<?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 Rajat Dangi 🛠️ on Medium]]></title>
        <description><![CDATA[Stories by Rajat Dangi 🛠️ on Medium]]></description>
        <link>https://medium.com/@rajatkumardangi?source=rss-4fff48917921------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*sC7cSbxWSYfiXjYGKvM5HQ.jpeg</url>
            <title>Stories by Rajat Dangi 🛠️ on Medium</title>
            <link>https://medium.com/@rajatkumardangi?source=rss-4fff48917921------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 06:22:13 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rajatkumardangi/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[We built our companies on AI. Now we’re doing it for other founders.]]></title>
            <link>https://rajatkumardangi.medium.com/we-built-our-companies-on-ai-now-were-doing-it-for-other-founders-43a2e46862ae?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/43a2e46862ae</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai-services]]></category>
            <category><![CDATA[ai-automation]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Mon, 02 Nov 2026 20:15:30 GMT</pubDate>
            <atom:updated>2026-11-02T20:19:26.736Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>Introducing </em><a href="https://xyzlab.in/"><em>XYZ Lab</em></a><em>: a two-person team that puts AI to work inside businesses doing $2M to $20M a year.</em></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*11FEKGyXxcOumLsraCpsAQ.png" /></figure><p>I run Shhots AI alone. AI runs analytics, revenue tracking, SEO, content, outreach and support, and I check in once a day or once a week.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wqfW2BO3_V6OwNqHmNNyzg.png" /><figcaption><a href="https://shhots.ai/">Shhots AI</a></figcaption></figure><p>My friend <a href="https://linkedin.com/in/ansarimofid">Mofid</a> runs <a href="https://mockey.ai/">Mockey</a> with a small team. In one year they created 30,000+ AI mockups, a library that took tratditional mockup companies 5 to 10 years to build.</p><p>So we started <a href="https://xyzlab.in/"><strong>XYZ Lab</strong></a> to do the same inside other businesses. We find the work AI can do, build the systems that do it, and stay on to run them. We’re starting with three clients.</p><h3>How I got here</h3><p>I started using AI in 2023 to write blog posts. SEO is where I have the most experience, so it was the obvious first job.</p><p>Then the jobs got bigger. Market research. Editing my writing. Brainstorming. Writing PRDs for product ideas.</p><p>Then AI started finding things for me. Two jobs I still remember:</p><ul><li>Go through 50,000 keywords, keep the ones that matter for my product, and plan the content.</li><li>Read an ebook and a set of subreddits, hunt for one specific insight, and propose a product that solves it.</li></ul><p>When <a href="https://metaschool.so/">Metaschool</a> pivoted into <a href="https://okara.ai/">Okara</a>, I went deep on open source models, voice AI, generative AI, privacy and security. Okara was building an AI agent builder, so I saw thousands of use cases up close.</p><p>That’s where agents took over my thinking. I watched teams build loops where AI finds the work, does it, reports on it, learns from the result and does it better next time. No human needed in between.</p><p>After Okara I spent two months looking for the next thing to build. I briefly joined <a href="https://hunto.ai/">Hunto AI</a>, a friend’s cybersecurity startup, and took it through the <a href="https://www.antler.co/location/singapore">Antler Singapore AI Residency</a>. More use cases, and wilder ones.</p><h3>What running a company on AI looks like</h3><p>Shhots AI makes ad creative with AI: UGC video ads, product films, photoshoots and image ads. I run it as a solo founder.</p><p>I have a CS degree but never wrote code for a living. I’ve always been the product and growth person. AI is why someone like me can build and run a software company alone.</p><p>Here’s what runs on its own today:</p><ul><li>Product analytics and revenue tracking</li><li>Revenue recovery</li><li>SEO, and showing up in AI answers like ChatGPT</li><li>Product and app development</li><li>Content creation and posting</li><li>Hunting for new opportunities</li><li>B2B outreach and client handling</li><li>Customer support</li></ul><p>Every part of the business has AI behind it. My job is to check the reports and make the calls.</p><h3>What Mofid built at Mockey</h3><p>Mofid is the design and product half of XYZ Lab. He founded Mockey.ai and takes products from first sketch to launch.</p><p>His team is small. In one year they created 30,000+ AI mockups. At the same time they automated SEO, kept shipping product updates and launched an iOS app.</p><p>Mockey is our first XYZ Lab case study:</p><ul><li><a href="https://xyzlab.in/recipes/mockey-back-office/">328 blog posts moved off WordPress onto a system that publishes by itself</a></li><li>14 new posts published in October 2026</li><li>70% of site code changes co-written by Claude (Sep to Oct 2026)</li></ul><p>One developer and one marketer now review the work instead of doing it.</p><h3>Why we started XYZ Lab</h3><p>Mofid and I have built products side by side for 8+ years. <a href="https://www.expresscomputer.in/news/1-mn-usd-invested-by-anand-mahindra-in-indian-startup-hapramp/57711/">Hapramp raised a $1M seed and reached 500K+ installs</a> (2019 to 2021). <a href="https://blinkstore.in/">Blinkstore</a> powered 100K+ merch stores (2022 to 2023). We also ran side projects like <a href="http://letterflix.com/">Letterflix</a>, under a small studio called theXYZ.</p><p>We kept comparing notes on how much of our work AI now does. The same pattern showed up at Okara, at Antler and in our own companies. The people putting AI to work spend less, waste less time and make more money than their peers.</p><p>Getting access to the tools takes a credit card. Building the agents, fitting them into the business and stepping out of the weekly loop takes skill and time. Most owners have neither to spare.</p><p>There are millions of hardworking business owners doing the same work every week. If AI takes that work, they get time back for their health and families. They spend less on tech, keep better margins and serve more customers.</p><p>That’s the job we want.</p><h3>What <a href="https://xyzlab.in/">XYZ Lab</a> does, and who it’s for</h3><p>XYZ Lab is the two of us. The two founders on the first call are the two who do the work. There’s no account manager and no juniors.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uqXHV34M3p4ROPnZGnEuFQ.png" /><figcaption><a href="https://xyzlab.in/">https://xyzlab.in/</a></figcaption></figure><h4>We work in three areas:</h4><ol><li><strong>Product:</strong> planning, analytics, AI features, internal tools and apps, and code that ships faster with AI writing and a person reviewing.</li><li><strong>Growth:</strong> content, showing up in Google and AI answers like ChatGPT, ad creative, outbound and lead follow-up.</li><li><strong>Operations:</strong> the repeat admin, such as invoices, reports, hiring, the inbox and documents.</li></ol><h4><strong>It’s a fit if:</strong></h4><ul><li>You do $2M to $20M a year, in any industry</li><li>A founder or senior leader owns the result</li><li>Nobody inside the business owns AI yet, and the same work repeats every week</li></ul><p>It’s not a fit if you’re under $2M (a good tool will serve you better), you want a chatbot demo, or your buying process runs through long enterprise procurement.</p><h3>How we work</h3><ol><li><strong>Diagnose (2 weeks, fixed fee).</strong> We find where AI pays off in your business, and what to leave alone. You get a plan with numbers. The fee counts toward the build if you go ahead within 30 days.</li><li><strong>Build (3 to 6 weeks, fixed scope).</strong> We build the systems in your own accounts and agree the numbers we’ll report on.</li><li><strong>Run (monthly, six-month minimum).</strong> We stay on as your satellite team. We run the systems and send a written report every month. When a better AI model or a new opportunity appears, we propose it and build it.</li></ol><h4>The rules we work by:</h4><ul><li><strong>You own everything.</strong> The systems live in your accounts. If you stop, you keep it all, plus a written guide.</li><li><strong>A person stays in the loop where a mistake is costly:</strong> messages to clients, hiring decisions, anything that moves money.</li><li><strong>Numbers come first.</strong> Every system has numbers agreed before we build. If they don’t move, the report says so.</li><li><strong>The monthly fee costs less than one senior hire.</strong> AI tool costs are included up to an agreed cap. Exact figures come on the first call.</li></ul><h3>Three clients to start</h3><p>We both still run our own companies, so we’re capping XYZ Lab at three clients for now.</p><p>We’re based in India and work with businesses in the US, UK, UAE, Australia and Singapore. Pricing is in US dollars. <a href="https://cal.com/xyzlab-ai/30min">Calls</a> happen in your hours, and everything else happens in writing.</p><h4>To start, <a href="https://cal.com/xyzlab-ai/30min">book a free 30-minute call</a> with both of us at <a href="https://xyzlab.in/">xyzlab.in</a>. If there’s a fit, a one-page agreement follows within 24 hours.</h4><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=43a2e46862ae" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Brand Needs an AI Micro Drama — Here’s why and how?]]></title>
            <link>https://rajatkumardangi.medium.com/your-brand-needs-an-ai-micro-drama-heres-why-and-how-2649d28b1719?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/2649d28b1719</guid>
            <category><![CDATA[ai-microdrama]]></category>
            <category><![CDATA[ai-video-generator]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Thu, 15 Oct 2026 11:41:43 GMT</pubDate>
            <atom:updated>2026-10-15T11:41:43.042Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why AI-generated mini-series could become a new marketing channel for consumer brands, and why we’re exploring it at Shhots AI.</h4><p><em>AI microdramas could be the next big opportunity in brand marketing. And AI is making them more accessible than ever.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Efdz4MmYw1Xkf8_jUi3yFA.png" /><figcaption><a href="https://shhots.ai/blog/ai-microdramas/">AI Micro Drama</a></figcaption></figure><p>What if your brand’s next marketing campaign wasn’t an ad?</p><p>What if it was a five-episode romantic comedy? Or a murder mystery? Or a workplace drama with recurring characters and unexpected plot twists?</p><p>Each episode is 60–90 seconds long. Published on Instagram Reels, TikTok, and YouTube Shorts.</p><h3>I am looking for Brands Interested in creating their own AI Microdramas</h3><blockquote>I’m Rajat, founder of <a href="https://shhots.ai/">Shhots AI</a>.</blockquote><blockquote>We’re exploring <strong>done-for-you AI microdrama production for brands</strong>, and I’d love to speak with founders, marketing teams, and agencies who think episodic storytelling could fit their marketing strategy.</blockquote><blockquote>If you’re planning a product launch, looking for new ways to engage your audience, or simply curious about what an AI-generated series could look like for your brand, I’d be happy to brainstorm some concepts with you.</blockquote><blockquote><a href="https://calendly.com/shhotsai/30min"><strong>Book a 30-minute exploration call with me →</strong></a></blockquote><blockquote>Let’s explore what kind of show your brand could create.</blockquote><iframe src="https://cdn.embedly.com/widgets/media.html?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DRCCQ52COgKM&amp;type=text%2Fhtml&amp;schema=youtube&amp;display_name=YouTube&amp;src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FRCCQ52COgKM%3Ffeature%3Doembed" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/7e8c79d7da2da4aae75ab4bec24a489f/href">https://medium.com/media/7e8c79d7da2da4aae75ab4bec24a489f/href</a></iframe><p><strong>Welcome to the world of AI microdramas.</strong></p><p>I’ve been researching this space recently, and I think there’s an interesting opportunity here for brands, especially those spending thousands of dollars every month producing social media content and advertising creatives.</p><p>Let me explain.</p><h3>First, Microdramas Are Already a Massive Business</h3><p>According to <a href="https://sensortower.com/blog/state-of-short-drama-apps-2026-report">Sensor Tower’s 2026 report</a>:</p><ul><li>Short-drama apps recorded over <strong>850 million downloads in Q1 2026</strong>, growing 140% year over year.</li><li>The category generated approximately <strong>$750 million in quarterly in-app purchase revenue</strong>.</li><li>Platforms like ReelShort, DramaBox, and FreeReels are attracting enormous audiences.</li></ul><p><strong>People are willing to spend serious time, attention, and money consuming short-form fictional entertainment on their phones.</strong></p><p>And brands have started noticing.</p><h3>Crocs, P&amp;G, and Other Brands Are Already Experimenting</h3><p>Earlier this year, Crocs launched a microdrama called <em>Charmed to Meet You</em> on ReelShort.</p><p>The story follows a woman who develops an unexpected connection with her neighbor through Crocs Jibbitz charms.</p><p>It’s a romantic story.</p><p>The product isn’t awkwardly inserted into an unrelated plot. The charms actually help drive the story forward.</p><p>According to <a href="https://www.marketingdive.com/news/brands-content-gold-microdramas-will-they-muck-up-moment/819961/">Marketing Dive</a>, the series attracted nearly 10 million views and led to further microdrama experiments from Crocs.</p><p>Then there’s Procter &amp; Gamble.</p><p>Its personal care brand Native launched <em>The Golden Pear Affair</em>, a 55-episode branded microsoap totaling nearly 80 minutes (55 episodes).</p><p>The company essentially produced its own entertainment property around a personal care brand.</p><p><a href="https://www.nativecos.com/blogs/press/native-launches-feature-length-microsoap-the-golden-pair-affair">Native’s official announcement</a> describes a complete cinematic story, designed specifically for vertical viewing.</p><p>In India, Tata Motors Commercial Vehicles and JK Super Cement have also backed branded microdrama productions.</p><p><a href="https://www.livemint.com/industry/from-ads-to-storylines-how-brands-are-embracing-microdramas-11790142136046.html">Mint reported</a> that JK Super Cement’s campaign generated approximately 93 million views across platforms, according to the company.</p><p>But that’s exactly what makes the next development interesting.</p><h3>Now Imagine Producing These Stories With AI</h3><p>Today, tools like <a href="https://shhots.ai/models/seedance-2-5/">Seedance</a>, <a href="https://shhots.ai/models/kling-3-0/">Kling</a>, <a href="https://shhots.ai/models/veo-3/">Veo</a>, and Runway can generate increasingly realistic cinematic scenes, characters, environments, and dialogue.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F4bG0RxlPZAY%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fshorts%2F4bG0RxlPZAY&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F4bG0RxlPZAY%2Fhq2.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/7c1595bf413125b297c2f7c61494603c/href">https://medium.com/media/7c1595bf413125b297c2f7c61494603c/href</a></iframe><p>AI writing tools can assist with screenplays and storyboards. Voice-generation tools can help with narration and localization.</p><p>You can already combine these capabilities to create short fictional videos without physically filming every scene.</p><p>And the underlying technology is improving rapidly.</p><p>In September 2026, <a href="https://www.reuters.com/business/media-telecom/china-fuels-rush-turn-ai-video-into-an-industry-2026-09-25/">Reuters reported</a> that more than 221,900 AI shows had launched on China’s Douyin in the first half of the year.</p><p>That’s an extraordinary amount of AI-generated entertainment.</p><h3>What Could an AI Microdrama Campaign Look Like for Your Brand?</h3><p>Imagine you’re running a jewelry brand.</p><p>Instead of producing another video showing someone wearing a necklace against a beautiful background, you create a mystery series called <em>The Wrong Ring</em>.</p><p><strong>Episode 1:</strong> A woman receives an expensive engagement ring addressed to someone else.</p><p><strong>Episode 2:</strong> She tries to return it, but discovers the intended recipient disappeared three years ago.</p><p><strong>Episode 3:</strong> A stranger offers her $50,000 to forget about the ring.</p><p><strong>Episode 4:</strong> She discovers the ring contains something far more valuable than a diamond.</p><p><strong>Episode 5:</strong> The mystery is resolved, with an ending that makes the entire story memorable.</p><p>Your jewelry collection becomes part of the narrative.</p><p>Now imagine another example.</p><p>You’re a fashion brand. You create a workplace comedy around three friends trying to survive their first jobs in New York. Every episode brings a new disastrous situation, an entertaining character interaction, and outfits naturally integrated into the story.</p><p>Or perhaps you’re a beauty brand.</p><p>You create a romantic comedy about two rival makeup artists competing for the same opportunity.</p><p>The products are part of their professional world, not the reason every character suddenly stops to deliver a sales pitch.</p><p>These are hypothetical concepts, but you get the idea.</p><p>The goal isn’t to disguise an infomercial as entertainment.</p><p>It’s to create something people would genuinely enjoy watching, with the brand woven naturally into the fictional world.</p><h3>Why I Think This Could Work for Brands</h3><p>There are three reasons I’m particularly interested in this format.</p><h3>1. It gives people a reason to come back</h3><p>Most social media ads are independent pieces of content. You see an ad, watch it, and move on.</p><p>Even if you enjoy it, there’s usually no reason to look for another video from that brand. A microdrama works differently.</p><p>If someone watches episode one and becomes invested in the story, they have a reason to find episode two. Then episode three.</p><p>And eventually, they might start recognizing the brand behind the series.</p><p>Instead of trying to win attention from scratch with every video, you can potentially build recurring attention around characters and stories.</p><p>That doesn’t guarantee sales, but it’s a different relationship with the audience.</p><h3>2. One storyline creates multiple marketing assets</h3><p>Imagine producing a six-episode series.</p><p>You don’t just have six videos.</p><p>You also have opening hooks, individual character moments, dramatic scenes, trailers, alternate endings, and behind-the-scenes content.</p><p>Some of those can become organic social posts.</p><p>Others can be edited into performance ad creatives, subject to the platform’s requirements.</p><p>You can test which scenes attract attention, which characters audiences respond to, and which creative angles generate meaningful traffic.</p><p>One narrative concept can become the foundation for a broader content campaign.</p><h3>3. AI makes experimentation more accessible</h3><p>This is probably the most exciting part for me.</p><p>A brand doesn’t necessarily have to commit to a lengthy traditional shoot before discovering whether an audience likes a particular story.</p><p>With AI-assisted production, it may be possible to test a small pilot, evaluate the response, and improve the concept before producing a larger series.</p><p>Of course, AI production isn’t effortless.</p><p>Character consistency, believable facial expressions, dialogue, continuity, and editing still require substantial work.</p><p>But reducing the logistical burden of production could open opportunities for smaller brands that wouldn’t otherwise consider scripted entertainment.</p><h3>But Would AI Microdramas Actually Generate Revenue?</h3><p>I wouldn’t recommend treating microdramas as a guaranteed replacement for performance advertising.</p><p>I’d treat them as an experiment in branded entertainment with clearly defined commercial goals.</p><p>For example, a brand could test a four-episode series and measure:</p><ul><li><strong>Viewer retention:</strong> How much of each episode do people actually watch?</li><li><strong>Episode progression:</strong> How many viewers move from episode one to episode two?</li><li><strong>Brand engagement:</strong> Are viewers remembering and engaging with the brand?</li><li><strong>Website traffic:</strong> Does the campaign generate visits from the intended audience?</li><li><strong>Commercial impact:</strong> Are viewers taking measurable actions such as signing up, adding products to carts, or purchasing?</li></ul><p>You could also compare the campaign with your existing video creatives.</p><p>Not every brand needs an elaborate fictional universe.</p><p>For some products, a straightforward demonstration will still be the better marketing investment.</p><p>But for categories where emotion, identity, lifestyle, and storytelling influence buying decisions, I think microdramas are worth experimenting with.</p><h3>Shhots AI Powered Microdrama</h3><p>This is something we’re currently exploring at <a href="https://shhots.ai/">Shhots AI</a>.</p><p>We’ve been building AI video creation tools for brands, marketers, and creators.</p><p>And the more I look at how AI filmmaking is evolving, the more interested I become in what’s possible beyond standalone advertisements.</p><p>What if a brand could approach us with its product, audience, and campaign objectives, and we helped turn that into an original AI-generated microdrama series?</p><p>Not just individual video clips.</p><p>A connected story with consistent characters, professional editing, cinematic visuals, and a distribution plan.</p><p>For an initial experiment, I’m thinking of something deliberately manageable:</p><p><strong>A three-to-five-episode branded AI microdrama pilot.</strong></p><p>Each episode could run approximately 45–90 seconds.</p><p>We’d explore the creative concept, develop the scripts and characters, produce the AI-generated scenes, add audio, and assemble the finished episodes.</p><p>We could also create shorter promotional edits and alternative opening hooks for social media.</p><p>The goal would be to test whether the story attracts an audience and whether that attention creates value for the brand.</p><p>It’s still early, and I’m not pretending we’ve solved every creative or production challenge.</p><p>We’re exploring the opportunity, speaking with potential partners, and figuring out what a genuinely useful service should look like.</p><p>Because the important question isn’t simply whether AI can create an entire fictional series.</p><p>It’s whether that series can help a brand grow.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2649d28b1719" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Shhots AI now makes 30-second animated, 3D, and podcast-style ads]]></title>
            <link>https://rajatkumardangi.medium.com/shhots-ai-now-makes-30-second-animated-3d-and-podcast-style-ads-2475bc26ddc1?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/2475bc26ddc1</guid>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[ai-video-generator]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-solution]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Sun, 04 Oct 2026 19:42:02 GMT</pubDate>
            <atom:updated>2026-10-04T19:42:02.560Z</atom:updated>
            <content:encoded><![CDATA[<h4>Shhots AI now generates animated, 3D and podcast-style ads up to 30 seconds long, from one product photo.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rxz-3XZVhgGHviqR8y9AwQ.png" /></figure><h4>In the last few weeks we shipped a fewthings on <a href="https://shhots.ai/">Shhots AI</a>:</h4><ul><li><a href="https://shhots.ai/blog/30-second-ai-video-launch/"><strong>Generate</strong> <strong>30-second video ads</strong></a> on Seedance 2.5, in animated, Pixar-style 3D and podcast formats</li><li><a href="https://shhots.ai/blog/real-estate-studio-launch/"><strong>AI Studio for Real Estate Agents</strong></a><strong> </strong>for listing photos, virtual staging and walkthrough videos</li><li><a href="https://shhots.ai/blog/ai-photoshoot-templates-launch/"><strong>18 new AI photoshoot templates</strong></a> across 5 product categories</li><li><a href="https://shhots.ai/video-caption-generator/"><strong>Free auto-captions</strong> for any video</a>, unlimited</li></ul><p>Here is what each one does and who it is for.</p><h3>What new ad styles can Shhots AI make?</h3><p>Three new styles, each up to <strong>30 seconds</strong> long, rendered on <a href="https://shhots.ai/models/seedance-2-5/">Seedance 2.5</a>.</p><ul><li><strong>Animated ads:</strong> your product in a fully animated scene, built for feeds full of talking-head UGC.</li></ul><iframe src="https://cdn.embedly.com/widgets/media.html?type=text%2Fhtml&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;schema=twitter&amp;url=https%3A//x.com/shhotsAI/status/2100670968159260855&amp;image=https%3A//i.embed.ly/1/image%3Furl%3Dhttps%253A%252F%252Fpbs.twimg.com%252Famplify_video_thumb%252F2100669796320989184%252Fimg%252F8P3oZlTb2VvzOzG0%253Fformat%253Dwebp%2526name%253Dlarge%26key%3Da19fcc184b9711e1b4764040d3dc5c07" width="500" height="281" frameborder="0" scrolling="no"><a href="https://medium.com/media/bb4bb7a1e696ce5f89a6fd9d75fee5a7/href">https://medium.com/media/bb4bb7a1e696ce5f89a6fd9d75fee5a7/href</a></iframe><ul><li><strong>3D ads (Pixar style):</strong> your product inside a 3D animated world with characters around it.</li></ul><iframe src="https://cdn.embedly.com/widgets/media.html?type=text%2Fhtml&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;schema=twitter&amp;url=https%3A//x.com/shhotsAI/status/2106713294979645585&amp;image=https%3A//i.embed.ly/1/image%3Furl%3Dhttps%253A%252F%252Fpbs.twimg.com%252Famplify_video_thumb%252F2106712889277251584%252Fimg%252FeOpAirG5wk0b5-w_%253Fformat%253Dwebp%2526name%253Dlarge%26key%3Da19fcc184b9711e1b4764040d3dc5c07" width="500" height="281" frameborder="0" scrolling="no"><a href="https://medium.com/media/6683fb0c9a07d12b3213ce3066243a38/href">https://medium.com/media/6683fb0c9a07d12b3213ce3066243a38/href</a></iframe><ul><li><strong>Podcast-style ads:</strong> your product pitched inside a clip that looks cut from a podcast episode.</li></ul><p><a href="https://x.com/shhotsAI/status/2105987451043164442">https://x.com/shhotsAI/status/2105987451043164442</a></p><p><strong>Why 30 seconds matters:</strong> a 15-second ad fits a hook and a reveal. A 30-second ad fits the hook, the problem, the demo and the offer. That is the full structure of most ads I’ve seen scale on Meta.</p><p>You still start from one real product photo. Shhots builds the scene around your real product.</p><h3>What does Real Estate Studio do?</h3><p>It turns raw listing photos into marketing-ready photos and video. No prompts. You upload, pick the job, and download.</p><ul><li><strong>Photo correction:</strong> clean up raw listing shots.</li><li><strong>Virtual staging:</strong> furnish empty rooms.</li><li><strong>Object removal:</strong> remove unwanted objects from a photo.</li><li><strong>Day to dusk:</strong> turn a daytime exterior into a twilight shot.</li><li><strong>Walkthrough videos:</strong> narrated property videos generated from the photos.</li></ul><p>Real Estate Studio is on the <strong>Pro ($49/mo)</strong> and <strong>Scale ($99/mo)</strong> plans, as of October 2026. <a href="https://shhots.ai/ai-for-real-estate-agents/">See Real Estate Studio</a>.</p><p><a href="https://x.com/shhotsAI/status/2102002393760297104">https://x.com/shhotsAI/status/2102002393760297104</a></p><h3>Which AI photoshoot templates are new?</h3><p>18 templates across 5 categories. Upload one product photo, pick a template, and Shhots builds the scene, lighting and props. You write no prompt.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*baIlDyOW5Sj_E13DKN2pJQ.png" /></figure><p>I built these per category because a speaker and a dog leash need different proof. A white-background packshot rarely sells either.</p><p>Templates are on every plan. Starter is <strong>$19/mo</strong> for up to <strong>175 images</strong>, about <strong>$0.11 per image</strong> (as of October 2026).</p><p>Full breakdown: <a href="https://shhots.ai/blog/ai-photoshoot-templates-launch/">18 AI photoshoot templates launch post</a>.</p><h3>Are Shhots auto-captions really free?</h3><p>Yes. Auto-captions are <strong>free and unlimited</strong>, for any video.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wMu5mSEU82DAXfN3qode1A.png" /></figure><ul><li>Upload a video.</li><li>Shhots transcribes it and adds captions.</li><li>Download and post.</li></ul><p>A lot of people watch feeds on mute. An ad without captions loses them. Captions shouldn’t cost credits, so they don’t. <a href="https://shhots.ai/video-caption-generator/">Try the caption generator</a>.</p><h3>Why put all of this in one tool?</h3><p>I’ve bought Meta ads for 8 years across 70+ brands. The bottleneck was rarely one good ad. It was the volume of creative a test needs.</p><ul><li>A real test needs UGC, animated, 3D, static images and product photos.</li><li>Each of those used to mean a different vendor, invoice and turnaround.</li><li>Shhots now makes all of them from one product photo, on one credit balance, from <strong>$19/month</strong>.</li></ul><p>Pick your best-selling product and turn its photo into a 30-second 3D ad: <a href="https://app.shhots.ai/auth/signup">start on Shhots AI</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2475bc26ddc1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Apply for Production Access After 14-Day Testing: Complete Google Play Guide for 2026]]></title>
            <link>https://rajatkumardangi.medium.com/goohow-to-apply-for-production-access-after-14-day-testing-complete-google-play-guide-for-2026-02b4b2363078?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/02b4b2363078</guid>
            <category><![CDATA[android]]></category>
            <category><![CDATA[google-play-console]]></category>
            <category><![CDATA[developer]]></category>
            <category><![CDATA[androiddev]]></category>
            <category><![CDATA[android-app-development]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Tue, 29 Sep 2026 07:27:13 GMT</pubDate>
            <atom:updated>2026-09-29T19:01:50.404Z</atom:updated>
            <content:encoded><![CDATA[<h4>Everything an independent Android developer needs to know about Google Play’s 12-testers-for-14-days requirement, from setting up the test to finally applying for production.</h4><figure><img alt="How to Apply for Production Access After 14-Day Testing" src="https://cdn-images-1.medium.com/max/1024/1*fcLJoL5B7m-UgX6PnYDyfg.png" /></figure><p>If you’ve built your first Android app, finished your Play Store listing, uploaded your release and then discovered that you <strong>can’t publish it yet</strong>, you’re not alone.</p><p>For many independent developers, the final obstacle is now <a href="https://testpogo.com/google-play-closed-testing/">Google Play closed testing</a>.</p><p>And the confusing part isn’t simply finding the right button inside Play Console.</p><p>You need to understand <a href="https://testpogo.com/production-access/"><strong>how to apply for production access after 14-day testing</strong></a>, how the 12-tester requirement works, what your testers actually need to do, and what Google asks when you finally submit your production-access application.</p><p>I recently looked into the process because I’ve seen more independent developers getting stuck at this exact stage.</p><p>Here’s the process, simplified.</p><h3>Table of contents</h3><ol><li>Why Google Play requires closed testing</li><li>Who needs 12 testers for 14 days?</li><li>Internal testing vs closed testing</li><li>How to set up your closed test</li><li>How to find 12 Android testers</li><li>What testers need to do for 14 days</li><li>What happens after the 14-day test</li><li>How to apply for production access</li><li>Common mistakes that delay production access</li><li>A simpler option if you don’t have 12 testers</li><li>Frequently asked questions</li></ol><h3>Why does Google Play require <a href="https://google.play/business/closed-testing/">closed testing</a>?</h3><p>Google introduced additional testing requirements for newer personal developer accounts as part of its effort to improve app quality before applications reach the public Play Store.</p><figure><img alt="Google Play Closed Testing" src="https://cdn-images-1.medium.com/max/1024/1*BMJTdcY8eKOAN6beEuUnCQ.png" /></figure><p>For affected accounts, you can’t simply upload your Android App Bundle and immediately move it into production.</p><p>You first need to demonstrate that the app has gone through a real closed test.</p><p>Google’s current documentation says developers with personal accounts created after <strong>November 13, 2023</strong> must run a closed test involving at least <strong>12 testers who have remained opted in continuously for at least 14 days</strong>.</p><p>Once you’ve met those conditions, you can <strong>apply</strong> for production access.</p><p>That’s an important distinction:</p><p><strong>Finishing the 14-day closed test doesn’t automatically publish your app.</strong></p><p>It unlocks the ability to apply for production access.</p><p>Google then reviews the information you provide about your app, your testing process and whether you believe the app is ready for production.</p><h3>Who needs 12 testers for 14 days?</h3><p>According to <a href="https://support.google.com/googleplay/android-developer/answer/14151465">Google’s current Play Console documentation</a>, this requirement applies to <strong>personal developer accounts created after November 13, 2023</strong>.</p><p>Your own Play Console dashboard should be the final source of truth for your account because Google can change requirements over time.</p><p>If the requirement applies to you, you’ll generally need: <strong>12 or more testers</strong></p><p>who are: <strong>opted into your closed test</strong></p><p>for: <strong>at least 14 continuous days</strong></p><p>before you can submit your production-access application.</p><p>If you’re sitting in Play Console wondering why Production is unavailable even though your app itself is finished, this is probably the requirement blocking you.</p><h3>Internal testing vs closed testing: what’s the difference?</h3><p>This catches a lot of first-time developers.</p><p>Google Play provides several testing tracks, but they don’t all serve the same purpose.</p><h3>Internal testing</h3><p>Internal testing is useful while you’re still actively developing your app.</p><p>You can distribute a build quickly to a small group of people you trust, test features and find obvious bugs.</p><p>Google says internal testing is optional but recommends it as an early testing stage.</p><h3>Closed testing</h3><p><a href="https://testpogo.com/google-play-closed-testing/">Closed testing</a> is different.</p><p>You distribute your app to a controlled set of users, but for affected new personal developer accounts, this is also the testing track that matters for unlocking production access.</p><p>Google specifically requires the qualifying <strong>12 testers for 14 days</strong> to participate through the closed-testing process.</p><h3>Open testing</h3><p>Open testing makes the test accessible more broadly.</p><p>However, Google says open testing becomes available after you gain production access, so it isn’t a shortcut around the closed-testing requirement for affected accounts.</p><p>In simple terms:</p><p><strong>Internal testing → useful during development</strong></p><p><strong>Closed testing → required before production for affected accounts</strong></p><p><strong>Open testing → available after production access</strong></p><p><strong>Production → public Play Store release</strong></p><h3>Step 1: Finish your app setup</h3><p>Before worrying about testers, get the basics right.</p><p>Your application should be in a state where a real person can install it and use its core functionality.</p><p>That means checking things like:</p><ul><li>app crashes</li><li>broken navigation</li><li>login and registration</li><li>permissions</li><li>subscription or payment flows</li><li>API failures</li><li>unusable screens</li><li>unfinished placeholder content</li><li>privacy-policy requirements</li><li>store-listing information</li></ul><p>Closed testing shouldn’t be treated as a bureaucratic countdown.</p><p>It’s a chance to find the things you missed while developing the application yourself.</p><p>As the developer, you already know how everything is <em>supposed</em> to work.</p><p>A tester doesn’t.</p><p>That’s exactly why their experience can be valuable.</p><h3>Step 2: Create the closed testing release</h3><p>Inside Google Play Console, create your closed-testing track and upload the build you want people to test.</p><p>You’ll then need a way to control who gets access.</p><p>For many developers this involves creating a Google Group containing the testers and connecting that group to the closed test.</p><p>Your testers then receive an opt-in link.</p><p>The important part is that they actually complete the opt-in process.</p><p>Simply collecting someone’s Gmail address does not mean you’ve completed the requirement.</p><h3>Step 3: Find at least 12 Android app testers</h3><p>This is where the process becomes surprisingly difficult for solo developers.</p><p>Finding 12 people sounds easy until you add the actual constraints:</p><p>They need an Android device.</p><p>They need to join your test correctly.</p><p>They need to stay opted in.</p><p>And they need to remain there for the required testing period.</p><p>Your realistic options fall into four categories.</p><h3>Option 1: Friends and family</h3><p>This is the obvious starting point.</p><p>If you know enough Android users, ask them directly.</p><p>The advantage is trust. You can message them if something goes wrong.</p><p>The disadvantage is that many developers simply don’t have 12 people around them who use Android and are willing to participate in an app test for two weeks.</p><h3>Option 2: Other developers</h3><p>A common approach is <strong>test-for-test</strong>.</p><p>You test another developer’s app and they test yours.</p><p>There are dedicated communities built around this.</p><p>In fact, recent posts in Android testing communities show developers continuing to exchange apps specifically to satisfy the 14-day closed-testing requirement.</p><p>This method costs nothing financially.</p><p>But it costs time.</p><p>If you join 12 reciprocal arrangements, you may also end up with 12 apps that you’re expected to test in return.</p><h3>Option 3: Reddit and developer communities</h3><p>Communities like Reddit can work surprisingly well.</p><p>Developers regularly post their Google Group, test opt-in and Play Store links and offer to test other apps in return.</p><p>The problem is coordination.</p><p>Getting 12 people to comment on a Reddit post is one thing.</p><p>Knowing whether all 12 will still be participating two weeks later is another.</p><p>For a hobby project without a deadline, that may be perfectly acceptable.</p><p>For a client project or a launch with a date attached, it’s less predictable.</p><h3>Option 4: Use an Android testing service</h3><p>The final option is a managed <a href="https://testpogo.com/"><strong>Android app testing service</strong></a>.</p><p>Instead of recruiting testers individually, you pay a service to provide the tester group and coordinate the testing period.</p><p>This is the approach that makes the most sense when your time is worth more than the cost of recruiting and managing the testers yourself.</p><p>I’ll come back to one inexpensive option later in the article.</p><h3>Should you recruit exactly 12 testers?</h3><p>I wouldn’t.</p><p>Twelve is Google’s minimum for the affected accounts, not necessarily the number I’d personally start with.</p><p>If you’re relying on volunteer testers and exactly one person doesn’t opt in correctly, you’ve suddenly lost your safety margin.</p><p>Recruiting a few extra Android testers gives you room for mistakes and dropouts.</p><p>You can see the same behaviour in developer communities: many developers deliberately look for 15–20 participants rather than stopping at exactly 12.</p><p>The goal isn’t to collect the largest possible number.</p><p>It’s to make sure you still satisfy the requirement when you’re ready to apply.</p><h3>Step 4: Actually use the 14 days as a test</h3><p>Here’s where I’d avoid some of the questionable advice you’ll find online.</p><p>Google’s published requirement is clear about the continuous <strong>opt-in</strong> requirement.</p><p>It also asks you about your testing process and production readiness when you apply.</p><p>So don’t treat the 14 days as:</p><blockquote><em>Get 12 installs → ignore app → wait → apply.</em></blockquote><p>Give testers things to test.</p><p>For example:</p><h3>Day 1–3: onboarding</h3><p>Ask testers whether:</p><ul><li>installation worked</li><li>signup was understandable</li><li>permissions made sense</li><li>onboarding explained the product</li></ul><h3>Day 4–7: core functionality</h3><p>Ask them to use the main reason your app exists.</p><p>For a finance app, create or categorize a transaction.</p><p>For a workout app, complete a workout.</p><p>For an editing app, create and export something.</p><h3>Day 8–11: edge cases</h3><p>Try:</p><ul><li>bad internet connections</li><li>wrong inputs</li><li>backgrounding the app</li><li>returning after several days</li><li>different Android devices</li><li>permission denial</li></ul><h3>Day 12–14: final feedback</h3><p>Ask:</p><ul><li>What confused you?</li><li>What broke?</li><li>Which feature felt unfinished?</li><li>Would you feel comfortable using this version?</li><li>What should change before release?</li></ul><p>Now you have something meaningful to say in your production-access application.</p><h3>Step 5: Fix problems found during closed testing</h3><p>This is the part where closed testing becomes useful rather than annoying.</p><p>Keep a simple issue list.</p><p>Something like:</p><p>IssueSeverityTesterActionLogin error on older deviceHighTester 4FixedButton text unclearLowTester 7Changed copyCrash after image uploadCriticalTester 2FixedOnboarding too longMediumTester 9Reduced to 3 steps</p><p>You don’t need an enterprise QA system.</p><p>You just need evidence that you actually tested the application, learned something and improved it.</p><h3>What happens after 14 days of closed testing?</h3><p>Once at least 12 testers satisfy Google’s continuous opt-in requirement, Play Console should allow you to move on to the production-access application.</p><p>Again, the key phrase is:</p><p><strong>apply for production access.</strong></p><p>You aren’t automatically approved simply because the clock reached day 14.</p><p>Google says you’ll answer questions intended to understand your app, your testing process and whether the application is ready for production.</p><p>So don’t rush through these questions.</p><h3>How to apply for production access after 14-day testing</h3><p>Once your closed test qualifies, open your <strong>Google Play Console Dashboard</strong> and begin the <a href="https://testpogo.com/production-access/">production-access</a> application.</p><p>Google says the application asks questions in several areas, including your testing process, your app and its production readiness.</p><p>Your exact questions can change, so answer what’s actually shown in your account.</p><p>But you should be ready to explain three things.</p><h3>1. Who tested your application?</h3><p>Describe your tester group honestly.</p><p>For example:</p><blockquote><em>We recruited Android users matching our expected audience and gave them access through our closed-testing track. Testers used the core workflows of the application across the testing period.</em></blockquote><p>Don’t invent demographics or testing behaviour that didn’t happen.</p><h3>2. What did you learn during the test?</h3><p>This is why collecting feedback matters.</p><p>A useful answer is specific:</p><blockquote><em>Testers identified confusion during onboarding and an issue affecting image uploads on two devices. We shortened the onboarding flow and fixed the upload error before completing the test.</em></blockquote><p>That says much more than:</p><blockquote><em>Everything worked great.</em></blockquote><h3>3. Why is the app ready for production?</h3><p>Explain what you validated.</p><p>For example:</p><blockquote><em>We tested account creation, authentication, the primary workflow, error states and repeat usage across multiple Android devices. Issues identified during testing were fixed and the resulting build was retested before applying for production.</em></blockquote><p>Specificity makes the answer more useful and accurate.</p><h3>Does 12 testers for 14 days guarantee approval?</h3><p>No.</p><p>Google’s own documentation describes the closed test as a requirement you must meet <strong>before applying</strong> for production access. Google then reviews your application.</p><p>That’s why I’d be skeptical of anyone claiming they can directly “approve” your Google Play application.</p><p>They can’t.</p><p>Google controls production access.</p><p>A testing service can provide testers.</p><p>A developer can improve the app.</p><p>But Google’s review decision remains Google’s.</p><h3>What if you don’t have 12 people to test your Android app?</h3><p>This is probably the most common problem for an independent developer.</p><p>You finished the difficult part.</p><p>And now you’re stuck asking people: <strong>“Do you happen to own an Android phone, and can you keep testing my app for the next two weeks?”</strong></p><p>You can absolutely solve this through Reddit, developer groups and reciprocal testing.</p><p>If you have more time than money, I’d start there.</p><p>But while researching the available Android testing services, I also came across <a href="http://testpogo.com/"><strong>TestPogo</strong></a>, which is built specifically around this Google Play requirement.</p><p><a href="http://testpogo.com/">TestPogo</a> provides real Android testers for a 14-day closed test. Its current plans start at <strong>$5 for five testers</strong> if you already have some participants, while the <strong>$10 plan provides 15 testers</strong> for developers starting from scratch. TestPogo says testers can begin within six hours, uses real people on their own Android phones, and doesn’t require your Play Console password or account access.</p><p>You can see how it works here — <a href="https://testpogo.com/#how-it-works">https://testpogo.com/#how-it-works</a></p><p>What I like about this model is that the service isn’t pretending it can grant production access.</p><p>Google still controls that.</p><p>It’s simply solving the frustrating operational problem:</p><p><strong>finding enough reliable Android testers and managing the 14-day test.</strong></p><p>If you’d rather spend your time fixing your app than chasing twelve people through DMs, that may be worth the few dollars.</p><h3>Common Google Play closed-testing mistakes</h3><h4>Mistake 1: Starting with exactly 12 unreliable testers</h4><p>Give yourself some buffer.</p><p>A few extra testers are much easier than discovering that your test isn’t where you expected it to be.</p><h4>Mistake 2: Confusing internal testing with closed testing</h4><p>Internal testing is useful, but Google’s production-access requirement for affected personal accounts specifically involves closed testing.</p><h4>Mistake 3: Waiting until day 14 to ask for feedback</h4><p>Collect feedback while people still remember what happened.</p><h4>Mistake 4: Treating the requirement as 12 downloads</h4><p>The published rule refers to testers remaining opted into the closed test continuously for the required period.</p><h4>Mistake 5: Fabricating your production-access answers</h4><p>Tell Google what you actually did.</p><p>A small but genuine test is better than an elaborate fictional story.</p><h3>Mistake 6: Sharing sensitive Play Console access</h3><p>Recruiting testers shouldn’t require giving strangers your Google password.</p><p>Use the testing mechanisms Google provides.</p><h3>A simple checklist before applying for production</h3><p>Before clicking the production-access application, I’d verify:</p><ul><li>at least 12 qualifying testers are still opted in</li><li>the 14-day requirement has been completed</li><li>your production candidate installs correctly</li><li>important crashes have been fixed</li><li>you’ve collected real tester feedback</li><li>you’ve documented the important changes made during testing</li><li>your Play Store listing is complete</li><li>your privacy and app-content information is accurate</li><li>you’re prepared to explain your testing process</li><li>you’re prepared to explain why the app is ready</li></ul><p>Then apply.</p><h3>FAQs — Frequently asked questions</h3><h4>How many testers do I need for Google Play closed testing?</h4><p>For personal developer accounts covered by Google’s current requirement, Google says at least <strong>12 testers</strong> must remain continuously opted into the closed test for at least <strong>14 days</strong> before the developer can apply for production access.</p><h4>How long does Google Play closed testing take?</h4><p>The qualifying period itself is at least 14 days for affected developer accounts. Recruiting and onboarding your testers can add additional time.</p><h4>Can I use friends as Google Play testers?</h4><p>Yes. The important thing is that the testers properly participate in your closed test.</p><h4>Where can I find Android app testers?</h4><p>You can recruit friends, find other developers through communities such as Reddit, participate in test-for-test groups, recruit your own users or use a managed Android testing service.</p><h4>Can I pay for Android app testing?</h4><p>There are Android app testing services that recruit and manage real testers. The important distinction is between paying for genuine testing and expecting someone to somehow bypass Google’s review process.</p><h4>What is the difference between internal and closed testing?</h4><p>Internal testing is an optional way to quickly test builds with trusted people. Closed testing is the controlled testing track Google requires affected new personal developer accounts to complete before applying for production access.</p><h4>What happens after the 14-day test?</h4><p>Once you satisfy Google’s requirement, you can apply for production access from Play Console and answer questions about your testing, application and production readiness.</p><h3>Final thoughts</h3><p>Google Play’s closed-testing requirement can feel strange when you’ve spent months building an application and the final thing keeping it off the Play Store is finding twelve people with Android phones.</p><p>But there are really two ways to approach it.</p><p>If you have time, use communities, other developers, friends and test-for-test arrangements.</p><p>If you want to remove the coordination work, a managed testing service such as <a href="https://testpogo.com/?utm_source=chatgpt.com"><strong>TestPogo</strong></a> can supply the Android testers while you focus on fixing whatever they find.</p><p>Either way, don’t think of closed testing as fourteen days you have to survive.</p><p>Use it to answer the more important question: <strong>Would I actually trust this version of my app in the hands of thousands of strangers?</strong></p><p>If the answer is yes, and your 12 testers have completed the required period, it’s time to make the production-access application.</p><p>All the best :)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=02b4b2363078" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[These 6 Free Claude Skills Can Replace an Entire Ad Creative Team]]></title>
            <link>https://rajatkumardangi.medium.com/these-6-free-claude-skills-canreplace-an-entire-ad-creative-team-5257000599b8?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/5257000599b8</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-tools]]></category>
            <category><![CDATA[calude-skills]]></category>
            <category><![CDATA[ai-ads-generator]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Thu, 20 Aug 2026 08:55:51 GMT</pubDate>
            <atom:updated>2026-08-20T08:56:12.887Z</atom:updated>
            <content:encoded><![CDATA[<h4>Six free Claude skills that write, direct, and (with <a href="https://shhots.ai/mcp/">Shhots MCP</a> connection) generate your ads. No prompt engineering required.</h4><p>Most people trying to make a UGC ad with AI get stuck before generation starts.</p><p>The hooks don’t land. The script makes a claim that gets the ad rejected. Nobody knows if this product needs a talking head or claymation. A real agency solves this with a scriptwriter, an art director, and a media buyer. Most founders and small teams have none of the three.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RNufSrZQOYZUzl7M2h8Imw.png" /></figure><p>Shhots built <a href="https://github.com/shhots/shhots-claude-skills">six Claude skills</a> to close that gap. Connect them to the <a href="https://shhots.ai/mcp/">Shhots MCP</a> and <a href="https://claude.ai/">Claude</a> stops describing your ad. It generates it.</p><h4><strong>What’s In The Six Skills</strong></h4><pre>Shhots Skills for Claude<br>├─ shhots-ugc-ads<br>│   └ Hooks, claim-safe scripts, visual treatments across all 7 video formats<br>├─ shhots-photoshoot<br>│   └ Art-director direction for studio, lifestyle, editorial, bold product shots<br>├─ shhots-campaign<br>│   └ Credit budgeting, sequenced generation, a testing plan<br>├─ shhots-creative-refresh<br>│   └ One-variable variants of a winning ad to beat creative fatigue<br>├─ shhots-ugc-styles<br>│   └ 10 distinct styles: claymation, stop motion, whiteboard, podcast clip, and more<br>└─ shhots-image-ads<br>    └ Hundreds of static ad varieties across 14 formats, category playbooks, and funnel stages</pre><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FPPOwBVBSsw4%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DPPOwBVBSsw4&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FPPOwBVBSsw4%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/471ec5e326fe892f7ebb5eba651c1f2f/href">https://medium.com/media/471ec5e326fe892f7ebb5eba651c1f2f/href</a></iframe><h4><strong>Skill Versus MCP</strong></h4><p>A skill is instructions. Claude knows what a scroll-stopping hook sounds like. It knows which claims survive ad review. It knows which of Shhots’ seven video formats fits your product. An MCP is access. It lets Claude call Shhots’ generation engine and hand back a finished video or image.</p><p>Connect both and one conversation does the job an agency used to do. Describe your product. Claude writes the direction. The same chat generates the ad. Skip the MCP and you still get the creative direction. Paste it into the Shhots web app and generate it yourself.</p><h4><strong>Installing The Skills</strong></h4><p>On Claude Code or Cowork, run these two commands:</p><p>/plugin marketplace add shhots/shhots-claude-skills<br> /plugin install shhots@shhots-claude-skills</p><p>On Claude.ai, download the skill packages from the GitHub repo’s Releases page and upload them under Settings, Skills.</p><p>The MCP connection is separate and lives on Shhots’ Pro and Scale plans. Authorize it in Claude’s Settings under Connectors. Full setup is at shhots.ai/mcp.</p><h4><strong>How It Works</strong></h4><p>Say you’re launching a vitamin C serum for TikTok. You tell shhots-ugc-ads what the product is, who it’s for, and where it’s running. It returns hook options, a claim-safe script, and a format recommendation. If an existing ad is fatiguing, shhots-creative-refresh changes one variable at a time: the hook, the setting, the avatar. That’s how you find out what’s actually driving performance instead of guessing. If you’re planning a full campaign, shhots-campaign sequences the generation and budgets the credits for you.</p><p>With the MCP connected, Claude picks the underlying model for the format you need: VEO, Kling, Seedance, Minimax, GPT-image, or Nano Banana. It assembles the final ad in the same chat.</p><h4><strong>The Numbers</strong></h4><p>Product image generations cost 11 credits. Videos run 551 to 852 credits depending on length. A new photo analysis costs 1 credit. Video formats include UGC, cinematic, animated, product-speaking, storefront, lifestyle, and unboxing or ASMR, in 5, 10, or 15 seconds, across 9:16, 1:1, and 16:9. UGC voiceovers cover 12 languages. Shhots blocks invented product claims at the tool level, so a script never asserts something your product page doesn’t back up.</p><h4><strong>Why This Is Free</strong></h4><p>A scriptwriter and an art director should not cost a monthly retainer to access. That’s why the skills are open and free. Shhots makes money on generation: turning the strategy into a finished, watermark-free video or image with a full commercial license.</p><p>The Starter plan is $17 a month billed annually, for 2,000 credits: enough for roughly 175 image generations or a handful of UGC and cinematic video ads. New accounts get 20% off their first generation with code UGC20.</p><h4><strong>Get It</strong></h4><p>Install from <a href="https://github.com/shhots/shhots-claude-skills">github.com/shhots/shhots-claude-skills</a>. Six commands, free either way you use it. Connect the MCP and the next ad you make comes out of a chat, not an agency invoice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5257000599b8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI for Real Estate Agents: The Four Listing Jobs Worth Automating in 2026]]></title>
            <link>https://rajatkumardangi.medium.com/ai-for-real-estate-agents-the-four-listing-jobs-worth-automating-in-2026-d80ad63152e0?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/d80ad63152e0</guid>
            <category><![CDATA[ai-video-generator]]></category>
            <category><![CDATA[real-estate-agent]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[real-estate-marketing]]></category>
            <category><![CDATA[real-estate]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Wed, 19 Aug 2026 21:31:27 GMT</pubDate>
            <atom:updated>2026-08-19T21:31:27.918Z</atom:updated>
            <content:encoded><![CDATA[<h4><strong>Real estate agents kept signing up to Shhots AI and hitting a prompt box. So we built Real Estate Studio. Here is what it does, what a listing actually costs, and where AI still has no business being.</strong></h4><p>Real estate agents started showing up on <a href="https://shhots.ai/">Shhots AI</a> before we had built anything for them.</p><p>They came for a product-ad tool. They uploaded a photo of an empty living room and typed something like “add a sofa and make it look nice.” The model gave them a living room. A different living room, with the window moved and the fireplace gone. Then they tried again, got a third room, and left.</p><p>The prompt box was the problem. A listing shoot is thirty photos of the same house, and the whole point is that photo ten looks like it belongs with photo one. A text box gives you thirty different houses.</p><p>So we built <a href="https://shhots.ai/ai-for-real-estate-agents/">Real Estate Studio</a>: a property-first workspace with sixteen named presets, zero prompt writing, and a hard rule that the AI may never move a wall. It shipped on the Pro and Scale plans, and this is the full breakdown of what it does.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fkgc9ZGRcfIY%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dkgc9ZGRcfIY&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fkgc9ZGRcfIY%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/46b7ea1dab553589daca892ffe031733/href">https://medium.com/media/46b7ea1dab553589daca892ffe031733/href</a></iframe><h3>The short version</h3><p>Four listing jobs are now genuinely automatable: correcting photos, staging empty rooms, removing clutter, and building listing video.</p><p>Each one used to mean a vendor, an invoice, and a wait of one to three days. Each one now takes under a minute.</p><ul><li><strong>Correction:</strong> nine named presets covering HDR from a single exposure, sky replacement, white balance, perspective, window pull, reflection removal, denoise, and 4K upscaling.</li><li><strong>Staging:</strong> three modes across eight styles, roughly $0.45 per staged image against $16 to $75 at a staging service.</li><li><strong>Removal:</strong> four presets for furniture, single objects, general clutter, and privacy blur.</li><li><strong>Video:</strong> three or four photos become a 15-second narrated walkthrough in 16:9 with an on-camera presenter.</li></ul><p>Everything files itself against a property record. The untouched original is never overwritten. Pro is $49 a month for 6,000 credits, Scale is $99 for 12,000, and one credit pool covers all four jobs.</p><h3>What AI for real estate agents actually does today</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nVm3NZ320v2NAXkuBjE-tA.png" /><figcaption>AI for real estate agents — Shhots AI</figcaption></figure><p>Search “AI for real estate agents” and you get a wall of results, most of them selling four different products under one label: CRM, lead capture, valuation, and media. Those are separate software categories with separate buyers, and a tool that claims all four usually has one strong quarter.</p><p>The part AI genuinely changed in the last eighteen months is listing marketing. Photo correction, virtual staging, object removal, upscaling, and listing video all crossed from “outsource it” to “do it before you leave the property.”</p><p>The part AI has not changed is the part clients pay a commission for. Pricing a house in a thin market. Reading what a seller is actually anxious about. Holding a deal together after an inspection comes back ugly. No current model is close, and anyone telling you otherwise is selling something.</p><p>The useful framing when you are evaluating <a href="https://shhots.ai/ai-for-real-estate-agents/">AI tools for real estate agents</a>: buy the media studio and the CRM separately, and judge the media studio on cost per listing rather than cost per month.</p><h3>Why presets beat prompts for listing photos</h3><p>General image models want a written description and reward the people who get good at writing one. That is the wrong shape for listing work.</p><p>A listing gallery has one requirement above all others: internal consistency. The Coastal bedroom and the Coastal living room need to read as the same house on the same afternoon. A prompt cannot guarantee that, because a prompt is a fresh negotiation with the model every time you run it.</p><p>A named preset is the same instruction on every image. Run “Sky Replacement” across six exteriors and you get six exteriors that match. Run “Stage empty room, Scandinavian” across four bare rooms and the four rooms share a design language.</p><p>Real Estate Studio is promptless for that reason. There is one optional comments field per generation, one line, for the rare case where something specific has to survive the edit. It is never required.</p><h3>The workspace: one property, one gallery</h3><p>Every listing gets its own record: property name, full address, beds, baths, and price. That record does two jobs. It keeps media filed per listing, so a brokerage running forty active properties never digs through a shared drive for final_v3_edited.jpg. And it feeds the walkthrough video script later, so nobody types the specs twice.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SK4D2_hINHxLTGsFnlcyvg.png" /><figcaption>Shhots AI for Real Estate Agents</figcaption></figure><p>Upload the full shoot at once and every photo is classified by room type: exterior front, living room, kitchen, dining, bedroom, bathroom, office. A forty-photo shoot arranges itself into a grouped gallery. Any photo can be reassigned from one dropdown if the classifier reads a den as a bedroom.</p><p>Select an image and the edit panel offers three actions: <strong>Correction</strong>, <strong>Staging</strong>, <strong>Removal</strong>. Pick one, pick a preset, set how many variations you want, and generate. The credit cost is shown before you commit, and every generation saves as a new version alongside the untouched original.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/1*YEYz3XVW73wCo3t30ht0og.png" /><figcaption>AI Virtual Staging, Correction, and Removal for Real Estate</figcaption></figure><h3>Job one: AI real estate photo editing</h3><p>Listings live or die in the first eight photos, and the bottleneck was never the camera. It was the twelve to forty-eight hours between shoot day and the gallery hitting the MLS while an editor somewhere worked through a queue.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9K3kFlhMARYK-QqMPU5zbQ.png" /><figcaption>AI Real Estate Photo Editing</figcaption></figure><p><a href="https://shhots.ai/ai-real-estate-photo-editing/">AI real estate photo editing</a> runs on nine named correction presets:</p><ol><li><strong>Auto Enhance.</strong> A balanced overall correction in one click. Enough for most photos.</li><li><strong>HDR / Exposure Balance.</strong> Recovers blown-out windows and dark corners from a single exposure. No bracketing.</li><li><strong>White Balance / Colour Fix.</strong> Removes the colour cast mixed indoor lighting always leaves.</li><li><strong>Perspective / Straighten.</strong> Corrects converging verticals and tilted horizons from wide-angle interiors.</li><li><strong>Window Pull.</strong> Replaces washed-out window glare with the actual outdoor view.</li><li><strong>Sharpen &amp; Denoise.</strong> Cleans soft or grainy shots, including phone photos in low light.</li><li><strong>Upscale to 4K.</strong> Lifts low-resolution images to print and MLS quality for brochures and signage.</li><li><strong>Reflection Removal.</strong> Clears mirror and glass reflections, including the photographer.</li><li><strong>Sky Replacement.</strong> Swaps grey overcast for a clean sky, masking around rooflines, trees, and power lines.</li></ol><p>Traditional HDR needs three bracketed exposures blended by hand in Lightroom. The HDR preset synthesises it from one frame, which matters most for the agent shooting a listing on an iPhone at 4pm.</p><p>Each generation costs 55 credits, roughly $0.45 on Pro, and returns in under a minute. An outsourced editing service charges $1.50 to $3.00 per photo and takes twelve to forty-eight hours.</p><p>For brokerages, portals, and photography companies running hundreds of properties, there is a batch path: <a href="https://shhots.ai/bulk-real-estate-photo-editing/">bulk real estate photo editing</a>.</p><h3>Job two: AI virtual staging in eight styles</h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fs3XJknUXy1Y%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Ds3XJknUXy1Y&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fs3XJknUXy1Y%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/aec34bab10a6956cfbabfebd8fdd0651/href">https://medium.com/media/aec34bab10a6956cfbabfebd8fdd0651/href</a></iframe><p>An empty room is the hardest thing you can ask a buyer to judge. Without a bed or a sofa in frame there is no reference object, so a generous bedroom photographs as a box and a narrow one photographs as a corridor. Staging supplies the reference. That is the entire mechanism. NAR’s 2024 data puts empty listings on market 73% longer than staged ones.</p><p><a href="https://shhots.ai/ai-virtual-staging/">AI virtual staging</a> on Shhots AI runs three modes:</p><p><strong>Stage empty room.</strong> Furnishes a bare space from scratch, matching furniture lighting and shadow to the photograph so the sofa sits in the same light as the floor under it. This is the mode for new builds and vacant resales.</p><p><strong>Restyle room.</strong> Keeps the existing furniture layout and swaps the pieces and decor for a new aesthetic. Useful when a room reads as dated rather than empty, and the mode interior designers use to show a client their actual living room in a different direction.</p><p><strong>Occupied to vacant.</strong> Strips a lived-in room back to an empty shell, then stages it fresh in whichever style you pick. Reach for this when the seller still lives in the house and their furniture is working against the listing. Most tools charge for this as a premium add-on. Here it is a mode at the same price.</p><p>Every mode carries the same eight styles: <strong>Modern, Scandinavian, Luxury, Coastal, Farmhouse, Minimalist, Boho, and Mid-century.</strong> Three modes times eight styles gives you twenty-four combinations before you have typed a word.</p><h3>Staging room by room</h3><p>Each room asks the buyer a different question, so the brief changes with it.</p><ul><li><strong>Bedroom.</strong> Staging settles whether a king fits. Scandinavian and Minimalist keep a small bedroom open. Luxury suits a primary suite.</li><li><strong>Living room.</strong> Usually the hero shot and the portal thumbnail, so it earns the most attention. Run it in two or three styles and pick for the neighbourhood’s buyer profile rather than your own taste.</li><li><strong>Dining room.</strong> Empty dining rooms photograph as corridors. A table and six chairs answers the question a family buyer is actually asking.</li><li><strong>Kitchen.</strong> Rarely needs furniture. It needs Removal plus a light styling pass so the worktop looks usable rather than bare.</li><li><strong>Apartments and condos.</strong> Perceived square footage decides everything. Minimalist and Modern keep the floor visible. Heavy Farmhouse pieces eat a studio in a wide-angle shot.</li><li><strong>Commercial.</strong> Office suites and retail units stage the same way. Structure preservation matters more here, because a tenant is measuring the space off your photograph.</li></ul><h3>The rule the whole studio is built around</h3><p>Plenty of image models will happily widen a room, add a window, or move a wall to make a composition work. In a property listing that is a liability.</p><p>Walls, windows, doors, floors, ceilings, built-ins, and sightlines stay exactly where the camera found them. Only the contents change. If a buyer measures the room off your photo, the measurement holds at the showing.</p><p>The untouched original is saved beside every generated version, so producing the before file for a disclosure request takes one click.</p><p>More on the format itself in our guides on <a href="https://shhots.ai/blog/what-is-virtual-staging/">what virtual staging is</a> and <a href="https://shhots.ai/blog/virtual-staging-before-and-after/">virtual staging before and after</a>, which grades six real listing examples against a checklist.</p><h3>Job three: removal, clutter, and privacy</h3><p>The fourth listing job hides inside the other three, which is why it rarely gets its own tool. Four presets handle it:</p><ul><li><strong>Remove furniture.</strong> Empties a lived-in room so buyers see the space rather than the seller’s sofa.</li><li><strong>Object removal.</strong> Takes out one specific thing: the bin, the pet bed, the cables, the car on the driveway.</li><li><strong>Declutter.</strong> Clears countertops, shelves, and floors across the whole frame in one pass.</li><li><strong>Privacy Blur.</strong> Obscures family photos, documents, and personal details before the listing goes public.</li></ul><p>That last one is the difference between a listing and a complaint from your seller. Nobody thinks about the medical letter on the fridge until it is on Zillow at full resolution.</p><h3>Job four: the AI real estate video generator</h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FcoLrNSgms7c%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DcoLrNSgms7c&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FcoLrNSgms7c%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/bb856675fc1b65220a0a2c3c7711ed77/href">https://medium.com/media/bb856675fc1b65220a0a2c3c7711ed77/href</a></iframe><p>Video used to be the thing you commissioned for the listing that justified it, which in most markets meant the top ten percent by price. Everything else got a photo gallery. That split existed because a videographer costs several hundred dollars and returns the edit in three to seven days.</p><p>The <a href="https://shhots.ai/ai-real-estate-video-generator/">AI real estate video generator</a> takes three or four photos from the property gallery and returns a 15-second narrated walkthrough in 16:9 with an on-camera presenter.</p><p><strong>Step one.</strong> Pick three or four images from the grouped gallery. Edited versions are the ones to use, because the video is only as good as the stills behind it. The tour order panel is drag-and-drop, so you decide which room opens and which one closes. Lead with the exterior for a portal cut, lead with the kitchen for social.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jfXVR-96Idb_4yiSFSk5BA.png" /><figcaption>AI Real Estate Walkthrough Video Generator</figcaption></figure><p><strong>Step two.</strong> The script writes itself from the property record for 1 credit. Address, beds and baths, price, and agent name are four separate toggles, and anything you switch off is never mentioned in the narration. Switch price off on a listing where the number is not public yet and the presenter will not say it. Every line stays editable up to 1,800 characters, and what you see in the editor is exactly what the video model receives.</p><p><strong>Step three.</strong> Render.</p><h3>Why fifteen seconds, and why a presenter</h3><p>Both constraints are deliberate, and both come from how listing video gets watched rather than how it gets produced.</p><p>A room-by-room deep tour buries the hook and loses the viewer before the kitchen. A short narrated overview with the address, the specs, and the price gets watched to the end, and the buyer who wants more clicks through to the gallery. The video is the trailer. The listing is the film.</p><p>On the presenter: slideshow videos with a music bed read as filler on a social feed, because nothing in frame is asking to be looked at. A face and a voice give the clip a subject. It will never replace an agent’s own piece to camera on a personal brand video. What it replaces is the music-bed slideshow most listings currently ship. We compared seven options on this in our guide to <a href="https://shhots.ai/blog/ai-avatar-tools-real-estate-walkthroughs/">AI avatar tools for real estate walkthroughs</a>.</p><h3>Six places a listing video earns its keep</h3><p>Instagram Reels and TikTok. Listing portal video slots. WhatsApp and email straight to a buyer list the hour the listing goes live. Geotargeted Facebook ads in the week before an open house. Vendor pitches, where showing a seller the video you would make for their property wins instructions. And brokerage team pages, where every walkthrough comes out of the same generator instead of a different freelancer each time.</p><p>Because there is no per-video invoice, making three cuts of the same listing is normal. Exterior-led for the portal, kitchen-led for Instagram, price-led for the brokerage page. Same gallery, different order.</p><p>Walkthroughs render in 16:9. For vertical 9:16 cuts, run individual photos through <a href="https://shhots.ai/image-to-video/">image to video</a> for camera motion, then burn in captions with the <a href="https://shhots.ai/video-caption-generator/">caption generator</a>, because most social viewing happens with the sound off.</p><p>Two deeper reads if this is the job you care about: <a href="https://shhots.ai/blog/how-to-make-a-real-estate-video/">how to make a real estate video in five steps</a> and the <a href="https://shhots.ai/blog/real-estate-video-marketing-guide/">real estate video marketing playbook</a>.</p><h3>What a listing actually costs</h3><p>Cost per month is the wrong number to compare. Cost per listing is the one that decides whether you use the tool on listing day.</p><p>Every image generation is 55 credits. Pro is $49 a month for 6,000 credits, which is about 109 generated images, so roughly $0.45 each. Scale is $99 a month for 12,000 credits, about 218 images. Walkthrough scripts cost 1 credit, and the render is billed on top from the same pool. One credit pool covers all four jobs, so $0.45 is a ceiling rather than a forecast.</p><p>Worked example on a standard listing:</p><ul><li>Twenty photos through Auto Enhance, Sky Replacement on the exteriors, Declutter on the kitchen and bathrooms: 1,100 credits, about $9.</li><li>Four empty rooms staged in two styles each so you can choose: 440 credits, about $3.60.</li><li>One walkthrough script: 1 credit, plus the render.</li></ul><p>That is roughly $12.60 in generations for the still gallery on a full listing, which puts close to four listings a month inside the Pro plan before you count video renders.</p><p>Against the alternative: outsourced photo editing at $1.50 to $3.00 per photo, virtual staging services at $16 to $75 per image, physical staging at $1,500 to $4,000 per listing, and a videographer at several hundred dollars with a three to seven day turnaround.</p><p>The interesting part is behavioural rather than financial. At $60 an image you stage three rooms and pick blind. At $0.45 you stage every room, run the important ones in three styles, and list the version that actually sells the space. Full numbers are on the <a href="https://shhots.ai/pricing/">pricing page</a>, and we broke down all three market tiers in the <a href="https://shhots.ai/blog/virtual-staging-cost/">virtual staging cost guide</a>.</p><p>For the video side specifically, <a href="https://shhots.ai/blog/real-estate-videographer-cost-vs-ai/">real estate videographer cost versus AI</a> does the same maths and says plainly where a videographer still wins.</p><h3>Free virtual staging tools and the three walls</h3><p>Free tools are useful for exactly one thing, which is finding out whether you like the workflow before you pay for it.</p><p>For a live listing they hit the same three walls every time. A watermark burned into the output. Resolution capped below what the MLS accepts. A daily generation limit that makes a full shoot impossible. None of those produce a file you can put on a portal.</p><p>Shhots AI has no free tier and no per-image billing. Real Estate Studio is on Pro at $49 a month and Scale at $99, at full resolution, watermark-free, with a full commercial license on every plan. The same file works on the MLS, Zillow, Realtor.com, Redfin, your brokerage site, printed brochures, and paid Facebook or Google ads.</p><h3>Is virtual staging legal, and what do you have to disclose?</h3><p>Virtual staging is legal and widely used. In almost every jurisdiction the rule is the same: stage freely, and say that you staged.</p><ul><li><strong>NAR Article 12</strong> requires a true picture in advertising.</li><li><strong>Most MLS systems</strong> require a label such as “virtually staged” on any materially altered photo.</li><li><strong>Several US states</strong> have added their own statutes. California’s <strong>AB 723</strong>, in effect since 1 January 2026, requires a disclosure near any digitally altered listing image plus a link to the original, unaltered version.</li></ul><p>Agents get into trouble over two specific things: structural changes and hidden defects. Nobody has been sanctioned for a replaced sky or a labelled sofa. Someone will be sanctioned for an edit that widened a room or painted over water damage.</p><p>Which is why the studio cannot move a wall, and why every generated version is stored alongside the untouched original. When AB 723 asks you to link the unaltered file, the unaltered file is one click away.</p><p>Staging adds furniture. It is not a repair tool. And it should never imply the sofa conveys with the sale.</p><p>The full picture is in our guide to <a href="https://shhots.ai/blog/is-virtual-staging-legal-mls-rules/">virtual staging rules and MLS disclosure</a>.</p><h3>What Real Estate Studio does not do</h3><p>Most results for “AI tools for real estate agents” are selling something from this list. These are real categories with good products in them. They are not this, and pretending otherwise would waste your evaluation time.</p><ul><li><strong>CRM and pipeline.</strong> Contact records, deal stages, transaction management.</li><li><strong>Lead capture and chatbots.</strong> Site widgets, qualification bots, AI voice agents that answer enquiries.</li><li><strong>Valuation and CMA.</strong> Automated valuation models and pricing tools, which need MLS data feeds we do not touch.</li><li><strong>Title, underwriting, and compliance.</strong> Regulated workflows that belong with specialists.</li></ul><p>Shhots AI is a creative studio for listing marketing. Photos, staging, removal, video, and ads. That is the whole scope, and it is on the <a href="https://shhots.ai/ai-for-real-estate-agents/">feature page</a> in the same words.</p><h3>Will AI replace real estate agents?</h3><p>AI is strong at producing media and drafting text. That is a real slice of the job, maybe a fifth of it, and it is now largely automatable. AI is weak at pricing a property when there are four comparables and none of them fit, at reading what a seller is actually worried about when they say something else, and at keeping a deal alive through a bad inspection.</p><p>What AI replaces is the vendor list. The photo editor, the virtual stager, and the listing videographer the agent used to hire. That is a smaller and far less dramatic story than the headlines, and it is the one that is actually happening.</p><p>The second-order effect is more interesting. When the good treatment stops being rationed, the question stops being “which listings deserve video” and becomes “why would any listing go out without one.”</p><h3>Run one listing through it</h3><p>The evaluation that works is one real property on the day it goes live. A month of tool comparisons tells you less.</p><ol><li><strong>Pick a listing you are about to shoot.</strong> Not an old one. You want to feel it on the day the seller is asking when the photos will be live.</li><li><strong>Run the whole shoot through correction.</strong> Auto Enhance across everything, Sky Replacement on the exteriors, Declutter on the kitchen and bathrooms. Twenty photos costs about $9 and takes about twenty minutes.</li><li><strong>Stage whatever is empty.</strong> Two styles per bare room, then choose. This is the step that changes how the listing performs on a portal thumbnail.</li><li><strong>Build the walkthrough and post it.</strong> Three finished photos, one script, one render. Post it as a Reel the hour the listing goes live, and compare the reach against your last photo-only listing.</li></ol><p>If it does not save you the vendor invoice and the two-day wait, you will know inside an afternoon.</p><h3>FAQ</h3><h4><strong>What is virtual staging?</strong></h4><p>Virtual staging adds furniture, lighting, and decor to a photograph of a room so buyers see a furnished space instead of bare walls. The room is never rebuilt. Only the contents change.</p><h4><strong>How much does AI virtual staging cost?</strong></h4><p>55 credits per generated image on Shhots AI, roughly $0.45 on the $49 Pro plan. Dedicated staging services charge $16 to $75 per image. Physical staging runs $1,500 to $4,000 per listing.</p><h4><strong>Is there a free AI virtual staging tool?</strong></h4><p>Free tools exist and all share the same limits: watermarks, capped resolution, and daily generation caps. Shhots AI has no free tier. Real Estate Studio is on Pro at $49 a month and Scale at $99.</p><h4><strong>Can AI stage a room that already has furniture in it?</strong></h4><p>Yes, two ways. Restyle Room keeps the layout and swaps the pieces. Occupied to Vacant strips the room back to empty first, then stages it fresh.</p><h4><strong>Will the AI change the layout of my room?</strong></h4><p>No. Walls, windows, doors, floors, ceilings, built-ins, and sightlines stay exactly as photographed.</p><h4><strong>Can AI make a real estate video from photos?</strong></h4><p>Yes. Three or four photos from the property gallery become a 15-second narrated walkthrough in 16:9 with an on-camera presenter. The script is written from the property record and stays editable.</p><h4><strong>How long are the walkthrough videos, and what aspect ratio?</strong></h4><p>Fifteen seconds, 16:9. For vertical Reels and TikTok cuts, use <a href="https://shhots.ai/image-to-video/">image to video</a> on individual photos.</p><h4><strong>Do I need to write prompts?</strong></h4><p>No. Every edit runs through a named preset. There is one optional comments field per generation, and it is never required.</p><h4><strong>Does it work with phone photos?</strong></h4><p>Yes, and better if you correct them first. Run the shoot through the correction presets, then build the walkthrough from the corrected versions.</p><h4><strong>Can I use AI-edited photos on the MLS and Zillow?</strong></h4><p>Yes, with the disclosure label your MLS requires. Every plan ships watermark-free at full resolution with a full commercial license.</p><h4><strong>Do I have to disclose AI-edited listing photos?</strong></h4><p>Yes in almost every jurisdiction. Label virtually staged images as virtually staged. NAR Article 12 requires a true picture in advertising, most MLS systems require a label on materially altered photos, and California AB 723 requires a link to the original.</p><h4><strong>Does Shhots AI do lead generation or CRM?</strong></h4><p>No, deliberately. It is a creative studio for listing marketing only.</p><h4><strong>What are the best AI tools for real estate agents?</strong></h4><p>It depends which job you are removing. For listing media, a studio that handles correction, staging, and video on one subscription is the efficient answer. For pipeline and follow-up, that is a CRM category with its own specialists.</p><h4><strong>Does this work for commercial real estate?</strong></h4><p>The photo and video workflows do. Office suites, retail units, and mixed-use floors correct and stage the same way residential rooms do.</p><h3>Where to go next</h3><p>Real Estate Studio is live on the Pro and Scale plans. The <a href="https://shhots.ai/blog/real-estate-studio-launch/">launch announcement</a> has the full feature list, and the <a href="https://shhots.ai/ai-for-real-estate-agents/">AI for real estate agents guide</a> covers how the four jobs fit together on one listing.</p><p>If you want to go straight to the job you care about:</p><ul><li><a href="https://shhots.ai/ai-real-estate-photo-editing/">AI real estate photo editing</a> for correction, sky replacement, and HDR</li><li><a href="https://shhots.ai/ai-virtual-staging/">AI virtual staging</a> for the three modes and eight styles</li><li><a href="https://shhots.ai/ai-real-estate-video-generator/">AI real estate video generator</a> for 15-second walkthroughs</li><li><a href="https://shhots.ai/bulk-real-estate-photo-editing/">Bulk real estate photo editing</a> for brokerages and portals</li><li><a href="https://shhots.ai/blog/ai-ugc-video-for-real-estate/">AI UGC video for real estate</a> for creator-style listing formats</li><li><a href="https://shhots.ai/pricing/">Pricing</a> for the credit maths</li></ul><p>Take one listing. Correct the gallery, stage what is empty, build the walkthrough, and judge it on that.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d80ad63152e0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Shhots AI MCP Server: Generate Ads and Product Photos via ChatGPT & Claude]]></title>
            <link>https://rajatkumardangi.medium.com/shhots-ai-mcp-server-generate-ads-and-product-photos-via-chatgpt-claude-5c001bceb4b1?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/5c001bceb4b1</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[mcp-server]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Wed, 29 Jul 2026 10:32:45 GMT</pubDate>
            <atom:updated>2026-07-29T10:32:45.871Z</atom:updated>
            <content:encoded><![CDATA[<p>Shhots AI’s MCP integration lets you ask your AI assistant to create product photos and ad videos on the fly. In short, an MCP server is a “universal plug” that connects an AI (like Claude or ChatGPT) to external tools.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zR-f3j6nXvfL1XAVnunN9Q.png" /><figcaption>Shhots AI MCP for Ads Generation</figcaption></figure><p>Shhots provides an MCP endpoint (<a href="http://mcp.shhots.ai/mpc/">https://mcp.shhots.ai/mcp/</a>) so you can brief your assistant in natural language and receive finished UGC ads, cinematic videos, or studio-quality product photos in return – all without switching apps.</p><p>For example, you might tell <a href="https://claude.ai/new">Claude</a> or <a href="https://chat.openai.com/">ChatGPT</a>: <em>“Create 5 UGC ad variations for this product photo, 9:16 for TikTok,”</em> or <em>“Turn my landing page into a 15-second cinematic ad.”</em> The assistant handles the rest.</p><h3>What is MCP (Model Context Protocol)?</h3><p>MCP stands for <strong>Model Context Protocol</strong>, an open standard introduced by Anthropic. It defines how AI assistants can <strong>discover and call external tools</strong> in a structured way. Think of MCP as a plug-and-play interface: once a tool (like Shhots) exposes an MCP server, any MCP-capable assistant can use its functions.</p><p>This means Claude Code or ChatGPT doesn’t need custom code for each integration — it can simply query the MCP server’s tool catalog and invoke functions with generated inputs. In practice, when Shhots’ MCP server is connected, your assistant can list the available tools (e.g. image or video generators), call them with your prompts, and return structured results. Nothing happens without your permission: the assistant shows the estimated credit cost before each generation and waits for your approval.</p><h3>Shhots MCP Tools: Images, Video, and Account Data</h3><p>Shhots exposes a set of callable <strong>tools</strong> via MCP (seven or eight in total) that cover creative media and account info. These include:</p><ul><li><strong>Product Image tools</strong>: for example, shhots_create_product_image shoots professional product photos in various styles (studio, lifestyle, editorial, bold) and aspect ratios (1:1, 16:9, 9:16, etc.). It also includes editing and asset management commands.</li><li><strong>Product Video tools</strong>: e.g. shhots_create_product_video generates AI ads and videos (UGC-style, cinematic product demos, animated explainer, unboxing/ASMR, etc.) in ratios like 9:16, 1:1, or 16:9 and lengths of 5–15 seconds (with scripts and visuals approved pre-render).</li><li><strong>Account &amp; History tools (read-only)</strong>: such as shhots_get_account (returns your plan and credit balance), shhots_list_generations (lists recent jobs), and shhots_get_generation (fetches status/results of a specific render).</li></ul><p>In technical terms, the Shhots MCP documentation shows a <strong>“Generate”</strong> category with commands like generate_ugc_ad, generate_cinematic_ad, generate_product_photoshoot, and animate_image, plus <strong>Library/Status</strong> commands (list_generations, get_generation_status) and <strong>Account/Credits</strong> commands (get_credit_balance, get_plan_details, estimate_credit_cost). Together, these let the assistant create watermark-free ads and photoshoots (which land directly in your Shhots workspace) and also check costs and credits ahead of time.</p><h3>How to Connect and Use Shhots MCP</h3><p>Getting started takes under a minute and requires only a Shhots Pro or Scale account (Plans at $49 and $99 per month). To connect:</p><ol><li><strong>Sign up to Shhots</strong> and upgrade to the Pro ($49/mo, 6,000 credits) or Scale ($99/mo, 12,000 credits) plan, if you haven’t already. (MCP is included in these plans.)</li><li><strong>Copy the MCP server URL:</strong> <a href="https://mcp.shhots.ai/mcp/">https://mcp.shhots.ai/mcp/.</a></li><li><strong>In Claude (Web/Desktop/Code)</strong> — open Settings → Connectors → “Add custom connector,” paste the URL, and sign in with your Shhots credentials.</li><li><strong>In ChatGPT (Web/Desktop)</strong> — open Settings → Connectors (enable <em>Developer Mode</em>) → add a new connector, paste the URL, and sign in.</li></ol><p>No API keys or coding are needed. Once connected, your assistant will list Shhots’ tools and you can start asking it to generate content. (Shhots’ own pages summarize this: just one URL bridges your AI to the full Shhots creative engine.)</p><p>After connecting, just speak or type like you would to a creative brief. For example: “<em>Generate a 1-minute product reel for this bike helmet showing morning commute scenes</em>,” or “<em>Animate this product image with a rotating 360° spin and holiday background</em>.” The assistant uses the MCP tools to carry out the request. You will see a confirmation dialog with the <strong>credit cost estimate</strong> (drawing from your monthly pool), and after you approve it, Shhots returns the final asset.</p><h3>Outputs, Credits and Cost</h3><p>All Shhots MCP outputs are <strong>watermark-free and come with a commercial license</strong>. As Shhots notes, the finished assets land in your workspace “watermark-free” so you can use them in any campaign. Credit usage follows the same rules as the app: for example, each AI-generated product photo costs <strong>11 credits plus 1 extra credit</strong> for the initial analysis. A 10-second product video ad costs on the order of <strong>550–600 credits</strong> (15s videos ~850 credits). You can verify costs ahead of time with the estimate_credit_cost tool.</p><p>Because Shhots MCP uses your existing plan credits, there’s no surprise billing — generation cost is shown first and must be approved before running. (And yes: one of the account tools, get_credit_balance, lets your AI assistant tell you how many credits remain.) In short, if you have 6,000–12,000 credits a month on Pro/Scale, you can pump out dozens of images or a handful of video ads each billing cycle, all managed from chat. For context, even the Starter plan ($19/mo) can do basic campaigns, but only Pro/Scale include MCP.</p><h3>Examples of Generated Content</h3><p>Imagine the time and money saved: a $5,000 photoshoot is now a prompt away. With Shhots MCP, just one product photo upload can yield a library of styled shots and ad clips. For example, a single white-background shampoo bottle could produce studio-style product shots in 1:1, 16:9, or 9:16 formats, plus animated UGC-style TikTok ads — all without hiring photographers or editors. (See the Shhots case study: they delivered <em>200+ AI images across 3 aspect ratios in 4K</em> within a week.)</p><p>You can also request lifestyle and UGC scenes. In the image above, Shhots generated a realistic spray ad for a perfume can (the model and lighting were AI-created). In other experiments, users have had their assistants create videos like unboxing clips or founder’s messages, simply by describing the scenario. Because the workflow happens in your familiar AI chat, you iterate by conversation instead of menu.</p><h3>Beyond Claude &amp; ChatGPT</h3><p>As of now, Shhots MCP is <em>live</em> for Claude and ChatGPT. (Support for more MCP-capable assistants is coming soon.) Any AI agent that implements Anthropic’s MCP spec — for example, Google’s Gemini or other platforms that add custom connectors — should be able to plug in Shhots in the same way.</p><p>Even without switching tools, your favorite AI agent can become a one-stop creative engine: with Shhots MCP, telling your AI “find a winning ad idea and make it” is actually possible.</p><p><strong>In summary:</strong> Shhots AI’s MCP server embeds its photo/video ad generation engine directly into AI chat. With a single connector and simple prompts, Shopify and Ecommerce brands can go from <em>brief to broadcast-ready creative</em> in minutes.</p><p><strong>Sources:</strong></p><ol><li><a href="https://shhots.ai/mcp/">https://shhots.ai/mcp/</a></li><li><a href="https://shhots.ai/blog/shhots-ai-mcp-server-for-claude-and-chatgpt/">https://shhots.ai/blog/shhots-ai-mcp-server-for-claude-and-chatgpt/</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5c001bceb4b1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Making video using claude at 10x speed and 1/10th token]]></title>
            <link>https://rajatkumardangi.medium.com/how-i-make-videos-with-remotion-and-claude-my-exact-6-step-process-db0632086b1a?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/db0632086b1a</guid>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[ai-video-generator]]></category>
            <category><![CDATA[remotion]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Mon, 13 Jul 2026 13:43:58 GMT</pubDate>
            <atom:updated>2026-07-13T13:50:51.382Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aPJTvJjdQpxP6EDxidzh8A.png" /></figure><p>I recently figured out making videos in code.</p><p>Remotion lets you build a video as a React component, which means the whole thing becomes text, logic, and assets that you can version, reuse, and regenerate on demand. Claude writes that code.</p><p>This blog expalins my process, that I arrived at after a few trial and error. I use it to produce videos for Shhots, and I have run it enough times that the steps below are the actual sequence that I used.</p><p>Here’s a video that I made recently for <a href="https://shhots.ai/">shhots.ai</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DhFNaq3ZE_dI&amp;type=text%2Fhtml&amp;schema=youtube&amp;display_name=YouTube&amp;src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FhFNaq3ZE_dI" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/b569842b4d30bfb1d98c226755979949/href">https://medium.com/media/b569842b4d30bfb1d98c226755979949/href</a></iframe><p>A quick note before the steps. <a href="https://www.remotion.dev/">Remotion</a> is a framework for making videos with React. Every frame is rendered from code, so animations, text, timing, and asset placement are all defined programmatically. If you can describe a video precisely, you can build it. <a href="https://claude.ai/">Claude</a> is good at turning a precise description into that code. The six steps are really about getting to a precise description, then handing it off cleanly.</p><h3>Step 1: Turn a reference video into a Remotion prompt</h3><p>I never start from a blank page. I start from a video I already like.</p><p>I give Claude a reference video and ask it to break the video down before it writes anything. The goal here is analysis, not code. I want Claude to look at structure: how long each shot holds, how text enters and exits, how the pacing moves, what the transitions are, where the motion comes from, and what the color and rhythm feel like.</p><p>The prompt looks like this:</p><blockquote>Here is a reference video. Watch it and describe its structure in detail. Go shot by shot. For each section, tell me the timing in seconds, the transition in and out, how text animates, the motion of any elements, the pacing, and the color treatment. Once you have described it, write a Remotion prompt that would recreate this video’s structure and style.</blockquote><p>The output of this step is a first-draft Remotion prompt that captures the skeleton of a video I already know works. Copying a proven structure is the fastest way to a good result. Originality comes later, in the assets and the message.</p><h3>Step 2: Bend the prompt to your branding</h3><p>The reference video is somebody else’s brand. Mine is not that. So the next step is to make Claude rewrite its own prompt through the lens of my brand.</p><p>I give Claude my branding: the color palette with exact hex values, the fonts, the tone, the logo treatment, and any motion or style rules I follow. Then I ask it to audit the prompt from step one and adjust every visual decision so the final video comes out on-brand when Remotion renders it.</p><p>The prompt looks like this:</p><blockquote>Here is my brand system. Primary color, secondary color, exact hex values, display font, body font, tone of voice, logo usage, and motion style are all below. Review the Remotion prompt you just wrote. Rewrite it so that when it runs in Remotion, the video matches my branding exactly. Replace generic color and type choices with mine. Flag anything in the reference structure that fights my brand and suggest a fix.</blockquote><p>Now the prompt describes my video, built on a structure I trust. This two-step split matters. If I hand Claude the reference and the branding at the same time, it tends to blend them into something muddy. Separating structure from brand keeps both sharp.</p><h3>Step 3: Ask Claude what assets it needs</h3><p>Here is the step most people skip, and it is the one that saves the most time.</p><p>Before I gather a single file, I ask Claude to tell me exactly what the video will need. Claude reads its own final prompt and returns a complete inventory of assets: screenshots, product images, background videos, font files, icons, logos, and anything else the render depends on.</p><p>The prompt looks like this:</p><blockquote>I am going to prepare a folder with every asset this video needs, then hand it to you and to Remotion. Read the prompt and give me a complete list of required assets. Include screenshots, images, videos, font files, icons, logos, and audio. For each asset, tell me what it is for, roughly what dimensions or format it should be, and where in the video it appears.</blockquote><p>The list is my checklist. I know before I start collecting exactly what I need and why, which means no half-built render that stalls because a font file is missing.</p><h3>Step 4: Build the asset folder</h3><p>This step happens outside Claude. I go down the list from step three and put every asset into one folder.</p><p>I name files clearly and consistently. Not IMG_2931.png, but product-hero.png, logo-white.svg, bg-loop.mp4, font-display.woff2. The names are going to end up inside the prompt and inside the code, so a clean naming scheme now removes an entire class of confusion later. If an asset on the list does not exist yet, this is when I create or source it.</p><p>The output is a single folder that matches the checklist one to one.</p><h3>Step 5: Get the final prompt as a Markdown file</h3><p>With the folder ready, I go back to Claude and close the loop.</p><p>I give it the exact file names I used for every asset. I ask it to fold those real file names into the prompt so the code will reference the right files. I also tell Claude that during the final run I will provide ElevenLabs API keys, so it should plan for music, sound effects, and voiceover where the video calls for them. Then I ask for the whole thing as a Markdown file.</p><p>The prompt looks like this:</p><blockquote>Here are the exact file names in my asset folder, mapped to each asset from your list. Update the prompt to reference these real file names. During the final build I will provide ElevenLabs API keys, so include music, sound effects, and voiceover wherever the video needs them, and note where each audio cue belongs. Give me the final result as a single Markdown file I can hand to Claude Code.</blockquote><p>A Markdown file is the handoff artifact. It travels cleanly into a coding environment, it is easy to read and edit, and it becomes the single source of truth for the build. Everything the render needs is now written down in one place: structure, branding, assets by real file name, and audio intent.</p><h3>Step 6: Build it in Claude Code</h3><p>The last step is the render. I open Claude Code, though any IDE with an agent works.</p><p>I ask it to read the prompt from the Markdown file, I provide the ElevenLabs API keys, I point it at the asset folder, and I ask it to build the video with Remotion. One instruction at the end of this prompt changes the quality of the whole run.</p><p>The prompt looks like this:</p><blockquote>Read the prompt in this Markdown file. Here are my ElevenLabs API keys for audio. All assets are in this folder, referenced by the file names in the prompt. Build the video using Remotion. Ask clarifying questions for further context, and ask me for files or data instead of assuming anything.</blockquote><p>That final line is the most important sentence in the entire process.</p><p>Left to fill gaps on its own, an agent will invent a placeholder, guess a value, or fake an asset, and you find out three renders later. Telling it to ask instead of assume turns a monologue into a working session. It stops, asks a sharp question, waits for the real answer, and builds on solid ground. The videos come out right on the first or second pass because nothing important was guessed.</p><h3>Why this works</h3><p>The whole process is one idea repeated: get to a precise, written specification before any code runs, and never let the agent guess.</p><p>Each step removes a specific kind of guesswork.</p><p>Step one borrows a proven structure so you are not guessing at pacing. Step two locks the brand so you are not guessing at look. Step three lets the prompt define the assets so you are not guessing at what to gather. Step four names files cleanly so the code is not guessing at paths. Step five writes everything into one file so the handoff is not guessing at intent. Step six tells the agent to ask so the render is not guessing at gaps.</p><p>The output is a video defined entirely in code and assets. Want a second version for a different product? Swap the folder and the file names. Want it in another language? Change the voiceover and the on-screen text. The expensive part, the structure and the system, gets built once and reused forever.</p><p>Try it yourself and let me know what kind of results you got :)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db0632086b1a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[$161B Generative AI Industry’s Market Map Deep Dive]]></title>
            <link>https://rajatkumardangi.medium.com/161b-generative-ai-industrys-market-map-deep-dive-b45770974ff0?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/b45770974ff0</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[market-research-reports]]></category>
            <category><![CDATA[ai-ads]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Wed, 20 May 2026 09:26:49 GMT</pubDate>
            <atom:updated>2026-05-20T09:26:49.029Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why no one has won AI Ads Generation yet, and where the real value is!</h4><h4>Key Takeaways</h4><ol><li>Generative AI usage for generating ads, image, videos, and complete marketing campaigns is a large, emerging, and highly competitive industry.</li><li>There are 100+ players in the AI Ads generation across categories, but no clear winner.</li><li>The AI Labs that are building fundamental generative AI models are powering this $161B industry, but they are not the clear winners.</li><li>Detailed notes on how VCs should assess generative AI companies in image, video, and ads.</li><li>These are my learnings based on experiences and a month of building <a href="https://shhots.ai/">Shhots AI (an AI Ads generator)</a>.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E93K68uSaCFvrQrYU4UGKw.png" /></figure><p>In a recent earnings call, Mark Zuckerberg laid out a future where an advertiser uploads a product image and a budget, and Meta’s systems generate, test, and optimise the entire campaign by themselves. The brand shows up with intent. The platform delivers performance. Agencies, production houses, designers, all of it gets compressed into a software layer that runs in the background.</p><p>If that future plays out even halfway, the global advertising industry with over $1 trillion in annual spend, with digital accounting for 73.2 percent of it, gets restructured at its foundation. Creative production becomes the layer where most of the cost compression happens. Studios, freelancers, and agencies start competing with software that ships ten variants in two minutes for less than the cost of a coffee.</p><p>This shift is already underway. Crunchbase data cited in early 2026 put 2025 funding into AI-related video companies at $3.08 billion, up 94.6 percent year on year. The generative AI market sits at roughly $161 billion today, with a CAGR of 28.6 percent and a projected size of $1.3 trillion by 2030. Inside that giant number, the ad creative stack is its own emerging category, and over 100 companies are actively building in it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_U05UUKcDHax4y_J.png" /><figcaption><em>Generative AI Startups Market Map — AI Ads Generator</em></figcaption></figure><p>But we do not have a clear winner in this massive segment where Generative AI is being used for making ads and campaigns.</p><h3>Five lanes, no champion</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ppi86DTr4LkvLFFa.png" /><figcaption><em>AI Generated Ecommerce photos and videos</em></figcaption></figure><p>The generative AI creative market has split into five overlapping lanes. Each one solves a different slice of the workflow, and each is drawing serious capital.</p><h3>Text to Video Generation</h3><p>General AI Video is the foundational layer. Runway, OpenAI’s Sora, Pika, Luma, Kling, Hailuo, Google’s Veo, Adobe Firefly, and Synthesia compete on raw video generation quality. Runway raised $308 million in 2025 at a $3 billion-plus valuation. Synthesia closed a $200 million Series E in 2026 at $4 billion. Pika is at $135 million in total funding. The money is real, but most of these companies are infrastructure. They power workflows. They do not own them.</p><h3>AI UGC Ads</h3><p>AI UGC Ads is the lane that exploded between 2024 and 2025. Creatify announced a $15.5 million Series A in May 2025 with $9 million in ARR. Arcads raised €16 million in late 2025 and now serves 6,000-plus clients. Captions, via its parent Mirage, announced $75 million in growth financing in 2026, with 20 million users and 250 million videos created on the platform. These companies manufacture talking-head avatars, hooks, and scripts at scale. Where they struggle is when the product itself, rather than the actor, is supposed to be the hero of the ad.</p><h3>AI Product Photography</h3><p>AI Product Photography is the catalogue and ecommerce specialist lane. PhotoRoom, Flair, Claid, StudioShot, Pebblely, Pixelcut, and ZMO AI focus on preserving product accuracy, generating lifestyle scenes, and producing PDP-ready imagery. These companies solve the “I don’t want to fly to Bali for a photoshoot” problem for Shopify merchants. They are strong on fidelity. They are weaker on ad workflow, video, and performance.</p><h3>Creative Operating Systems</h3><p>Creative Operating Systems is the most ambitious lane and arguably the most crowded. Jasper, Canva, AdCreative.ai, Predis.ai, Typeface, Writesonic, Simplified, and Ocoya want to be the place teams plan, generate, manage, and publish creative. Canva has the distribution. AdCreative has the performance-first positioning. Jasper has the enterprise muscle. The challenge with this lane is breadth. Owning every workflow often means specialising in none.</p><h3>Product-First Commerce Creative</h3><p>Product-First Commerce Creative is the smallest lane today but, in my view, the most strategically interesting. PhotoRoom, Dressipi, Kittl, Mokker, Pencil, Remove.bg, Zyro, and a handful of newer entrants build AI tools specifically tuned to ecommerce conversion. They speak the language of Shopify, ROAS, and PDPs. They sit closest to the buyer’s wallet.</p><p>Map these against two axes, commerce specificity on the horizontal, closed-loop marketing workflow on the vertical, and something strange shows up. The top-right quadrant, where the most valuable companies should sit, is still mostly empty.</p><h3>The strategic gap</h3><p>Most generative AI creative startups own one loop. The winner of this category will own three.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cLwKuYazMFasJXU3.png" /><figcaption><em>AI Creative Companies Market Positioning</em></figcaption></figure><h3>Creative Loop</h3><p>The creative loop is asset generation. Make an image, a video, a UGC clip. Almost everyone is good enough here now. Models have commoditised faster than anyone expected. Outputs that wowed people in 2023 are basic table stakes in 2026.</p><h3>Truth Loop</h3><p>The truth loop is fidelity. Does the generated asset show the actual product? Does the logo stay legible? Does the packaging match across variants? Does the cap shape, the colour, the claims panel, the variant hierarchy, all survive the generation? This is where almost every platform breaks down. A skincare brand cannot ship a video where the bottle morphs between frames. A supplements brand cannot run ads where the claim text becomes gibberish. Product truth is a hard problem, and most tools have not solved it.</p><h3>Performance loop</h3><p>The performance loop is what happens after the asset is shipped. Did it convert? Which hook worked? Which scene reduced CPA? Which thumbnail pattern lifted CTR? Which variant scaled? This is the closest thing to a real moat in advertising, because the data only gets richer over time, and it is genuinely defensible. The Brief, Creatify, and Zeely have started building this. None has nailed it.</p><p>The company that owns all three loops at once — generation that is fast and beautiful, truth that is reliable across hundreds of outputs, and performance that learns from live spend — becomes the operating system for AI-made advertising. That company does not exist yet. It is the most interesting open opportunity in applied AI right now.</p><h3>Generative AI is about workflow, not just models</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GBC5Xhidup8AuCJt.png" /><figcaption><em>Shhots AI Ads generator dashboard</em></figcaption></figure><p>The instinct of most investors looking at this category is to back model labs. It is a familiar pattern. The model gets better. Capabilities expand. Defensibility comes from the science.</p><p>This category does not work that way. Models in the creative stack are largely commoditised, and the gap between the best and the second-best narrows every quarter. The defensibility lives in the workflow above the model. SKU memory. Brand memory. Performance memory. Team controls. Approval flows. Rights management. Integration with the merchant’s actual stack: Shopify, Meta Ads Manager, Google Ads, Amazon.</p><p>The buyer is not paying for the model. The buyer is paying for time saved, test velocity, predictable output, and lift in ROAS.</p><p>That framing also explains why some of the most interesting recent rounds in this space have gone to application-layer companies rather than model labs. Creatify, Arcads, The Brief, Captions, Higgsfield. None of them is training a frontier video model. All of them are building workflow around models that already exist. Higgsfield in particular is a useful proof point, Reuters reported 2026 funding at a $1.3 billion valuation with a $200 million annualised revenue run-rate, and 85 percent of users came from social-media marketing.</p><p>Workflow is the moat. Model access is a feature.</p><h3>What Indian founders should notice</h3><p>A few things stand out about this market from where I sit, building in India.</p><p>The wallet is global, and the buyer is reachable from anywhere. Shopify merchants in Austin, Berlin, and Bangalore have the same problem and the same willingness to pay. Distribution is online. Onboarding is self-serve. India’s strongest founders in this space do not need a US office to win US revenue, the way an enterprise SaaS founder might have a decade ago.</p><p>The cost structure favours Indian builders. Generative AI workflow companies are render-heavy and engineering-heavy. Indian teams can ship faster and burn less, which matters in a category where foundational models are getting cheaper but the cost of building a defensible workflow on top is still significant.</p><p>The buyer-side market in India is finally maturing. The first wave of Indian D2C brands has graduated past “we need a Mumbai studio for our hero shoot.” Mamaearth, boAt, Sugar, The Whole Truth, Mokobara, and hundreds of smaller brands ship dozens of new ads per week. They have the same creative fatigue as their US counterparts and a sharper price sensitivity. That makes India a strong proving ground for product-first creative tools before scaling globally.</p><p>The regulatory environment is sharpening, and that creates an opening rather than a threat. The European Commission’s AI Act transparency rules kick in from August 2026 for certain synthetic content and deepfakes. The FTC continues to emphasise disclosure for endorsements and influencer marketing. The US Copyright Office has clarified that purely AI-generated outputs are copyrightable only when sufficient human authorship is present. Indian founders building in this category need to bake provenance, disclosure, and rights management into the product from day one. The companies that do this become enterprise-ready faster.</p><h3>What investors should ask</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*d_fnIY0NI9YKHmYi.png" /><figcaption><em>Evaluating AI Ads Generation Landscape</em></figcaption></figure><p>If you are a VC looking at this category, the questions that separate winners from very good demos are not the obvious ones. The obvious questions are about model access, output quality, and prompt engineering. The real questions sit elsewhere.</p><h3>Cost per usable creative</h3><p>Ask about cost per usable creative, not cost per raw generation. A platform with a 30 percent retry rate is twice as expensive as one with 60 percent first-pass usability, even when both list the same per-credit price.</p><p>Ask about product memory. Can the system remember a SKU across hundreds of generations? Does the cap shape, the label, the variant identity persist? If it cannot, the platform is replaceable.</p><h3>Performance feedback</h3><p>Ask about performance feedback. Is there a path to closing the loop between asset and outcome? A platform that only makes assets without learning from how they perform has a shallow moat.</p><h3>Pricing architecture</h3><p>Ask about pricing architecture. One-time credit packs are excellent for acquisition but suppress ARR. The scaled players have all converged on hybrid models: entry-level packs or trials for activation, subscriptions for self-serve use, seats for teams, and API or managed plans for enterprise. Venture-grade revenue lives in the upper tiers.</p><h3>Integration depth</h3><p>Ask about integration depth. Shopify sync, Meta DPA generation, Google ad-account export, Amazon catalogue API. The deeper the integration, the higher the switching cost.</p><p>These are the questions that surface real moats. The rest is theatre.</p><h3>The honest take on Generative AI</h3><p>The generative AI creative stack is the rare category that is simultaneously huge, real, and unwon. The wallet is enormous. The funding is flowing. The technology is good enough. And no single company has yet built the operating system that combines product truth, multi-asset generation, team workflow, and performance feedback in one place.</p><p>That is the gap. Whoever fills it will likely build one of the most valuable AI application-layer companies of this cycle.</p><p>For founders, the playbook is not “build another AI video tool.” It is to pick a specific buyer with a specific pain, anchor the product around the workflow loops that are still weak in the market, and build defensibility through proprietary data, integration depth, and pricing architecture. The category is large enough to support multiple winners, but only across distinct segments. Ecommerce, agencies, enterprise, and creators each have different gravity.</p><p>For investors, the right underwriting model is closer to vertical SaaS than to foundation models. The companies that will compound are the ones whose moats deepen with usage. Truth memory. Performance memory. Workflow lock-in. Integration depth. These are the things that get harder to replicate every quarter, not easier.</p><p>I am building in this space. Shhots.AI is my attempt at a product-first creative engine for Shopify and DTC brands, and the last 2 months of operating in the market have taught me more about how this category really behaves than any analyst report could. I will write separately about what I have learned about pricing, activation, and channel mix.</p><p>For now, the most useful thing I can offer is this: the AI ad market is not won. It is open, it is fast, and the next 18 to 24 months will decide who gets to call themselves the operating system for AI-made advertising.</p><p>That is the bet worth watching.</p><h4><em>Sources:</em></h4><p><em>Crunchbase 2026 AI video funding data; company funding announcements (Runway, Synthesia, HeyGen, Pika, Creatify, Arcads, Captions/Mirage, Higgsfield, Recraft); Grand View Research and Fortune Business Insights generative AI market sizing; public product and pricing pages of platforms cited; European Commission AI Act transparency provisions; US Copyright Office AI authorship guidance; FTC endorsement guidelines.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b45770974ff0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Claude Mythos Actually Means for Indian Banks (And Why I Stopped Sleeping on This)]]></title>
            <link>https://rajatkumardangi.medium.com/what-claude-mythos-actually-means-for-indian-banks-and-why-i-stopped-sleeping-on-this-5c496d90d30f?source=rss-4fff48917921------2</link>
            <guid isPermaLink="false">https://medium.com/p/5c496d90d30f</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[finance-and-banking]]></category>
            <category><![CDATA[security-operation-center]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[claude-mythos]]></category>
            <dc:creator><![CDATA[Rajat Dangi ️]]></dc:creator>
            <pubDate>Thu, 30 Apr 2026 09:19:21 GMT</pubDate>
            <atom:updated>2026-04-30T09:19:21.131Z</atom:updated>
            <content:encoded><![CDATA[<h4>India’s Finance Minister called an emergency meeting over an AI model that has not even been released. Here is what is going on, in plain English.</h4><p>A week ago, I was on a call with the head of security at a mid-sized Indian bank. Halfway through, he stopped and asked, “Have you heard about this Mythos thing? Should I be worried?”</p><p>That is kinds of happening across different banks, different stakeholders, but same question.</p><p>So I am writing this post once, for everyone who is searching for the answers on “what they need to do to be prepared on security front”.</p><p>If you work in banking, fintech, insurance, or just care about how AI is changing cybersecurity in India, this is for you.</p><h3>The meeting that started the panic</h3><p>On April 23, 2026, India’s Finance Minister Nirmala Sitharaman sat down with the heads of major banks. The RBI, NPCI, CERT-In and the IT Minister was there.</p><p>The topic of the meeting was cyber risks becasue of AI and newer models like Claude Mythos.</p><p>The Ministry of Finance later put out a statement calling the threat “unprecedented.” Sitharaman herself said the risk is “as big as war.” She asked for cybersecurity frameworks to become “far more versatile.”</p><p>That is not normal language for a finance minister talking about a tool that is not even publicly available. So what is this thing, and why is everyone losing their mind?</p><h3>What is Claude Mythos?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kqbcrrVDu-w_uGk1ipVV0g.png" /></figure><p>Claude Mythos is the latest AI model from Anthropic, the company that makes Claude. The full name is <a href="https://red.anthropic.com/2026/mythos-preview/">Claude Mythos Preview</a>. The word “preview” matters here, because Anthropic has decided not to release it to the public at all.</p><p>In tests, Mythos was able to:</p><ul><li>Find software bugs that nobody knew were there</li><li>Write code to break in using those bugs</li><li>Stitch many small bugs together into one big attack</li></ul><p>The UK AI Security Institute ran it on hard hacking tests. Mythos scored 73 percent on tasks that are usually solved only by expert security researchers. That is a huge jump from the previous generation of AI models.</p><p>Anthropic looked at these results and decided the model was too powerful to ship to anyone with a credit card. So they did not.</p><h3>Project Glasswing: the safe-mode release</h3><p>Instead of a public launch, <a href="https://www.anthropic.com/glasswing">Anthropic started Project Glasswing</a>.</p><p>Glasswing is a closed group. Around 40 firms have access to Mythos. The list reads like a who’s who of global tech and finance: AWS, Microsoft, Google, Apple, Cisco, NVIDIA, Palo Alto Networks, JPMorgan Chase, the Linux Foundation. They are using Mythos to find and fix bugs in critical software before bad actors get the same power.</p><p>Here is the part that should make every Indian BFSI leader sit up: no Indian company is on the Glasswing list. The Indian government is now in talks with the US and Anthropic to get fair access. NPCI, which runs UPI, has formally asked for early access to test UPI’s own code for day-zero risks.</p><p>The defenders abroad have a head start. Indian banks do not. Yet.</p><h3>Why this hits BFSI harder than any other sector</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e-AbMMMNQegxy71nx70kBA.png" /></figure><p>Banks, fintechs, and insurers sit on top of three weak spots that AI like Mythos is built to exploit.</p><p><strong>Old code.</strong> Many core banking systems in India are 10, 20, or 30 years old. Mythos has already shown it can find bugs that are 16 and 27 years old in widely used software. Old code is its hunting ground.</p><p><strong>Many connections.</strong> A single bank links to dozens of partners, payment networks, KYC vendors, and outsourced services. Each link is a door. AI can map and test every door at once.</p><p><strong>High value.</strong> Money is inside. Customer data is inside. Both are worth a lot. Attackers will spend more effort on banks than on almost any other sector.</p><p>Layer one more thing on top. Indian banks live with strict rules. Every change goes through approvals. Patches take time. Attackers do not have that problem.</p><h3>A 5-minute video version, if you prefer to watch</h3><p>Business Standard put out a clear explainer on this same topic on April 28. If you want the news version of what I covered above, watch this:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FE5L_lKSoYtc%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DE5L_lKSoYtc&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FE5L_lKSoYtc%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/cc9ee381032e829bf2af4c4d6390e926/href">https://medium.com/media/cc9ee381032e829bf2af4c4d6390e926/href</a></iframe><p>It is a good complement to this post. It walks through the government meeting, the Mythos capabilities, and the global context in about five minutes.</p><h3>The shift that almost everyone is missing</h3><p>Most coverage I have seen treats Mythos as just another scary AI story. It is not. The shift is more specific.</p><p>Cybersecurity used to reward deep skill. The best hackers and the best defenders were rare. Both sides spent weeks on each task. That gap is collapsing.</p><p>Mythos can find a serious bug in hours. It can write a working exploit in the same session. What once took a small team a month now takes one AI agent a single day.</p><p>The window from “bug found” to “bank breached” is shrinking.</p><p>This is the part that matters for security teams. Most bank security setups today are slow by design. Alerts come in. People look at them. People decide what to act on. People take the action. That chain works fine for a human attacker. It breaks against an AI one.</p><h3>What I think will actually happen</h3><p>I have been thinking about this for a few weeks. Here is my honest read.</p><p><strong>Short term, attackers win.</strong> They move faster. They do not wait for compliance approvals. They do not need perfect systems. They just need one open door. Once Mythos-class capabilities leak (and a few reports suggest the Mythos model may have already been accessed by people it was not meant for), this gap will widen before it closes.</p><p><strong>Medium term, defenders catch up.</strong> The same AI that finds bugs can fix them. AI can also run defence. AI can sort alerts, investigate threats, and act in real time. The banks that adopt these tools early will pull ahead.</p><p><strong>Long term, this is AI versus AI.</strong> Human analysts will still matter, but they will supervise rather than process. The boring work goes to machines. The big calls stay with people.</p><p>The banks that get this transition right will define the next decade of BFSI security in India. The ones that wait will pay the price in breaches, fines, and lost trust.</p><h3>What Indian banks should actually do</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ASQWDvqLerqZ_Cre3PyXEQ.png" /></figure><p>The government has already given some clear early steps. Real-time threat sharing with CERT-In. Coordination through the Indian Banks’ Association. Stronger cyber talent. Mandatory incident reporting.</p><p>That is the base. Smart banks will go further. Here is what I would do if I ran security at a BFSI firm today.</p><p><strong>1. Cut alert noise before adding more tools.</strong> Most SOCs are drowning. Adding more dashboards adds more noise. Start by figuring out what share of your alerts result in real action. If it is below 5 percent, you have a noise problem. Fix that first.</p><p><strong>2. Move from human-speed to machine-speed response.</strong> If attackers use AI, manual review chains will not keep up. Look at where humans are still doing routine work. Triage. Log review. Tab-switching between tools. Replace those steps with AI agents that act in real time.</p><p><strong>3. Run attacks on yourself, every day.</strong> Annual pentests are not enough now. AI can run thousands of attack paths in parallel. Your defence has to be tested at the same pace. Use red-team AI agents against your own stack.</p><p><strong>4. Track time to respond, not just rate of detection.</strong> Detection without action is useless. Track three numbers each month: time to detect, time to respond, time to contain. Aim to cut all three.</p><p><strong>5. Build for change.</strong> Static rules will age out fast. Ask your vendors how their tools update with new threats. If the answer is “we push new rules every few months,” that is too slow.</p><h3>The bigger picture</h3><p>The risk to Indian BFSI is real. But the same capabilities that make this scary are also the ones that will make defence stronger than ever before. AI does not pick a side. The side that adopts it faster wins.</p><p>I have been talking to a lot of BFSI security leaders for my work at <a href="https://securitypulse.ai/">SecurityPulse AI</a>. The conversations are shifting. A year ago, the request was usually “give me one more dashboard.” Now it is “help me act in real time without drowning my team.” That is the right question to be asking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1sDV1WBwk8uLZfEQkcEigQ.png" /><figcaption>Security Pulse AI</figcaption></figure><p>If you are in banking, fintech, or insurance and you are thinking through what AI-speed defence looks like, I would love to chat. <a href="https://securitypulse.ai/about/">My DMs are open.</a></p><h3>A small ask before you go</h3><p>If this post helped you make sense of what is going on, hit the clap button (Medium lets you clap up to 50 times, by the way). Better yet, share it with one person at your bank or fintech who needs to read it.</p><p>The faster Indian BFSI catches up on this, the better off all of us are.</p><p><em>I write about AI, cybersecurity, and the future of tech. You can find more of my work on Medium or follow me on </em><a href="https://linkedin.com/in/rajatdangi"><em>LinkedIn</em></a><em>.</em></p><p><strong><em>Sources for this post:</em></strong><em> Anthropic’s Glasswing announcement (anthropic.com/glasswing), the Anthropic Frontier Red Team blog on Mythos (red.anthropic.com/2026/mythos-preview), Business Standard, The Indian Express, India Today, NBC News, and The Conversation.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5c496d90d30f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>