Claude has started color grading, but the bottleneck in video isn't color grading

Greg Isenberg posted a demo about using Claude Opus 5.5, Astra, and DaVinci for video editing, with color grading as a key focus. His claim in the tweet is big: Opus 5.5 will let people who want to be vloggers, make online videos, or build video marketing machines get started without a team or editing skills. According to him, millions of people will enter from the sidelines over the next 24 months.

The comment section didn't follow that direction. One user said editing has gotten cheaper, but knowing what is worth filming has not gotten cheaper; that has always been the harder part. Someone else added: a team and editing were never the reason nobody watched. Another reply was more specific: millions of people will use 4K to say things they never really had anything to say about. It is harsh, but it pulls the question from tools back to content.

The technical details are more worth looking at. Color grading is not just applying a filter. Matching colors across shots, preserving nodes that can be revised, handling LUTs and exports are all engineering problems. Some users asked whether the model is connected through Resolve's scripting API or simulates clicks on the color page. The stability of those two paths is very different. Jeff Bruchado mentioned a key fork: if the grading result is returned as nodes, you can still manually fix a shot that is wrong; if it is baked directly into the export, every revision note turns into another re-render. liveta put it more directly: the model only knows which node values it set, not what the image looks like; only by rendering a still after each change and checking it can you keep cross-shot color drift under control.

There are also pitfalls at the execution layer. Beka Makharoblishvili mentioned that timestamps can be off when cutting video by transcript; in one of his short videos, a word ended 0.14 seconds later than the transcript marked, and the cut point sliced right through the word. Jason Oakes's usage is closer to an off-the-shelf workflow: he used to spend hours editing business videos himself; now he has Grok Bots talk to Opus, and Opus drives DaVinci Resolve to do the boring parts. What is saved is not creativity, but repetitive operations.

So what this wave of tools really compresses is execution cost. For people with a development background, the focus is the interface layer: does the model output a rollback-capable node graph or LUT, or just a pile of parameters; can it read frames, render and check, and handle timestamp errors. For ordinary users, do not interpret zero editing skills as zero barrier. The tools will raise the floor, making more videos watchable, but the ceiling is still topic selection, judgment, and the willingness to shoot a fourth take. In the future there will probably be more videos that are cleanly edited, consistently color graded, and empty in content. Editors do not need to panic for now; the people who should panic first may be those who think the bottleneck is in the software.

Original tweet: https://x.com/gregisenberg/status/2117250560202035509