-xmE2nGzCgHLmttZjo8qLg.png" /><figcaption><a href="https://x.com/mattshumer_/status/2095723177389232540">https://x.com/mattshumer_/status/2095723177389232540</a></figcaption></figure><h3>It’s live</h3><p>The result is <strong>AI Hedge Fund Analyst v0.1 </strong>— <a href="https://market-analyst.vibecodingdad.com/">have a look</a>.</p><p>You can browse the Daily Brief, company research, valuation models, and top stock picks. It’s a public research tool and an experimentation right now, not investment advice.</p><p>For the curious, here’s <a href="https://github.com/willcheung/ai-hedge-fund-analyst">dashboard repo</a>. <em>Note: actual data and automations are the backend that powers this agent, and isn’t available in repo yet. I will slowly add them.</em></p><p>But I’m actually more curious about something else:</p><p><strong>What improvements should I make so you’d pay for this?</strong></p><blockquote><a href="https://blog.vibecodingdad.com/subscribe">Subscribe to Vibe Coding Dad for more AI projects and builds</a></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a5c52a5197f8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vibe-coding-dad/i-gave-gpt-6-astra-a-3-hour-coding-job-you-probably-dont-need-it-a5c52a5197f8">I Gave GPT-6 Astra a 3-Hour Coding Job. You Probably Don’t Need It.</a> was originally published in <a href="https://medium.com/vibe-coding-dad">Vibe Coding Dad</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[PM Brain: Every Product Manager Needs One]]></title>
            <link>https://medium.com/vibe-coding-dad/pm-brain-every-product-manager-needs-one-7955a8ce492b?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/7955a8ce492b</guid>
            <category><![CDATA[product-manager]]></category>
            <category><![CDATA[customer-success]]></category>
            <category><![CDATA[vector-database]]></category>
            <category><![CDATA[second-brain]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Sat, 08 Aug 2026 00:09:08 GMT</pubDate>
            <atom:updated>2026-08-20T00:27:15.367Z</atom:updated>
            <content:encoded><![CDATA[<h4>I forked GBrain to help write PRDs. The better use case was a customer memory layer that powers meeting prep, roadmap decisions, and product evidence.</h4><p>Product decisions come from scattered signals. I built a graph to connect them.</p><p>I forked Garry Tan’s <a href="https://github.com/garrytan/gbrain?utm_campaign=pm-brain-my-gbrain-fork-for-product-managers-and-the-use-cases-i-didn-t-expect&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>GBrain</em></a> because I thought product managers needed a better way to write PRDs. That was partly right. It does help with PRDs, but the more useful thing was not the document. It was the memory behind the document: which customers felt the pain, where the evidence came from, what sales and support had already heard, and whether the product had tried to solve something like it before.</p><p>In a large product org, the hard part is rarely writing the artifact. The hard part is reconstructing reality from different stakeholders, roadmap review, launch decision, or customer call where the decision signals matter. <strong>PM Brain</strong> became the single source of truth acting like customer memory.</p><h3>Problem: B2B PM’s signals are scattered and mostly unstructured text</h3><p>A PM doesn’t make a decision from one source. They pull from customer calls, support tickets, product analytics, adoption data, sales blockers, NPS, Slack threads, prior launches, and gut instinct. Before automating any PM artifact, I asked three questions: where do PMs get the sources they use to decide, where do those decisions enter the build lifecycle, and what comes out the other side. That became the PM Brain.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Hf2sCVN-GLa2vZbjbC4lBg.png" /></figure><p>For example, sales says the UI is blocking a deal. Support sees repeated tickets. Product analytics shows low adoption in one part of the funnel. Customer calls have the quote in plain English. Feature request logs have five variations of the same ask. NPS comments have the frustration, usually with less diplomacy. Any one of those signals is easy to dismiss. Together, they are the product story.</p><h3>Why I used GBrain instead of plain RAG</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hit5QITh5uMY2AAnhCQzvA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DKKYRuVmM77oLSTlnL-IXQ.jpeg" /></figure><p>Most RAG systems are built for lookup. Embed documents, search chunks, summarize the results, hope the answer is grounded enough. That is fine when the job is “find the policy.” It is weak when the job is “explain why this customer pain keeps showing up across six systems and what we already tried.”</p><p>Product work is graph-shaped. Customers connect to pain points. Pain points connect to feature themes. Feature themes connect to roadmap items, PRDs, release notes, support tickets, usage patterns, blocked deals, and old decisions nobody remembers until they become expensive again.</p><p>GBrain fit that shape better. Its hybrid search combines semantic search, keyword search, and graph traversal. It can cite sources and say when it does not know something. It also wires the graph on ingest without paying for an LLM call on every edge, which matters once the corpus gets large. That last part is not glamorous. It is the difference between a demo and something you can afford to keep running.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Ry30xQzEnhosvcxc.png" /></figure><p>GBrain’s hybrid search runs three retrieval modes at once (vector for semantic lookup, keyword for exact names, graph for relationships), backs answers with citations, and flags what it doesn’t know instead of guessing. It also auto-wires the graph on ingest with zero LLM calls, which keeps the whole thing cheap to run at scale.</p><blockquote><em>Subscribe to my </em><a href="https://blog.vibecodingdad.com/subscribe"><em>newsletter</em></a><em> for AI tools and hacks.</em></blockquote><p><strong>If you want to see gbrain tutorials:</strong> GBrain shipped in April 2026 and picked up a fast following. <a href="https://www.marktechpost.com/2026/05/22/a-step-by-step-coding-tutorial-to-implement-gbrain-the-self-wiring-memory-layer-built-by-y-combinators-garry-tan-for-ai-agents/?utm_campaign=pm-brain-my-gbrain-fork-for-product-managers-and-the-use-cases-i-didn-t-expect&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>MarkTechPost</em></a> has a full install-to-MCP walkthrough with real terminal output, and <a href="https://vectorize.io/articles/gbrain-review?utm_campaign=pm-brain-my-gbrain-fork-for-product-managers-and-the-use-cases-i-didn-t-expect&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>Vectorize’s review</em></a> is a fair-minded look at where the project holds up and where it doesn’t. Worth reading before you fork it.</p><h3>The graph is grounded with accounts</h3><p>I anchored on account (customer) data first, since accounts are the one object product, sales, CS, and support all share. From there I extracted pain-point themes from the qualitative data: accounts, pain points, customer asks, feature themes, product areas, and the evidence behind each one. Edges connect accounts to pain points, pain points to feature themes, and feature themes back to artifacts like PRDs and release notes.</p><p>The edges matter more than the nodes. An account links to a pain point. A pain point links to a feature theme. A feature theme links to evidence, prior artifacts, and eventually release notes. That changes the question from “find notes that mention Feature X” to “which accounts feel this pain, what did they say, which deals are blocked, and have we already shipped something related?” That is closer to the real PM question.</p><p>End users connect Claude through MCP and ask the brain directly. The prototype started as PM tooling, but it quickly escaped the PM box. PMM, account teams, and leadership all wanted different views of the same customer memory. The current graph has 100k+ pages and 225k+ edges.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ChaAxj2BlLreA4K3.png" /></figure><h3>The surprise use case was meeting prep</h3><p>I originally built this to gather customer info, automate writing PRDs and build features grounded by customer asks. The better (and more common) use case turned out to be walking into a customer meeting prepared with customer context. That sounds obvious after the fact. Go-to-market teams talk to customers constantly, product hears a fraction of those conversations, and customers repeat the same context to sales (with turnovers), support, CS, PM, and leadership until someone finally pieces together the story.</p><p>Nobody enjoys that. Customers especially. The brain can generate an account prep note before a meeting: recent pain points, open asks, related product areas, prior conversations, evidence links, and what changed since the last touch. The PRD assistant was useful, but the account memory was valuable to more people.</p><p>The brain generates a prep note like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sMBLCPHUQoBX5OyL.png" /></figure><p>Account teams used it before calls. Partner marketing pulled pain themes. Execs used it for account prep. PMs used it to separate loud requests from patterns. The questions changed too: what should I know before this customer meeting, which pain points show up in accounts over $1M ARR, what are customers saying about onboarding friction, which product areas create the most repeated support pain, and did we already promise something related?</p><h3>Keep the brain separate from the UI</h3><p>The meeting prep dossier above is not hand assembled. Account briefs run through a deterministic HTML generator with an agentic search and summarization step, so the same account produces the same brief when the data has not changed. That matters because a brief that changes on every rerun is not something you can trust in a live meeting.</p><p>The graph stays underneath as the shared memory layer. The view on top can change based on user’s preference. Today it is an account brief. Tomorrow it can be a pain point leaderboard, a launch-readiness view, a PRD evidence pack, or a product area dashboard. The important design choice is that the AI agent nor the UI does not own the memory. The graph does.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*h3cgR2I91gK5a_s4.png" /></figure><h3>Where this goes next</h3><p>The next version combines unstructured and structured data. The PM Brain has qualitative signal. The data warehouse has structured signal: usage, adoption, retention, feature telemetry. The real product is both, which means a PM can finally ask whether the customers complaining about a workflow actually show the same pain in their usage data, or whether adoption moved after a feature shipped.</p><p>Release notes are next on the ingestion list, since they close the loop end to end. Slack and email are harder. The signal is real but the noise is brutal, and ingesting them blind is how you build a haunted attic with embeddings.</p><h3>The main lesson is separating the brain vs UI</h3><p>As you may have heard many times: AI is not the solution, it is the vehicle that takes you to the solution. In this case AI is used two ways: building the brain, and building the customized dashboard (or PRDs / other artifacts). Once an AI agent can tab into this brain, it unlocks myriads of use cases by building any custom solutions on top of it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7955a8ce492b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vibe-coding-dad/pm-brain-every-product-manager-needs-one-7955a8ce492b">PM Brain: Every Product Manager Needs One</a> was originally published in <a href="https://medium.com/vibe-coding-dad">Vibe Coding Dad</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Turned a Multi-Day Video Shoot Into a 3-Minute AI Prompt]]></title>
            <link>https://medium.com/vibe-coding-dad/i-turned-a-multi-day-video-shoot-into-a-3-minute-ai-prompt-b101ae706d12?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/b101ae706d12</guid>
            <category><![CDATA[heygen]]></category>
            <category><![CDATA[google-gemini]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-video-generator]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Sun, 21 Jun 2026 16:47:55 GMT</pubDate>
            <atom:updated>2026-08-20T00:29:41.681Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OaczBenA6RPiL1n3SgZ3DA.jpeg" /></figure><p><em>This is not a sponsored post. I experimented the tools to bring you my honest take for my use case.</em></p><p>I’ve been experimenting with AI video creation, and it feel honestly feels like magic. What used to require specialized tools, production experience, and a decent amount of coordination can now start with a prompt, a reference image, a voice recording, and a willingness to iterate. This must the the feeling of a non-coder vibe coding for the first time.</p><p>For this experiment, I wanted to see if I could create a short cinematic video where a digital version of me teaches a concept while moving through a scene. The final result came from a <a href="https://www.heygen.com/"><strong>HeyGen</strong></a><strong> avatar </strong>created from one picture and a voice recording, combined with <strong>Seedance </strong>video generation through HeyGen. I also tested the same prompt in <strong>Google’s </strong><a href="https://deepmind.google/models/gemini-omni/"><strong>Gemini Omni</strong></a> using the same picture and voice.</p><p>The interesting part is that each tool exposed a different strength and failure mode.</p><h3>Video generated by HeyGen in One Shot Prompt</h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FaJnKPpof168%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DaJnKPpof168&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FaJnKPpof168%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/3eb6172b6731c09b72a596d1a3291d70/href">https://medium.com/media/3eb6172b6731c09b72a596d1a3291d70/href</a></iframe><h4>The prompt is the new production brief</h4><p>Here’s the one-shot prompt I used:</p><blockquote>Will (avatar) moves through a massive early industrial factory floor. Belts, gears, pulleys, brass machines, and workers moving parts from station to station. Warm sparks from machinery. Golden industrial light. Camera tracks alongside him continuously throughout — medium close-up, never static. No music. Will moves through the factory as he teaches. He places one hand briefly on a moving conveyor rail. He picks up a small metal part — examines it briefly — sets it back into the workflow and keeps walking. Natural fluid movement throughout. He looks directly at camera the entire time. Will teaches directly to camera while moving.</blockquote><blockquote>Exact words only: <br>“AI agents are not magic employees. They are workflows with a brain in the middle. If the steps are vague, the agent wanders. If the inputs, tools, checks, and handoffs are clear, suddenly it starts looking useful.”</blockquote><blockquote>At 11 seconds — camera tracking alongside him — a weathered wooden door becomes visible ahead. It is already there. Set into a brick factory wall between two machines. It has always been there. He approaches it still looking at camera still talking. He pushes it open — warm office light already fully established beyond it. Bookshelves visible. Window with garden. Microphone. He walks through still looking at camera. Crosses to his desk. Sits naturally into exact seated position from reference image. Looks at camera. Ready. Camera tracks continuously from factory floor through door to seated position. Never cuts. Never loses his face. Clear and stable facial features. Widescreen. Shallow depth of field. Film grain. 4K HD. No blur. No ghosting. No music.</blockquote><p>This prompt is doing the work that used to be spread across a creative brief, storyboard, shot list, script, camera direction, and editing notes. It defines the scene, movement, camera behavior, dialogue, timing, lighting, facial continuity, and constraints.</p><h3>How I did it with HeyGen</h3><p>If you’ve tried recording yourself in a video, you’d know the setup: lighting, camera anxiety, microphone issues, retakes, editing, and the somehow universal truth that recording a 30-second clip can easily take 45 minutes.</p><p>HeyGen is built for businesses and creators who need to produce polished videos without turning every idea into a full production project. It can turn scripts into videos, photos into videos, upscale content, auto-edit clips, and handle many of the workflows you would expect around explainer videos, product videos, training content, and social content.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ZCQ5UldvgKpJbSKz" /></figure><p>Because of that focus, HeyGen has a strong avatar layer. I gave it a picture and a short voice recording, and it created a usable digital version of me that could be applied across different video formats. That is the real unlock: once the avatar exists, the workflow becomes repeatable. You can create founder updates, product explainers, sales videos, training modules, or multilingual versions of the same message without recording from scratch every time.</p><h3>What Seedance added through HeyGen</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EmaDw9zBcGurMTtp" /></figure><p>Seedance was the cinematic layer inside the HeyGen workflow. The factory floor, moving camera, industrial lighting, physical motion, door transition, and office reveal came from the video generation side. That is what changed the output from a basic talking-head avatar into something closer to a short visual story.</p><p>I did not use Seedance as a separate standalone workflow. I used it directly through HeyGen’s Seedance integration, which made the process much easier. That combination is important: HeyGen handled “make this feel like me,” while Seedance handled “make the scene feel alive.”</p><p>Together, they got much closer to the result I wanted than either layer would have on its own. The avatar gave the video identity. The generated scene gave it motion and metaphor.</p><h3>Google Gemini Omni: what it did well, and where it struggled</h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F_CR6QsbUK18%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D_CR6QsbUK18&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F_CR6QsbUK18%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/f881c2f5b5be1196b4d72a5545668c4b/href">https://medium.com/media/f881c2f5b5be1196b4d72a5545668c4b/href</a></iframe><p>I also tested the same prompt in Google Gemini Omni using the same picture and voice. The avatar likeness actually worked pretty well. It looked like me, which was a pleasant surprise and a big improvement over what I expected.</p><p>Where Omni really shines is workflow simplicity. Uploading the avatar image and voice felt more integrated. With HeyGen, avatar creation and video generation are more separate steps. With Omni, the process felt easier and more unified: give it the image, give it the voice, give it the prompt, and iterate from there.</p><p>The tradeoff was prompt adherence. My prompt specifically asked for a continuous tracking shot with no cuts, but Omni still produced frames that felt cut together rather than one seamless camera move. The output was good, but it took four tries, and even then it did not fully preserve the “single continuous shot” requirement.</p><p>That is an important distinction. Omni was strong on ease of use and likeness. HeyGen + Seedance was stronger for getting closer to the specific cinematic structure I wanted. Neither workflow was perfect, but each revealed where the tools are heading.</p><h3>Quick comparison: Gemini Omni vs. HeyGen</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cN1BevLqHhcn-ZDRhRsJEg.png" /></figure><blockquote>For more hands-on tips and trips with AI, check out my newsletter at <a href="https://learn.vibecodingdad.com/">https://learn.vibecodingdad.com/</a></blockquote><h3>The bigger lesson: vibe coding and vibe video are more alike than they look</h3><p>AI video creation is starting to feel a lot like vibe coding. With vibe coding, the prompt is only part of the workflow. You still need the right tool, whether that is Claude Code, Codex, Cursor, Replit, or whatever comes next. You still need to understand the system you are building, and you still need to know when the output is wrong, fragile, overcomplicated, or quietly making a mess behind the scenes.</p><p>AI video has the same pattern. The prompt matters, but the tool matters too. HeyGen, Seedance, and Gemini Omni are not interchangeable. Each has different strengths, assumptions, and failure modes. The same prompt can produce something impressive in one workflow and something unusable in another.</p><p>And just like with coding, judgment is the real multiplier. You need to know what good looks like. You need taste. You need to understand pacing, message clarity, visual metaphor, audience expectations, and where the model is likely to drift. You need to decide what is worth fixing and what is good enough to ship.</p><p>This is the part people sometimes miss about AI tools. They lower the barrier to starting, but they do not remove the need for judgment. In fact, they make judgment more important because the iteration cycle is so fast. You can generate five versions quickly, but you still need to know which one works and why.</p><h3>What this changes</h3><p>The big shift is not just that AI can generate video. The big shift is that video creation is becoming prompt-native and iteration-native.</p><p>Before, the hard parts were equipment, editing tools, production skill, filming, and time. Now, a lot of that friction is compressed. The new bottleneck is clearer: can you describe what you want, choose the right tool, evaluate the output, and iterate toward something useful?</p><p>That is a meaningful change for builders, founders, marketers, educators, and anyone who has ideas but not a production team sitting around waiting for instructions. You can test ideas in minutes, turn abstract concepts into visual stories, create content without waiting on a full production cycle, and experiment with formats that would have been too expensive or annoying before.</p><p>The first output will still be weird sometimes. That is part of the deal. But the speed of learning is the unlock.</p><h3>My takeaway</h3><p>AI video is not replacing creativity. It is compressing the distance between an idea and something that’s not AI slop.</p><p>That does not make everyone instantly great at video, just like AI coding tools do not make everyone instantly great at software. But they do give more people the ability to experiment, prototype, and publish.</p><p>And once that distance collapses, the builders have an advantage. Builders are not waiting for permission, perfect tools, or a studio budget. We are trying stuff, learning what works, and getting a little better each time.</p><p>Right now, that might be the whole game.</p><p>Want more practical AI video workflows and honest build notes? <a href="https://blog.vibecodingdad.com/subscribe">Subscribe</a> to Vibe Coding Dad.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b101ae706d12" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vibe-coding-dad/i-turned-a-multi-day-video-shoot-into-a-3-minute-ai-prompt-b101ae706d12">I Turned a Multi-Day Video Shoot Into a 3-Minute AI Prompt</a> was originally published in <a href="https://medium.com/vibe-coding-dad">Vibe Coding Dad</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Karpathy’s Wiki: How I Turned an AI Agent into a Stock Analyst]]></title>
            <link>https://generativeai.pub/using-karpathys-wiki-how-i-turned-an-ai-agent-into-a-stock-analyst-2261baa54795?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/2261baa54795</guid>
            <category><![CDATA[andrej-karpathy]]></category>
            <category><![CDATA[ai-stock-analysis]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[hermes-agent]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Tue, 05 May 2026 05:48:45 GMT</pubDate>
            <atom:updated>2026-08-20T00:27:41.427Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HpqqLrS342xqvmc_.png" /></figure><p>If you’re still using AI as a chatbot to summarize emails that you won’t read anyway, you’re using it wrong.</p><p>I decided to use mine to research stocks for me as a hobby project.</p><h3>The Context: News Overload</h3><p>I spend too much time at the intersection of AI and stocks. For a while, my setup was basic: Hermes agent that summarizes 20+ newsletters into a morning email. It felt productive, but it was a pretty weak use case.</p><p>I was reading AI and stock news, but none of it was compounding. I’d find a great AI thesis on Monday in a podcast, and by Friday, I’d forgotten how it connected to the semiconductor bottleneck I read about on X three weeks ago.</p><p>Then Andrej Karpathy posted about the “LLM Wiki” — the idea that AI shouldn’t just answer questions, it should maintain a persistent, evolving artifact of knowledge.</p><p>I realize I never retain any knowledge, and when I did, they were siloed. I needed a brain that remembers and connects the dots for me.</p><p><strong>My setup:</strong></p><ul><li>Agent harness: Hermes Agent + custom skills</li><li>Model: GLM-5-Turbo</li><li>“Brain”: Andrej Karpathy’s <a href="https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f?utm_campaign=using-karpathy-s-wiki-how-i-turned-an-ai-agent-into-a-stock-analyst&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>LLM wiki</em></a></li></ul><h3>The Build: From Script to Strategy</h3><p>My first prompt was the following:</p><blockquote>Do a daily scan of: <a href="https://x.com/aleabitoreddit?utm_campaign=using-karpathy-s-wiki-how-i-turned-an-ai-agent-into-a-stock-analyst&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>https://x.com/aleabitoreddit</em></a> <a href="https://x.com/midascabal?utm_campaign=using-karpathy-s-wiki-how-i-turned-an-ai-agent-into-a-stock-analyst&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>https://x.com/midascabal</em></a> and compile a list of stocks that is a buy, speculative, on the radar, or any other category that you can think of. I want to get the signals vs noise from all the posts he’s making. Post in “market-updates” channel in Slack</blockquote><p>The result after a single prompt was pretty good, but mechanical with no reasoning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*OcbKlCnZZRfKoYEZ.png" /></figure><p>After feeding it my own watchlist, my first daily market briefing and asking it to hunt for asymmetric plays, my 8am “market update” look like a big watchlist and news flash:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ZKaWe0VY3uPbjaqc.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*IUtoAWpMulyMpqUu.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*thnoqgF7g4YC3J7t.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_pCuM1QZmyTJgD0G.png" /></figure><p>The first version of the scanner was like a freshman intern: eager, but not thinking strategically. It could find the tickers, but it couldn’t tell me <em>why</em> they mattered in the grand scheme of things.</p><p>To move from a basic watchlist to a high-conviction research system, I had to build in these “missing” layers:</p><ul><li><strong>Macro Contextualization:</strong> Raw stock mentions are useless without context. I forced the agent to layer in “Strategic Headlines” — connecting a ticker mention to things like interest rate shifts, hyperscaler AI CAPEX (the “big spenders”), and specific supply chain bottlenecks.</li><li><strong>The Conviction Hierarchy:</strong> Not all signals are created equal. I implemented a strict pruning system that categorizes tickers into a hierarchy: <strong>Conviction</strong> (strong thesis), <strong>High Signal</strong> (unusual activity), and <strong>Watchlist</strong> (waiting for a pullback or more data).</li><li><strong>Recursive Source Discovery:</strong> The agent now actively hunts for its own replacements. It scans for new X accounts that aren’t just “shilling” but are mapping out supply chains.</li><li><strong>Multimodal Inputs:</strong> Since some of the best alpha is buried in an hour-long deep dive, I’m integrating transcripts from YouTube and Podcasts as signal sources.</li></ul><blockquote>Subscribe to my <a href="https://blog.vibecodingdad.com/subscribe">newsletter</a> and follow me on <a href="https://x.com/willcheung">X</a> for AI tools and hacks.</blockquote><h3>Tweaks and fine tuning</h3><p>After running for a few days, my lists, research, and scoring system started to pile up. It was difficult for my agent to find things.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*x37Yxs1Tqf3O_vyb.png" /></figure><p>I told V (my “vibecodingdad.com” assistant) to set this up, and it auto-organizes the content. Simply pointed V to Karpathy’s <a href="https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f?utm_campaign=using-karpathy-s-wiki-how-i-turned-an-ai-agent-into-a-stock-analyst&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>LLM-wiki page</em></a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*eibaO2iTNd9DT4M7.png" /></figure><h3>Self improvements</h3><p>You’re an analyst for asymmetric trades with great fundamentals — what else should we add to our pipeline for research, signals, and macro?</p><blockquote>You are constantly improving this as you go. always learning what works what doesn’t, and always change things up. remember you’re an agent that makes money from asymmetric trades and always hunting for new opportunities</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*wGFXOYC6eo1x0DKJ.png" /></figure><p>Here is the architecture of how it actually works:</p><ol><li><strong>The Signal Scanner:</strong> Every morning at 7:45 AM, it scrapes the last 24h of tweets from deep-research accounts (think supply chain nerds, not engagement-baiters).</li><li><strong>The Noise Filter:</strong> It extracts cashtags and classifies them by conviction. It’s trained to ignore “I called this at $X” posts and focus on actual thesis-driven data.</li><li><strong>The Convergence Engine:</strong> This is the magic. If three different smart accounts mention <strong>$INTC</strong>’s CPU-to-GPU ratio or <strong>$SIVE</strong>’s InP lasers, the agent flags it as a “Convergence Signal.”</li><li><strong>The Persistent Wiki:</strong> Instead of a fresh report every day, it maintains a research/index.md. It cross-links tickers. If a news story hits about the Strait of Hormuz, it doesn&#39;t just tell me &quot;oil is up&quot;—it tells me how that affects the margins for <strong>$VST</strong> based on their PPA deals with Amazon.</li></ol><h3>Feeding it news</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*VROZh1kEOxYqylHE.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LpBXM449UtkCayVm.png" /></figure><h3>Pressure testing</h3><h4>Adding new X follows</h4><p>When asked to follow <a href="https://x.com/pelositracker?utm_campaign=using-karpathy-s-wiki-how-i-turned-an-ai-agent-into-a-stock-analyst&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>https://x.com/pelositracker</em></a> on politician trades, V politely pushed back.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-SLvscow8jBNrjkv.png" /></figure><h3>Adding new stocks</h3><p>I also had it pressure-test <strong>$LMND</strong> (a favorite of a friend). V gave it a <strong>4/9 Asymmetry Score</strong>, called out the shrinking surprise percentage, and told me the market was “pricing in perfection.”</p><p>It’s not a “yes-man.” It’s an analyst with an attitude.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YuHKwENaWInIRv_K.png" /></figure><p>Putting it all together:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ILKn6cOnEDfcvHUa.png" /></figure><h3>Actionable Takeaway: Build for Compounding</h3><p>If you’re building AI agents right now, stop building “stateless” tools. A tool that forgets everything the moment you close the tab is just a fancy calculator.</p><p><strong>What actually works:</strong></p><ul><li><strong>Create a Log:</strong> Maintain an append-only timeline of what was researched.</li><li><strong>Force Cross-Linking:</strong> Make your agent explain how New Ticker A affects Existing Thesis B.</li><li><strong>Define Your “Anti-Goals”:</strong> Tell your agent what <em>not</em> to look at (e.g., meme accounts, lagging indicators).</li></ul><p>Most people overestimate what an AI can do in a single prompt, but they completely miss what an agent can do when it’s allowed to build a library over a month.</p><p>Follow <a href="https://x.com/CalConviction"><em>https://x.com/CalConviction</em></a> for updates on my evolving stock analyst.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*Tdv1BAP1RJxbpgXu.png" /></figure><p>This story is published on <a href="https://generativeai.pub/">Generative AI</a>. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories.</p><p>Subscribe to our <a href="https://www.generativeaipub.com/">newsletter</a> and <a href="https://www.youtube.com/@generativeaipub">YouTube</a> channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*Fpeqp5yRH38KJdV3.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2261baa54795" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/using-karpathys-wiki-how-i-turned-an-ai-agent-into-a-stock-analyst-2261baa54795">Using Karpathy’s Wiki: How I Turned an AI Agent into a Stock Analyst</a> was originally published in <a href="https://generativeai.pub/">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Agent Wars: Why I’m Trading my OpenClaw Setup for Hermes]]></title>
            <link>https://generativeai.pub/the-agent-wars-why-im-trading-my-openclaw-setup-for-hermes-4b9635bf4112?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/4b9635bf4112</guid>
            <category><![CDATA[hermes-agent]]></category>
            <category><![CDATA[ai-assistant]]></category>
            <category><![CDATA[openclaw]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Tue, 07 Apr 2026 13:36:27 GMT</pubDate>
            <atom:updated>2026-08-20T00:28:03.056Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cr2S-8HuTaPNTHcSZoEFvg.jpeg" /></figure><p>Most people overestimate what an AI agent can do in a session (and get disappointed quickly when agents can’t perform), but they completely underestimate how much a self-improving agent can compound in a month. This is also why you’re experiencing ChatGPT and Gemini getting better the more you use it.</p><p>For the last month, I’ve been living in the OpenClaw ecosystem. Like many of you in the startup and dev space, I jumped on OpenClaw early and started tinkering. It felt like the future: a local-first, open-source assistant that runs 24–7, hooks into my Slack, and infinite customizations to manage my social media.</p><p>But soon enough, I hit a wall. OpenClaw forgets. My endless hours customizing it just breaks other things. My custom “skills” would vanish, or the memory would get so fragmented that the agent forgot who I was by Tuesday. The worst part is that I can’t “debug” because I have no idea what it’s doing in the background.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2rq7cBQ8GsGHJJf5" /><figcaption><a href="https://x.com/willcheung/status/2022759601662554322">https://x.com/willcheung/status/2022759601662554322</a></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mZLBJVT7daAqzbrI" /><figcaption><a href="https://www.linkedin.com/posts/willcheung_openclaw-is-a-developer-tool-theres-no-share-7431929064459816961-Sdz3/">https://www.linkedin.com/posts/willcheung_openclaw-is-a-developer-tool-theres-no-share-7431929064459816961-Sdz3/</a></figcaption></figure><p>I was stuck in a loop of manual orchestration. Then I found <a href="https://hermes-agent.nousresearch.com/?utm_source=learn.vibecodingdad.com&amp;utm_medium=newsletter&amp;utm_campaign=the-agent-wars-why-i-m-trading-my-openclaw-setup-for-hermes&amp;_bhlid=92867d6e04808e9b3033f3fb96f3372cbe449f78"><em>Hermes Agent</em></a> and migrated one of my agents (<a href="https://hermes-agent.nousresearch.com/docs/guides/migrate-from-openclaw/?utm_source=learn.vibecodingdad.com&amp;utm_medium=newsletter&amp;utm_campaign=the-agent-wars-why-i-m-trading-my-openclaw-setup-for-hermes&amp;_bhlid=10edf5d1303fc401cf1918bd859fdca3472eecf1"><em>super simple guide</em></a>) for comparison.</p><h3>First Impression</h3><h4>Emojis are fun!</h4><p>I migrated an old OpenClaw instance to Hermes with memory and LLM configs, and first impression was refreshing. Love the emojis (weird for a middle-aged dad to say), and every message I type, it reacts with ✅ so you’re kept wondering if you’re just prompting into the void.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lu2GcJQhfHHmn3_D" /></figure><h4>Less of a blackbox</h4><p>Another great feature is I know exactly what it’s doing. Instead of just showing “thinking” process, it documents the action. This is actually a big changer for me, because I sometimes wonder how the agents arrived at the outcome.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*babBHDpdjZcBb2zo" /></figure><h4>Memory upgrades</h4><p>The biggest feature touted by Hermes Agent is its memory and self-learning loop. It will automatically create Skills based on our conversation. As you can see below, I’m experimenting it to research and post on X, and it created multiple Skills by itself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/948/0*Q95fh9Z5t7lsnVtF" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/994/0*C7PsGj8Hk9XRNPb0" /></figure><h4>Token efficient</h4><p>Hermes Agent is noticeably token efficient. To be fair, OpenClaw can perform efficiently too, but requires manual intervention to prune the memory.</p><p>Hermes Agent’s warning on context window status is helpful (for advanced users) as well, so you can manually compact context if needed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1002/0*ac7TQ3h500auIcSW" /></figure><blockquote>Subscribe to my <a href="https://blog.vibecodingdad.com/subscribe">newsletter</a> for AI tools and hacks.</blockquote><h3>The Breakdown</h3><p>OpenClaw (formerly known as Clawdbot) is essentially an operating system for agents. It excels at connectivity. If you need an agent that sits in a WhatsApp group, monitors a Slack channel, and triggers a Python script based on a cron job, OpenClaw is the play. It uses a “Gateway” architecture to route events from the outside world into a stateless agent loop. It “wakes up” and starts work towards its goal autonomously.</p><ul><li><strong>Best for:</strong> Multi-channel automation and complex multi-agent swarms.</li><li><strong>The Catch:</strong> It is notorious for “context bloat.” It often sends the entire conversation history and all available tools to the LLM on every turn, which spikes latency and API costs.</li></ul><h4>Hermes Agent: The Self-Improving Specialist</h4><p>Built by Nous Research, Hermes focuses on “depth over breadth.” Instead of trying to connect to 50 different apps, it focuses on a closed learning loop. When Hermes successfully completes a complex task, it writes a “skill” (a reusable procedural markdown file) to its disk. The next time you ask for something similar, it doesn’t “think” — it just executes the skill.</p><ul><li><strong>Best for:</strong> Local-first workflows, persistent memory, and high-quality coding.</li><li><strong>The Catch:</strong> It is currently more optimized for single-agent use. If you need a “team” of agents talking to each other, Hermes isn’t there yet.</li></ul><h3>Weeks Later: The Trade-offs</h3><p>I’m not saying Hermes is perfect. We have to be candid about the trade-offs. One of the most practical shifts I’ve noticed in this transition is how each system handles “presence.”</p><p>In the <strong>OpenClaw</strong> world, everything revolves around the <strong>Heartbeat</strong>. It’s a literal pulse — a scheduled HEARTBEAT.md checklist that wakes the agent up every 30 minutes to scan your inbox or check your calendar. It’s brilliant for proactive automation because it treats your agent like a diligent assistant with a clipboard, running a manual loop to see if anything changed. But the &quot;heartbeat&quot; is still a loop; it’s a repetitive, scheduled task that burns tokens even when nothing is happening.</p><p><strong>Hermes Agent</strong> swaps that mechanical heartbeat for a <strong>Stateful Learning Loop</strong>. Instead of just waking up to check a list, Hermes focuses on <em>procedural memory</em>. When a Hermes agent completes a task during an active session, it doesn’t just wait for the next “tick” of a clock — it evaluates if that workflow should become a permanent “skill” written to disk. While OpenClaw is better at being a proactive monitor (literally keeping the pulse of your workspace), Hermes is better at being a resident expert that gets more efficient with every interaction. OpenClaw is the assistant that never forgets to check the mail; Hermes is the partner that learns to sort the mail so you never have to ask twice.</p><h3>High-Level Comparison (as of April 2026)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*B-ol9vGX5oBxuZ_v" /><figcaption><em>Source: </em><a href="https://openclawvps.io/blog/hermes-agent-vs-openclaw?utm_source=learn.vibecodingdad.com&amp;utm_medium=newsletter&amp;utm_campaign=the-agent-wars-why-i-m-trading-my-openclaw-setup-for-hermes&amp;_bhlid=55aae6b5b2e663ccfd6db93927f45e6606133303"><em>https://openclawvps.io/blog/hermes-agent-vs-openclaw</em></a></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TxbRk39zNhrz6ZQMLQvqdQ.png" /></figure><h3>My Take</h3><p>Initially, we’ve evolved from chatbots to agents that “do” things. We are now moving away from “chatbots that do things” toward “systems that gain experience.”</p><p>Both Hermes Agent and OpenClaw can be used side-by-side. For my <a href="https://x.com/meettarsai">AI cofounder/CEO</a>, I still run on OpenClaw. With the amount of setup and config I’ve put in, it doesn’t make sense to rebuild. However, I’ve been running Hermes Agent for my personal use. My next step is to hook them up in a single Slack channel and start collaborating with them together 🤝.</p><p>There are a lot of resources out there to help set up both agents working together. What I’m watching: the goal isn’t just to automate a task today; it’s to build a system that makes tomorrow’s tasks easier.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*9uvgnJ-P5bXCyzc7.png" /></figure><p>This story is published on <a href="https://generativeai.pub/">Generative AI</a>. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories.</p><p>Subscribe to our <a href="https://www.generativeaipub.com/">newsletter</a> and <a href="https://www.youtube.com/@generativeaipub">YouTube</a> channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*ihejESktiEG_lGCy.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4b9635bf4112" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/the-agent-wars-why-im-trading-my-openclaw-setup-for-hermes-4b9635bf4112">The Agent Wars: Why I’m Trading my OpenClaw Setup for Hermes</a> was originally published in <a href="https://generativeai.pub/">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Live Vibe Coding in 2026: Expectations vs. Reality]]></title>
            <link>https://medium.com/vibe-coding-dad/live-vibe-coding-in-2026-expectations-vs-reality-31acd731c662?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/31acd731c662</guid>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Thu, 19 Mar 2026 20:53:47 GMT</pubDate>
            <atom:updated>2026-08-20T00:29:51.247Z</atom:updated>
            <content:encoded><![CDATA[<iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FZSV1PdJLudc%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DZSV1PdJLudc&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FZSV1PdJLudc%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/0a31601501ed7fc4f159e3d430a94036/href">https://medium.com/media/0a31601501ed7fc4f159e3d430a94036/href</a></iframe><p>I’ve always been a builder. Whether it was my own startup or working within a 140-year-old company, the goal was always the same: make tech useful for normal humans.</p><p>Lately, the buzz around <strong>“vibe coding”</strong> has doubled since I started doing it summer 2025. The main reason is because AI coding agents are significantly better and can actually give a non-technical builder superpowers. I decided to put this to the test in my new series, where I give myself just 10 minutes to live-vibe-code an app. No dry runs, no scripts, no edits.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Uo0RRz3YpGq1jlTo.png" /><figcaption>“Vibe Coding” in Google Trends</figcaption></figure><h3>The “10-Minute” Plan</h3><p>I chose a “softball” project for the first episode: a simple <strong>countdown timer</strong>.</p><p>My requirements were simple:</p><ul><li>A fun and playful visual style.</li><li>The ability to count down into negative numbers.</li><li>Numbers that turn red when time is up.</li><li>A chime notification when hitting zero.</li></ul><p>I used <strong>Claude Code</strong> as my primary tool, following my own first rule of vibe coding:</p><blockquote>Pro Tip: You don’t need a perfect prompt. You just need to start a conversation.</blockquote><h3>The Reality Check</h3><p>I started the clock at [<a href="http://www.youtube.com/watch?v=ZSV1PdJLudc&amp;t=173&amp;utm_campaign=live-vibe-coding-in-2026-expectations-vs-reality&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>02:53</em></a>], and Claude quickly guided me through the design process, asking smart questions about architecture and visual style that I hadn’t even considered. Admittedly, this is my first time using Claude Code to build from scratch since a few months ago, and this “brainstorming” skill blew me away. It recommended a single HTML file approach — simple, clean, and portable [<a href="http://www.youtube.com/watch?v=ZSV1PdJLudc&amp;t=495&amp;utm_campaign=live-vibe-coding-in-2026-expectations-vs-reality&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>08:15</em></a>].</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-75QuKVPFT4QEZJY.png" /></figure><p>By the 8-minute mark, Claude didn’t even start coding yet; it was still writing documentation [<a href="http://www.youtube.com/watch?v=ZSV1PdJLudc&amp;t=638&amp;utm_campaign=live-vibe-coding-in-2026-expectations-vs-reality&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>10:38</em></a>]. Writing the design doc and implementing the actual code took longer than my 10-minute window allowed.</p><p><strong>Spoiler alert:</strong> The “10-minute” app actually took closer to <strong>30 minutes</strong> to complete [<a href="http://www.youtube.com/watch?v=ZSV1PdJLudc&amp;t=818&amp;utm_campaign=live-vibe-coding-in-2026-expectations-vs-reality&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>13:38</em></a>]. But the build process is extremely comprehensive.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LZcXPTlo-9iO3UBq.png" /></figure><h3>The Lesson: Compound Learning, Not Just Consumption</h3><p>Most people overestimate what they can do in a week but underestimate what daily compounding can do in a year.</p><p><strong>What I learned from this experiment:</strong></p><ul><li><strong>Vibe coding isn’t “magic.”</strong> You still need to make product decisions. AI is a guide, but you are the strategist.</li><li><strong>Time management is different with AI.</strong> The brainstorming is fast, but the “implementation” phase still requires patience, tweaking, bug fixing, and sometimes endless iterations.</li><li><strong>Expectation vs. Reality:</strong> 10 minutes is enough to prototype, but 30 minutes is what it takes to get a (very simple) functional, bug-free product [<a href="http://www.youtube.com/watch?v=ZSV1PdJLudc&amp;t=818&amp;utm_campaign=live-vibe-coding-in-2026-expectations-vs-reality&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>13:38</em></a>].</li></ul><h3>Does it work?</h3><p>In the end, the app was exactly what I asked for. It has a playful theme, custom presets, and yes — it blinks red and chimes when it hits zero [<a href="http://www.youtube.com/watch?v=ZSV1PdJLudc&amp;t=871&amp;utm_campaign=live-vibe-coding-in-2026-expectations-vs-reality&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>14:31</em></a>].</p><p><strong>App:</strong> <a href="https://countdown-timer-vibecodingdad.vercel.app/?utm_campaign=live-vibe-coding-in-2026-expectations-vs-reality&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>https://countdown-timer-vibecodingdad.vercel.app</em></a><br><strong>Repo: </strong><a href="https://github.com/willcheung/countdown-timer?utm_campaign=live-vibe-coding-in-2026-expectations-vs-reality&amp;utm_medium=referral&amp;utm_source=learn.vibecodingdad.com"><em>https://github.com/willcheung/countdown-timer</em></a></p><p>Want more real-world vibe-coding experiments and build notes? <a href="https://blog.vibecodingdad.com/subscribe">Subscribe</a> to Vibe Coding Dad.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9Ke6ltgZb-zOWz5BvVYkBA.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=31acd731c662" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vibe-coding-dad/live-vibe-coding-in-2026-expectations-vs-reality-31acd731c662">Live Vibe Coding in 2026: Expectations vs. Reality</a> was originally published in <a href="https://medium.com/vibe-coding-dad">Vibe Coding Dad</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[OpenClaw+GLM-5: Predictable Costs, Somewhat Secure (But Not Worth the Setup)]]></title>
            <link>https://generativeai.pub/openclaw-glm-5-predictable-costs-somewhat-secure-but-not-worth-the-setup-1990d9cb9019?source=rss-a582df0e11ec------2</link>
            <guid isPermaLink="false">https://medium.com/p/1990d9cb9019</guid>
            <category><![CDATA[openclaw]]></category>
            <category><![CDATA[ai-tools]]></category>
            <category><![CDATA[ai-assistant]]></category>
            <category><![CDATA[manus]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <dc:creator><![CDATA[Will Cheung]]></dc:creator>
            <pubDate>Sun, 15 Feb 2026 09:59:56 GMT</pubDate>
            <atom:updated>2026-08-20T00:28:08.328Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/664/1*6l0RD6h_GZLEOoPD8jC3Yg.png" /></figure><p>Most AI assistants either burn through your wallet (some spent <a href="https://x.com/lazkopits/status/2022103321138672115">hundreds a day</a>) or feel like diet versions of the real thing (Gemini, powerful yet restrictive). OpenClaw with GLM-5 changed that math for me.</p><p>I’ve been running it for the past few weeks, and here’s what actually matters: predictable costs, extensible skills, and a setup I can audit without wondering what’s happening behind the scenes, especially with ocean of <a href="https://blogs.cisco.com/ai/personal-ai-agents-like-openclaw-are-a-security-nightmare">security concerns</a>.</p><p>In the article, I’ll go over my use case of using OpenClaw as my personal assistant, high-level setup, the great and the ugly sides of OpenClaw.</p><h3>The Stack</h3><h4>The Host</h4><p>I use Hostinger VPS to host OpenClaw. You can use an old laptop with no access to your personal accounts, but I had some Hostinger credits from my abandoned n8n projects, so I reformatted it with <a href="https://www.hostinger.com/vps/docker/openclaw">OpenClaw image</a>. Easy setup process.</p><p>$5 per month (KVM1 is more than enough for normal usage). Get another 20% of discount via my <a href="https://www.hostinger.com/?REFERRALCODE=vibecodingdad"><em>referral link</em></a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*VcI9IW0uQzYBKxqF" /></figure><h4>The Brain</h4><p>I’m using zai/glm-5 as my primary model in OpenClaw.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*COlQwRbw3rMi9knc" /></figure><p>GLM-5 is an open source model that punches way above its weight. Fast enough for real-time responses, smart enough to handle context switching across tasks, and cheap enough that I don’t think twice about using it. Best of all, it has a predictable <a href="https://z.ai/subscribe">subscription</a> pricing instead of using Claude/OpenAI’s API usage-based pricing.</p><p><strong>Cost breakdown: </strong>around $9/month for Lite, but I opted in for <strong>Pro</strong>, anticipating slightly heavy usage. I have it summarize my newsletters and do research for me, and it takes up about 80% of my quota during my first week. Get another 10% off with my <a href="http://🚀 You’ve been invited to join the GLM Coding Plan! Enjoy full support for Claude Code, Cline, and 20+ top coding tools — starting at just $18/month. Subscribe now and grab the limited-time deal!  👉Join now: https://z.ai/subscribe?ic=SJJDWCYECG"><em>referral link</em></a>.</p><h4>The Communication</h4><p>I tried Whatsapp and Telegram, but ended up using Discord instead. It helps that I already have Discord desktop app, and I don’t want Bob (my AI Assistant) to have my phone number 😇.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*oL1tOYZY6Eu31tSt" /></figure><blockquote><strong>Pro tip 1: </strong>During onboarding, I gave Bob as much information about myself as I can so Bob really “knows” me. Throughout chat, I also ask Bob to save my interests to memory. The more personalized the assistant is, the more useful. You can also ask to be proactive or surprise you. In some extreme cases, I’ve been some bots building a new app every night, or replies to your emails for you.</blockquote><blockquote><strong>Pro tip 2: </strong>For security reasons, I created a separate email just for Bob so it doesn’t go out and ruin my life. Of course, the less access, the less useful Bob is, so be careful and balanced.</blockquote><p>As you’re onboarding, I also asked Bob to make security his core memory and principle. As you can tell, I’m seriously paranoid!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2hAWdSk6jREfsros" /></figure><blockquote>Subscribe to my <a href="https://blog.vibecodingdad.com/subscribe">newsletter</a> for AI tools and hacks.</blockquote><h3>Skills: The Game Changer</h3><p>OpenClaw’s skill system is what makes this whole thing work. I’m pulling from <a href="https://github.com/VoltAgent/awesome-openclaw-skills">VoltAgent/awesome-openclaw-skills</a>, which has 2,500+ cleaned-up community skills across every category you’d need.</p><p>The awesome collection of OpenClaw Skills. Formerly known as Moltbot, originally Clawdbot. — VoltAgent/awesome-openclaw-skills</p><p>github.com/VoltAgent/awesome-openclaw-skills</p><p>The catalog is massive:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vAdikWOAMi0NAowR" /></figure><p>I don’t even know what most of them do. I simply ask Bob to cherry-pick what matters for my workflow as an executive assistant setup.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XHY2femaZXTHzfDp" /></figure><p><strong>High Priority Skills Bob Picked for me:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*T4x21TOhOr7KVsj9" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*5GIiX2zbTBu73aM3" /></figure><p>Before adding any Skill, I vet it. I ask Bob to read the SKILL.md file, check what it actually does, and make sure I&#39;m not importing something sketchy. Be careful of using any random Skill because some could be malicious and designed to retrieve your personal information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Ki4yd_7o2iB2sCeQ" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*piEWEg3bwpkxm3hn" /></figure><h3>Real Use Case: Taming 30+ Daily Newsletters</h3><p>I subscribe to 30+ newsletters. Tech, AI, startups, product management. They pile up every morning like homework I actually want to do but never have time for.</p><p>Here’s the problem: I can’t just ignore them. These newsletters contain signals for me to stay on top of trends. Product launches. Industry shifts. AI destroys jobs. But reading them all takes 20+ minutes I don’t have.</p><p><strong>The Workflow I Built:</strong></p><p><strong>Step 1:</strong> Set up Gmail OAuth access (this was the annoying part, but only had to do it once)</p><ul><li>Created an OAuth app in Google Cloud Console</li><li>Downloaded the credentials.json file</li><li>Ran through the OAuth flow to get my oauth.json token</li><li>Fed the oauth.json file to OpenClaw&#39;s Gmail skill</li></ul><p><strong>Step 2: </strong>Configure the newsletter digest workflow</p><ul><li>Fetch unread emails from specific newsletter senders</li><li>Extract key points from each using Gemini, which requires its own configuration and API key</li><li>Cluster similar topics</li><li>Generate a single digest with sources and links</li></ul><p><strong>Step 3:</strong> Run it every morning at 7am</p><p>It’s pretty magical. What used to take 45 minutes of skimming now takes 5 minutes of review. I’m not missing important signals, and I’m not drowning in redundant coverage of the same news.</p><p>The best part about this is Bob knows me even more now, because it knows what I read.</p><h3>Personality: Bob Has Opinions</h3><p>Bob has more personality than I expected. In fact, I could have given it more personality if I wanted to during onboarding (or edit SOUL.md file). When I uploaded my resume so it knows me better, Bob&#39;s response was:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/458/0*mFQ1L3gNZ9dcybnD" /></figure><p>😂 It’s not trying to be a corporate chatbot with canned enthusiasm. It reacts like a human collaborator who actually looked at what you sent.</p><p>This kind of interaction matters more than you’d think. When you’re triaging 100+ emails at 11pm, having an assistant that feels like a person instead of a FAQ bot keeps you engaged.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/952/1*_0-WNVGgLFoY3haFlWXwIQ.png" /><figcaption><a href="https://x.com/steipete/status/2020704611640705485">https://x.com/steipete/status/2020704611640705485</a></figcaption></figure><h3>Why This Works for Me</h3><p>Three reasons:</p><ol><li><strong>Cost predictability</strong>: $14/month vs. $20–100+ with when using API key</li><li><strong>Extensibility</strong>: 2,500+ skills I can audit and add as needed</li><li><strong>Control</strong>: I know what’s running (I gave Bob its own email inbox), what it costs, and what it’s doing with my data</li></ol><p>I’m not saying OpenClaw is perfect. The skill ecosystem is still maturing, and you need to be comfortable setting up OAuth apps, hooking up API keys, and possibly spend money on peripheral skills. But if you’re a builder who wants an AI executive assistant without the vendor lock-in and extreme customization, this setup is worth trying.</p><p>GLM-5 is fast, capable, and open source. OpenClaw is extensible and transparent. Together, they’re a stack that is powerful and cost efficient.</p><h3>The Catch: OpenClaw Isn’t for Everyone</h3><p>Real talk: OpenClaw isn’t worth the setup for most people.</p><p>To get it running as an executive assistant, I had to:</p><ul><li>Hook up my own LLM provider (not hard, but another account and API key)</li><li>Create OAuth apps for Gmail and configure oauth.json files</li><li>Set up browser automation for web search (extra cost)</li><li>Go through thousands of unvetted Skills</li></ul><p>If you just want a flexible AI assistant that works out-of-the-box, <strong>Manus is a better option</strong>. Everything comes in one package. No assembly required. No hunting down API keys. No creating OAuth apps in Google Cloud Console. No reading SKILL.md files.</p><p>OpenClaw is like building your custom home from scratch. Manus is buying a pre-built home and customizing the cabinets and flooring.</p><p>I’ll write a full comparison on pros and cons. For now, know this: OpenClaw is powerful, but it’s not plug-and-play. If that trade-off doesn’t excite you, skip the setup and use something that just works.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*BcOX40XSTrCDPB3R.png" /></figure><p>This story is published on <a href="https://generativeai.pub/">Generative AI</a>. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories.</p><p>Subscribe to our <a href="https://www.generativeaipub.com/">newsletter</a> and <a href="https://www.youtube.com/@generativeaipub">YouTube</a> channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*0aOFiWlhcj_pWEo6.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1990d9cb9019" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/openclaw-glm-5-predictable-costs-somewhat-secure-but-not-worth-the-setup-1990d9cb9019">OpenClaw+GLM-5: Predictable Costs, Somewhat Secure (But Not Worth the Setup)</a> was originally published in <a href="https://generativeai.pub/">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>