The First Wave of Ai in Commercials VFX is pointing to an Integrated Hybrid Workflow
Geoffrey Hinton, the ‘Godfather of Ai’ warned that ‘the idea of it [Ai] wiping out humanity is not silly... it's not science fiction.’ He elaborated that when comparing two intelligent entities, ‘the one with more resources and more intelligence dominates,’ suggesting AI could pose an existential threat to humanity within 20 to 30 years.
Humanity’s Most Spectacular Own Goal?
AI writing Ai? It would create an exponential multiplier of an already vastly superior intelligence. Such a feedback loop would propel artificial intelligence so far beyond us that we cannot even conceive what that intelligence might become. In ‘Everything is Fucked,’ Mark Manson explores a thought-provoking scenario where artificial intelligence becomes so advanced that it essentially becomes god-like, and humans end up worshipping these digital deities. He suggests we might willingly submit to these AI gods, seeing them as all-knowing entities that can solve our existential problems.
Where all this is going is not clear. Joseph Tainter's Law of Diminishing Returns to Complexity suggests that societies evolve with increasing complexity, yielding fewer benefits but using more resources, until you reach the point of collapse. It is clear that this cannot go on forever. But there are many more unknowns: What are the risks of AI in the hands of terrorists? Can it solve our environmental problems? What about mass redundancies; doctors, lawyers, musicians, even master whisky tasters? Will it impact everyone in knowledge work?
While I cannot contribute to the existential debate, I can focus on what's directly in front of me. What practical insights can we draw right now from the first wave of AI in commercial VFX?
It can produce a file but not a project
Ai exists as digital abstraction within our physical human world. It has no executive influence—it simply resides in computers, platforms, and apps. A human could pull the plug (metaphorically at least, and only on a specific process, not on Ai as a whole). In practical terms, this means AI is task-based: it does what we tell it. It cannot yet act unilaterally outside its format constraints; for example, it can't independently access another computer unless we explicitly connect it.
To reach some Matrix-style zenith, all physical boundaries would have to be swept away. Ray Kurzweil, Google's Director of Engineering, has spoken of this "significant bottleneck" that would take decades to implement. We probably have a rough timeline to work with. This temporary limit leaves us humans with the upper hand—we will run the project while Ai remains the tool.
A Shift Toward the Generalist?
However, we will still need to adapt to make room for Ai's arrival. It is widely understood that specialisms have value in today's market, and it pays to be an authority in your niche. The terms ‘specialist’ and ‘generalist’ are problematic, requiring context to define their meaning properly, however we can go with idea that the Flame operator is a Generalist for now.
The Flame operator is a jack of all trades who knows a little about a lot and is therefore equipped to deliver complete projects. Flame ops modify content, assemble the parts that make up an ad, and finalize the whole piece. This skillset could work well with Ai. Instead of competing by specialising in areas such as beauty work, we could step up to manage the vast volumes of content that Ai will create. Let AI handle the tasks like the clean up & beauty work. The online editor of the future might instead become well-versed in every new format—extended reality, real-time rendering, and so forth—and turn this generalist overview into a specialisation itself.
Hype and the reality that Ai must address
David Jones, CEO of The Brandtech Group, predicted that AI would ‘upend the advertising industry in 2025’, suggesting production costs could be slashed by up to 50%. However, I have seen no evidence of this in my practice. When I discovered Ai could write code, I imagined building my website instantly. When I saw AI could write copy, I thought my blog posts would materialise effortlessly. Neither expectation proved true. Ai has improved my output, but it has saved me no time at all. Was David Jones suffering from the illusion we know too well in post-production—that the final 5% of work consumes the most time?
Ai prompts and widgets don’t give you precision control
At the beginning of a project, Ai is at its best. Our early prompts yield beautiful results, and we are off to a flying start. But this luxury of keeping content generic only occurs at the start. If these initial Ai creations could be delivered as-is, advertisers could make dramatic savings by skipping production altogether. But few advertisers can effectively convey their message with generic content. At the end of a project, Ai is actually at its weakest when we need it most. The more specific you get, such as with final amendments, the less control Ai affords us. You may have started with Ai and made superb progress, only to watch it fall apart.
Sometimes ‘Probable Next Words’ are Not Reliable Enough
As three professors at Glasgow University argued, AI's mistakes are better described as "bullshit" rather than hallucinations. A hallucination requires some awareness of truth, but Ai language models simply produce probable next words based on their training data, with no regard for truth. Like a used-car salesman using whatever talking points might sell a car, Ai generates plausible-sounding content without concern for accuracy. This "bullshit" framing helps explain why Ai can produce convincing but factually incorrect content - it's not trying to be truthful, just statistically probable.
Like everything else, Ai falls short of perfection, and this gap cannot be closed
Ai promises to fly us higher, this time to the sun. Oliver Burkeman notes that each new technology promises convenience, but often leads to diminishing returns. Like social media, promising easy connection but delivering loneliness, or email promising efficiency but creating endless overwhelm, each convenience solution seems to create new problems. Our constant reaching for technological fixes can leave us caught in an endless cycle of seeking yet more solutions.
The Emergence of the New Hybrid Workflow
Every production should start with Ai. Once we've exhausted its capabilities, we proceed with our traditional workflow, modifying its output until it's properly integrated into our project. The integration work is substantial. I have produced GenAi content that has rescued projects, but not one piece arrived ready to drop in as final. More than post-processing, the pre-processing elements that guide Ai are highly dependent on traditional skills. To get close to your desired composition and style, you'll likely need to build it to some degree first, in order to show Ai what it needs to do. Nuke's CopyCat works on this principle: first you build, then it replicates. This approach differs significantly from Photoshop's prompt-based generative fill and demands substantial VFX work to set it up. THe payoff is greater precision as you guide Ai to closer to delivering your final shot.
Workflows will continue to evolve dramatically, but this is what we have now. AI remains dependent on existing industry infrastructure. Success will depend on understanding both Ai's capabilities and limitations. We are in a transition period, and Ai workflows are still, at best, experimental.
It's neither the end of work as we know it, nor is it a passing fad.