Lately, I’ve been looking at Pexo, built by @evanLiaoQ—a project I think is worth discussing. The AI video space is now extremely crowded. Kling, Veo, Seedance, Sora, Runway—every few weeks, model capabilities advance further: image quality, duration, character consistency, and motion realism are all improving rapidly. But after actually using them, you realize a key problem: More models don’t mean making a complete video is any easier. You still need to choose a topic, write a script, break down shots, select a model, craft prompts, generate repeatedly, then handle voiceover, music, subtitles, and editing. Often, generating a single shot takes just minutes—but turning it into a polished, publishable, market-ready video is where most time is spent. That’s exactly what Pexo aims to solve. It’s not building another underlying video model. Instead, it’s creating an AI Video Agent on top of existing models. Simply put: Previously, you manually operated various AI tools. Now, you just tell Pexo: “I want to make a video like this.” The rest is handled by the Agent. It understands your request, reads web pages or assets, organizes information, generates scripts, splits scenes, and calls different image, video, and audio models as needed—then stitches everything into a complete video. So what’s truly interesting about Pexo isn’t how many models it integrates—but how it redefines the interaction with AI video: From “Generate a clip” to “Make me a video.” These two approaches are fundamentally different. The former sells “generation capability”; the latter sells “the outcome.” For example, Pexo can now take a product’s official website URL and have the Agent automatically extract product details to generate a professional Product Video. This is highly practical for SaaS, brands, e-commerce, and crypto projects. Previously, making a product video meant hiring copywriters, designers, editors—even outsourcing an entire team. In the future, you might just paste your website, product screenshots, and key selling points into an Agent—and refine it through conversation. Another feature I’m closely watching is “Mark to Fix.” The real challenge with AI video today isn’t “can it generate?”—it’s “how do you edit it?” What if the logo is too big? The character’s shirt color is wrong? A background needs replacing? The subtitle position is off? Re-generating the entire clip every time is incredibly inefficient. Pexo’s solution: simply circle the area on screen and tell the Agent: “Replace this.” “Leave this unchanged.” “Use this alternative version.” If Pexo can consistently maintain local editing and contextual awareness, AI video will truly evolve from a “lottery tool” into a production tool. Moreover, Pexo isn’t just building a standalone website anymore. They’ve started launching Pexo Skills—integrating with Agent environments like Claude Code, Codex, Cursor, Manus, and OpenClaw. This step is even more significant. In the future, we won’t actively open dozens of AI websites. Instead, we’ll tell our General Agent: “Help me build a product.” “Write my website.” “Then make me a launch video.” And the General Agent will quietly delegate the video production task to specialized agents like Pexo. The General Agent handles task comprehension; Pexo ensures video output is professionally executed. This may be the exact position Pexo is aiming to own. Pexo’s founder, Evan Liao, has long worked in AI video and content generation—with prior experience at Tencent, ByteDance, and Vidu—so this project follows a natural evolution. From Apex Context to today’s Pexo, they’ve consistently emphasized one concept: Context. Models will keep getting stronger and cheaper. But what truly determines whether an Agent is useful isn’t how many models it connects to—it’s whether it can genuinely understand what you mean, remember brand and character context, know which model to invoke for each scenario, and automatically correct failures. This is precisely where Vertical Agents deserve our closest attention. As foundational models become commoditized, competition will shift from “whose model is strongest?” to “who understands tasks best? Who understands users best? Who gets closest to the final result?” What Pexo is doing is essentially betting on this direction: From AI Video Generator → AI Video Agent. I’ll continue watching to see if they can truly connect generation, editing, context awareness, and agent workflows into one seamless end-to-end video production system. @evanLiaoQ