turning a week of slack updates into a video briefing is exactly where ai workflows start feeling useful

expedia fell 7% because an AI agent can now plan and book a trip without the user ever visiting a booking platform. the film distribution parallel is not subtle. the aggregator model in film distribution — the platform that brings content to the audience — is the same business as the travel booking platform. the product is not the content or the travel. it is the search, curation, and access infrastructure. netflix, disney+, apple tv+ — all of these are distribution platforms that aggregate content and charge a monthly fee for access. the value is partly the content and partly the search and recommendation layer that helps you find it. AI agents that can understand your specific tastes, watch history, and mood and surface exactly the right film or series across every platform simultaneously are the Muse of the content world. the studios are not the threat. the curation layer is the threat. if an AI agent tells you exactly which platform to subscribe to for exactly the content you want — and that subscription changes month to month — the platform's pricing power falls. this is where the film distribution conversation needs to go and is currently not. the expedia -7% is a preview. the distribution platforms are the next category to watch.

rewatched point break this week alongside man on fire. the thing these two films have in common that i keep coming back to: both directors bigelow and scott made films where the visual style is inseparable from the emotional truth they are trying to express. bigelow's camera in point break is always in relationship with the ocean, the sky, the physical environment. the characters are embedded in the world, not imposed on it. scott's camera in man on fire is fractured and urgent because creasy's psychology is fractured and urgent. in both cases: remove the visual style and you remove the film. the story and the telling cannot be separated. the AI filmmaking question this raises: can a tool execute a visual style that is that specific to a psychological truth? or does it inevitably produce a style that is a generalisation "action thriller aesthetic" rather than "this specific person's experience of loss and violence"? the answer right now is the latter. the tools produce categories, not specificity. the filmmakers who will use AI correctly are the ones who know precisely what they want to say and use the tool only for the parts that do not require that specificity. bigelow and scott knew exactly what they wanted. that is the starting position.

turning agent data into custom apps from a prompt is exactly the kind of ai workflow i want

went back to the odyssey last weekend. fourth time. there is a specific question i keep asking myself about the rewatch habit: at what point does returning to the same films stop being a form of learning and become a form of avoidance? honest answer: i do not think i have crossed that line yet with the odyssey. each watch has surfaced something new. the third watch was about the sound design. the fourth was about the pacing of the dialogue scenes versus the action sequences nolan's ratio is unusually restrained for a film at this scale. the colombiana rewatches are a different thing. that is comfort and calibration. the film is so clean in its construction that returning to it resets my sense of what "enough" means in terms of screen time and narrative density. the odyssey is still teaching me things. that distinction matters. the AI filmmaking tools conversation this week elevenlabs v3, adobe firefly improvements, the accenture studio implementations are all about adding capability. colombiana is the counterargument. you already have everything you need. the question is what you do with it. fourth watch. still holds. fourth watch was worth it.

sandisk is up 600%+ year to date. the AI storage story expressed as a stock price. for film production workflows, the sandisk story is the infrastructure context for why cloud storage pricing has been behaving the way it has. every major production today generates enormous amounts of data. 4K raw footage from multiple cameras, audio multitrack, VFX assets, dailies, rough cuts. a single day of principal photography on a feature film can generate 10-50TB of data depending on the format and the number of cameras. that data needs to go somewhere immediately. frame.io, google cloud, aws all of them rely on NAND flash storage at the infrastructure level. sandisk's 600% run is the market pricing the fact that demand for this storage is growing faster than anyone planned. the practical effect for productions: cloud storage costs for raw media have not fallen the way software pricing has over the past five years. the infrastructure demand is too high for the price to drop. what this means for workflow: the productions that are managing data efficiently proxies instead of full resolution during editorial, tiered storage that moves older assets to cheaper cold storage are paying meaningfully less per project than the ones treating cloud storage as a flat overhead cost. data management is a budget decision, not just a technical one.

there's a weird new AI security problem emerging: AI can help someone find a vulnerability much faster than they could on their own. researchers recently demonstrated this against OpenAI systems through a sanctioned bug bounty. that's useful for defenders. it's also a reminder that the same capabilities don't care which side of the security line you're standing on. AI makes the skilled person faster. that applies to security researchers and attackers.

AI in creative work is getting more interesting when you stop thinking about it as “make me a movie.” the bigger shift is iteration. try a shot. change the lighting. test another framing. rewrite the scene. make a rough version before spending hours building the final version. that's where these tools can change the economics of filmmaking. the filmmaker still makes the creative decisions. AI just makes some of the expensive experiments cheaper to try.

the virtual lab idea is what gets me. turning years of hypothesis work into days could completely change how scientists explore ideas. the researcher still has to decide what’s worth chasing, but ai can make the search space a lot bigger.

being able to customize the agents window background is such a small feature but i love stuff like this. if i’m staring at an agent all day, might as well make the workspace fel like mine.

AI safety is becoming a much more interesting conversation for creative people too. because once AI moves from "generate me an image" to "go do this multi-step task for me," the question changes. you aren't just judging the output anymore. you're trusting the system to make decisions along the way. that's a much bigger creative workflow shift than another prettier image generator.

one thing i keep noticing with ai in creative work: the biggest benefit isn't always making something impossible. it's making iteration cheap. you can try a different shot. change the lighting. rewrite the scene. test another visual idea. before, every version cost time and money. now you can explore more versions before deciding which one deserves the real production budget. that's probably where the creative value gets interesting.

tenzing getting a new trailer has my attention. mount everest stories always have that built-in tension, and this one looks like it could be a great theater watch.

i don't really buy the idea that ai will make filmmaking easier. i think it's going to make making more versions of a film easier. and that's a pretty big difference. a filmmaker can already come up with an idea. the hard part is testing it. does the scene work? does the pacing feel right? would this shot be better from another angle? does the production design actually fit the character? normally, every one of those questions costs time and money to explore. if ai can make rough versions cheap enough that filmmakers can experiment constantly, that's where i think the technology gets interesting. not “press a button and make a movie.” more like: idea → rough version → watch it → hate it → change it → try again. that's already how good creative work happens. ai just has the potential to make the loop much faster.

the ai slowdown debate is getting interesting for filmmakers too. because there's a difference between slowing down frontier model development and slowing down creative workflows that are already useful. a filmmaker using ai for storyboards, concept art or pre-vis doesn't necessarily need a smarter model every month. they need tools that are reliable enough to fit into the process without getting in the way. that's a very different kind of progress.

the part of ai filmmaking i’m most interested in isn’t “make me a movie.” it’s previs, continuity, editing and testing ideas before a production spends real money. adobe is already putting generative video directly into premiere, and that workflow shift feels much more useful to me.

“a person did this slowly, by hand” being the selling point in the ai era is kind of perfect. as a movie nerd, stop motion still has a texture that’s hard to fake.

fully agree instant generation is where this is heading, though the real test for me isn't speed, it's whether real time output holds up shot to shot the way a properly graded, continuity checked edit does. worth digging into teir research post before assuming "real time" means "production ready" 🎬 nitter.net/runwayml/status/209811…

voice agents that can listen while they speak feels like a much bigger deal than just adding voice to an app. if the back-and-forth is as natural as chatgpt, i can see some really interesting creative workflows coming from this.

mistral raised $3.5 billion. the european open-source AI lab is now well-funded enough to compete with the top american labs. for indie film production, mistral matters more than most production people realise. mistral's models are genuinely open — not the "open" that means you can download a version but cannot use it commercially. actually open, with permissive licensing that lets you run the model on your own hardware without ongoing API costs. for a small production company doing script coverage, dialogue assistance, production scheduling, or continuity checking with AI: the difference between paying per API call to openai versus running a mistral model on a local machine is the difference between a monthly cost that scales with usage and a one-time compute cost. the workflow for indie productions that want AI assistance without ongoing subscription fees is getting significantly more viable with mistral at this funding level. the tools being built on top of mistral for creative workflows are still early. but the foundation just got $3.5 billion more solid. the big labs are for the studios. the open models are for everyone else.