Brivael Le Pogam(@brivael):I did almost 1 billion views with text. Now I'm going to do the same with video. Here's what I've learned so far. I spent 2 years trying to automate content with AI, especially video. I tried everything. The science behind hooks. Systems that ingest content that already performed and reverse-engineer why. Pipelines that generate, score, iterate. Prompt chains on prompt chains. Etc etc. The reality: When there's no human in the loop, it always sounds flat. It always reads as AI-generated. Every time. You can feel it in the first 3 seconds. The structure is right, the hook "follows the rules," the pacing is optimized, and it's still dead. Because there's no one behind it. No angle, no conviction, no taste. What actually works, and what I do on my own account, is completely different: Use AI to compress your time. Not to replace your judgment. The video I posted the other day did 5 million views. I spent zero time on the editing. And for the writing, I didn't prompt my way to a script. I recorded voice notes. My raw thinking, my angle, my taste. Then I turned those into the script with AI. That's the whole trick. The part that comes from you stays yours. The part that's pure execution gets compressed to near zero. The mistake everyone makes is trying to automate the wrong half. They automate the thinking and keep doing the editing by hand. It should be the exact opposite. Your taste is the scarce asset. The execution is the commodity. AI is an incredible multiplier. But zero times anything is still zero. If there's no one in the loop, there's nothing to multiply. Keep the human signal. Delete the human labor.

2026.10.28 09:42

I did almost 1 billion views with text. Now I'm going to do the same with video. Here's what I've learned so far. I spent 2 years trying to automate content with AI, especially video. I tried everything. The science behind hooks. Systems that ingest content that already performed and reverse-engineer why. Pipelines that generate, score, iterate. Prompt chains on prompt chains. Etc etc. The reality: When there's no human in the loop, it always sounds flat. It always reads as AI-generated. Every time. You can feel it in the first 3 seconds. The structure is right, the hook "follows the rules," the pacing is optimized, and it's still dead. Because there's no one behind it. No angle, no conviction, no taste. What actually works, and what I do on my own account, is completely different: Use AI to compress your time. Not to replace your judgment. The video I posted the other day did 5 million views. I spent zero time on the editing. And for the writing, I didn't prompt my way to a script. I recorded voice notes. My raw thinking, my angle, my taste. Then I turned those into the script with AI. That's the whole trick. The part that comes from you stays yours. The part that's pure execution gets compressed to near zero. The mistake everyone makes is trying to automate the wrong half. They automate the thinking and keep doing the editing by hand. It should be the exact opposite. Your taste is the scarce asset. The execution is the commodity. AI is an incredible multiplier. But zero times anything is still zero. If there's no one in the loop, there's nothing to multiply. Keep the human signal. Delete the human labor.

Brivael Le Pogam@brivael

2026.10.28 09:29

There's a huge debate right now around AI video agents. Which model wins, which workflow, whether editors are going away, whether video editing software is dead. I think most people are looking in the wrong place. Here are my two axioms. Axiom 1: everything that can be expressed as data will get cracked. Or already has been. Editing, captions, b-roll, pacing, voice, avatars, color grading. The moment a craft can be described, measured and labeled, it ends up inside a model. It's not a question of if. It's a question of months. Betting your company on "we make the best captions on the market" is betting against the curve. Axiom 2: the real problem today isn't generation anymore. It's ontology. The models are here. The tools are here. The problem is how you orchestrate them to produce a video that's actually good, not just technically correct. An ontology, concretely, is three things. → How the elements of a video are expressed, stored and passed to models. A scene, a shot, a hook, a brand, a tone, a character. If an agent can't read your video as a structure, it can't work on it. It can only regenerate it. → Having the right system of record. The place where the truth lives: your assets, your brand guidelines, your characters, your history, what performed and what flopped. Without it, every agent starts from zero on every run. → Making sure users don't reinvent the wheel locally every single time. Right now everyone is hacking together their own prompts, templates and workflows in their corner. That's knowledge dying on a hard drive. Good ontology: agents understand what they're manipulating. Bad ontology: you have the best models in the world and you still ship generic videos. And this is where it gets interesting for the future of SaaS, especially in AI video editing. Tomorrow's SaaS isn't an interface full of buttons. Agents don't need buttons. It's two things. A system of record that agents can work with. A base where the video, the brand, the characters and the style exist in a form any agent can read, modify and compose. The ability to build the last 5 to 10%. The first 90% is going to be commoditized, that's axiom 1. What's left is taste. Singularity. The thing that makes your video look like you, and not like the average of the internet. And that last 5 to 10% is an infinite problem. There will never be a "solved." Taste moves, formats move, audiences get bored. Every time the average goes up, the gap that matters shifts. That's exactly why it's where you should build. Last piece: plug all of it into other bots. Your video agent won't live alone. It needs to talk to your marketing agent, your CRM, your distribution agent, your analytics. Video becomes a living object in a graph of agents, not an .mp4 you export and forget. To sum up: Models will crack everything, so value leaves generation. Value moves to ontology: how you represent, store and share video knowledge. The winning SaaS is an agent-native system of record, tools for the last 5 to 10% of taste, and native connection to the rest of the agent ecosystem. That's exactly what we're building at Argil.

커뮤니티로 전달