AI music videos are getting out of hand.

AI UGC Video Experiment 1 I gave four AI video agents the same 15-second product-ad brief. The video credits worked out to $7.36–$14.40 under my plans. The outputs included an extra cup, a misplaced product title and a changed bag shape. Here are the results, what missed the brief, and how I calculated the cost.

AI UGC Video Experiment 1 I gave four AI video agents the same 15-second product-ad brief. The video credits worked out to $7.36–$14.40 under my plans. The outputs included an extra cup, a misplaced product title and a changed bag shape. Here are the results, what missed the brief, and how I calculated the cost.

Real life, baby

I spent three months doing a content and growth internship at an AI video startup, working on its X, Instagram and TikTok accounts. These are my own reflections on what I learned, what I think I did well, and what I still haven't figured out. The part I think I did well was finding content people wanted to watch. Early on, I tried ideas around a football event. I thought playful remixes around it could work on TikTok and Instagram; I was less sure about the same thing on a company's X account. I kept trying, and one of those ideas got people watching and following the account. Later, I also found ideas I could keep developing into a series. So yes, I do think my taste improved: spotting a trend, picking what was worth trying, thinking about where it might work. But then what? I don't think getting content seen meant I'd figured the whole thing out. I learned a lot in those three months, and worked through some problems, but there were others I still couldn't answer. I think of this as a user conversion funnel. I wanted to understand how many potential users reached each stage, and where people were dropping off: 1. Content viewers — how many potential users see the content? 2. Account followers — how many content viewers follow the account? 3. Website visitors — how many viewers or followers visit the website or product page? 4. Registered users — how many website visitors sign up? 5. Users generating their first video — how many registered users generate a first video with the product? 6. Paying users — how many product users become paying users? 7. Retained paying users — how many paying users come back and keep using the product? I'm not saying everyone has to follow an account before visiting the site. People can skip steps or arrive another way. This was just how I separated the results I wanted to understand, because I was using the word "conversion" for things that weren't the same thing. I could see some follows and some clicks through to the product. But did that mean I'd done even those parts really well? I don't think I could say that. The part I felt best about was getting content seen. After the visit, I didn't know enough about registrations, first use, payment or people coming back. That doesn't mean nobody did those things. It means I couldn't tell you what happened, or connect it properly to my work. So I had to look at what I hadn't done enough of. I knew what other products were out there. I'd looked at competitors' social accounts and other creators' content. But actually sitting down and using those products to make a video, start to finish? I may have done a bit of that, but how do I put it... very occasionally, here and there. I hadn't gone through them in a systematic way. Knowing the features of the product I worked on wasn't enough. Where was it better than the alternatives? Where were they better? What would someone actually notice while making a video? If I didn't understand that, how well could I explain why they should choose it? This is why I think product and go-to-market are two sides of the same thing. A great product still needs people to know it exists. And getting attention doesn't do the whole job if the product doesn't meet what someone needs. Both come back to the person you're trying to serve. I don't think I can understand the growth side properly while leaving the product side alone. I also hadn't gone out and talked to real users myself. That's a big piece I was missing. Who is this for? What are they struggling with? Why use AI video at all, and then, why use this particular tool? When I say product-market fit, that's the question I'm trying to get at: does the product really fit a specific group's needs? I hadn't answered it just by knowing the features or looking at content. Maybe someone wants to spend less on making videos. Maybe they need to finish faster, or make more videos than they can now. Those all seem possible to me, but I can't just assume AI delivers that for them. I'd need to know what they were making, what the whole process involved, and whether they actually got something useful at the end. Those are things I need to ask and check. Using the product myself made the making part much more specific. At the beginning, I didn't know how to write prompts, what images or reference videos to provide, how to control camera movement, or how to get the overall mood and style I wanted. I had to start learning. I looked at other people's prompts and tried to understand how they were making things. Even choosing a reference image brought up questions. A face photo, a full-body shot, or front, side and back views? Do I need to show the clothes too? And how do I keep the character looking like the same person? Even now, I wouldn't say I'm great at making AI videos. For me, there was a lot to learn. I still need to hear which parts other people find difficult. That fed into how I thought about useful content for a product's account. My idea was an ongoing mix of four kinds of content: tips and tutorials, work people had made with the product, trends, and explanations of the features. A tutorial could help someone choose a reference image or work on character consistency. A user's video could show them a use they hadn't thought of. A feature explanation could give them something concrete to try. Trends still had a place in bringing people into the conversation. At the time, my feeling was that this kind of content was useful and helped some people take a next step, but didn't take them as far as I'd hoped. I also wondered whether I was paying too little attention to trends. That's how it felt to me, though. I hadn't connected it to registrations or payment, so I can't say that a particular mix improved those outcomes, or that fewer trends caused weaker results. There's another thing I keep thinking about: how clear the direction needs to be. If I don't know what I'm mainly trying to achieve, I don't know where to put my effort. And if what counts as progress keeps changing, it's hard to know whether the work is getting anywhere. I think that makes it harder to get positive feedback from the work, or feel a sense of achievement, even when a lot of effort is going in. So I think a direction needs enough time, and the work still needs quick iterations. Those can go together. Keep the main question clear for a while, then try different content, see what happens, and adjust. I don't mean never changing direction. I mean giving myself a chance to learn something before switching what I'm trying to learn. Take reach. If that's what I'm working on, I'd spend more time finding trends and testing ideas that might spread. If I'm trying to give a particular group a reason to follow, I need to keep being useful to those people, so they have a reason to care about the next post. If I'm looking for potential customers, I need to understand whose problem I'm speaking to. A smaller group with a real need could be more useful for that than a much bigger, less relevant audience. That's something I'd want to test, not a promise. Then there's the brand side. Someone might hear about a product somewhere else and look up its account. They see the work there and think, okay, I understand what this can do, and the quality looks good. I'd want the account to help build that understanding and trust. That can help with getting customers too. It just gives the content a different job from trying to make every individual post spread as far as possible. I don't think reach, followers, brand and customers have to fight each other. But I do need to know what I'm mainly working on at that point, and what I'd look at to decide whether it's working. Otherwise I end up wanting every post to do everything, and I can't even tell what to change when it doesn't. I saw this in my own account too. Starting from zero, I wanted reach, then followers, then a community, and all the other things I thought an account could become. When I focused on trends, I spent a lot of time looking at what was spreading on different platforms and finding things to share or build content around. Over one weekend, my own account went from a little over 200 followers to over 1,000. Reach went up, and followers came with it. That part did work. But then I also wanted it to bring me business conversations and opportunities. I was asking for more outcomes without being clear enough about who I was doing this for, or what I wanted to work on first. I wanted all of it. I don't think the lesson is that I can only ever want one thing. It's that wanting all of it didn't tell me how to do the work. That's why I keep coming back to positioning. Who exactly am I trying to serve? Can I describe that small group clearly enough to know what they need, what I should make for them, and what wouldn't be very useful to them? If I can get specific about those people, I can spend more time understanding them and trying to serve them well. I think that matters for an account, for finding customers, and for the product itself. So that's where I'm at after three months. I think I got better at finding content people want to watch, and I still have work to do on what comes after that. Some problems I've worked through. Some I haven't. I want to understand the people making those choices better, and I can't get all those answers just by looking at content. If you made it this far, let's connect. I'd love to swap notes with people building AI products or working on their growth.

花费16天，我用 Opus 5.5 将小王子的故事制作成了3D剧情游戏 结合Tripo+Blender，Opus 5.5 将小王子故事中的每一个星球变成了可探索的3D场景 希望大家能支持一手，后续会开源整个游戏，以及详细制作流程 Welcome to The Little Prince: Across the Stars✨：littleprince.dev/

3.2M impressions. Fewer than 100 follows. Two months on X taught me that reach and the right audience are two different things.

Gemini 4 Argon is here. It’s available globally today for developers and enterprises in the paid tier of the Gemini API and the Gemini platform, for Ultra subscribers in Google Antigravity, and Ultra and Pro subscribers in the Gemini app and Search. Hope you all enjoy it!

Higgsfield CEO @alexmashrabov admitted they'll NEVER win the $20 subscription market. They still went $1M → $1B in 18 months. A customer went $99/mo → $6M/yr in 6 months. My breakdown of how they got there, and who's actually paying for AI video 👇