AI Video Is Everywhere – Making It Actually Good Is Still Hard

01 /The Concept

Even with AI video generation platforms widely available, making a compelling branded video is still hard – even when the idea already exists.

The idea has to become a story first – broken into scenes, each with a detailed description, brought to a satisfying conclusion within a target runtime, without losing pace or the viewer's attention. Then there's the platform problem: no single AI video tool is best at everything. Producing something genuinely high-quality usually means generating different clips across multiple platforms and stitching them together. And then those clips have to be edited – cut, paced, color-graded to a consistent look, audio balanced, effects added. It sounds like polish, but it's critical work that takes real time and expertise.

Skipping the craft and using an AI platform that promises a finished video from a single prompt will produce something – but not something that moves viewers. Without emotional impact, virality doesn't happen and sales don't follow. At least not at the level that was possible.

Absurd has built what it calls a "creative AI agency" that produces viral, conversion-focused videos for brand clients – by combining human creativity with AI production. The agency handles:

- Concept and script development. The creative team learns the client's brand and product, then works through ideas in online sessions until a concept is agreed on.

- Production. The AI generates the script and storyboard, and individual clips are produced on whichever platform is best suited to each scene. Clients can monitor the entire process through a Figma-like digital workspace and leave feedback in real time – so revisions happen during production, not after.

- Edit and finishing. Once the raw material is approved, Absurd's editors bring it to a polished, ready-to-publish state.

Examples of produced work include a video for a customer onboarding service, one for a restaurant management platform, and the agency's own launch video.

Absurd has just entered Y Combinator acceleration, announced recently on the YC blog.

02 /Why It Matters

AI is structurally bad at novelty. Its outputs are a product of patterns found in existing content – so the best it can do is recombine what already exists. It can repeat something good, but it can't invent.

Invention, for now, remains a human capability. And here's hoping it stays that way.

The core trend is therefore clear: build services and platforms that combine human creative invention with AI's ability to turn that invention into a finished product.

And Then ([related review](/review/a-vot-jetim-malo-kto-zanimaetsja-nu-i-ochen-zrja)) just graduated from a16z's Speedrun accelerator with a platform for interactive voice games – dialogues between users and an AI engine. One game opens with the player sitting in a 24-hour diner when someone walks in with a bomb, threatening to detonate it. Police are on the way, but so is the fuse. The only option is to talk him down.

The game's concept was written by a human, because the founder is convinced AI can only produce formulaic plots. But the actual dialogue with each player is generated by the AI on the fly – which it handles very well, producing a unique conversation every time.

Crosby ([related review](/review/bolee-prostaja-model-dlja-sozdanija-perspektivnogo-ii-produkta)) operates in a more serious domain: an AI legal firm that reviews, verifies, and proposes edits to contracts clients receive from counterparties. AI does the initial pass; human lawyers then review, correct, and add nuance. Despite the layered review process, the guaranteed turnaround is four hours. Crosby raised $5.8 million initially and followed with $20 million just last week – still operating in closed beta.

Rocketable ([related review](/review/v-obshhem-sluchae-jeto-poka-fantastika-a-v-chastnom-vozmozhnost-na-milliard)) is pursuing a more dramatic version of the same idea. The startup is building a software holding company designed to run with a single human employee – with AI handling all product development. But the AI isn't inventing anything: Rocketable acquires existing cloud services built by human developers, with at least two years of operating history and between $500K and $2M in annual revenue, and hands them to AI for improvement, optimization, and feature expansion. The startup graduated from Y Combinator this year with $6.5 million raised.

03 /Opportunities

The most popular question about AI right now is whether it will displace people. The more useful framing is: in which areas will AI outperform humans on quality, speed, reliability, and cost? Those are the areas where displacement happens.

Strategically, the opportunity is to find areas where humans remain sharper than AI – or where the economics of human work still beat automation.

The first such area is creative invention. Leave the invention to people and let AI do everything downstream. A human creates the input; AI builds the output. With a possible human check at the end for quality control.

This can extend into a longer collaborative spiral: a person contributes an idea, AI develops it, the person adds a new twist, AI develops that – continuing until the result is something neither could have produced alone.

Enhance Labs ([related review](/review/dva-novyh-scenarija-raboty-proryvnyh-ii-produktov)) is building a voice-based platform explicitly for this kind of collaborative thinking – still in closed testing, having raised its first $1.52 million last month.

The second area is physical work. AI can't yet inhabit a robot that operates reliably across all the varied conditions of the real world – and building that hardware is still prohibitively expensive.

XOi ([related review](/review/tema-v-kotoroj-mozhno-i-horosho-zarabatyvat-i-horosho-prodatsja)) raised $230 million in February on an app for field service technicians. Point a phone camera at a piece of equipment, describe the symptom, and the app identifies the equipment and walks the technician through a step-by-step diagnostic and repair guide. Similar apps for maintenance workers and industrial personnel are multiplying fast.

Two large, strategically durable directions for the AI era – pick one and find the specific domain where you can make it work

Which direction will you go? And more specifically – where?