The enterprise AI conversation has spent the past two years focused on the workflows that are obviously susceptible to automation. Coding, document review, customer service, internal search, knowledge management. Those use cases produce clear productivity numbers and clear ROI math, which is why they dominate the headlines and most of the spending.
A quieter conversation is happening in the creative production industries, where the answer to “what does AI replace” is more nuanced. Film, television, theater, advertising, and live entertainment production all involve workflows that AI has changed substantially, alongside workflows it has not changed at all. Understanding why the boundary sits where it sits matters for any enterprise trying to model where AI adoption stops and where human craft persists.
The pattern that emerges from looking at the production sector applies more broadly than the production sector itself.
Where AI has won in creative production
Pre-production is the part of the workflow where generative tools have made the most progress. Concept art, storyboarding, location scouting reference material, costume mood boards, and early-stage visual development are all categories where AI tools have moved from novelty to standard production resource. The productivity gains are real. What used to take a concept artist three days now takes a few hours of prompting and iteration.
Script development has also absorbed AI assistance, though more cautiously. Studios use generative tools for first-pass scene rewrites, dialogue alternatives, and continuity checking. The Writers Guild contract negotiations in 2023 set boundaries around how those tools can be used, but the tools themselves are now part of the development workflow at most major studios.
Post-production has seen the largest measurable impact. AI-driven tools for color grading, audio cleanup, automatic subtitling, visual effects compositing, and rotoscoping have eliminated entire categories of manual work that used to consume thousands of hours per production. The boundary between traditional VFX and AI-assisted VFX has effectively disappeared. Most major productions now use AI tools across the post pipeline whether they advertise it or not.
If the productivity story stopped there, the analogy to other enterprise AI adoption would be straightforward. The story does not stop there.
Where the boundary holds
Practical effects work is where AI adoption hits a hard ceiling. This is the category that covers physical makeup, prosthetic appliances, on-set creature work, mechanical effects, costume construction, and the supporting physical craft that happens between the camera and the performer. Generative AI does not enter this category in any meaningful way, and the reasons are structural rather than temporary.
The first reason is the obvious one. AI generates digital content. Practical effects produce physical objects that exist on a set and interact with performers in real time. A generated image of a prosthetic appliance cannot be worn by an actor. A generated video of stage blood does not produce stage blood on a set. The physical-to-digital boundary is not a technology problem that better models will solve. It is a category boundary that defines what the technology is.
The second reason is more interesting and gets less attention. Practical effects work involves a continuous loop of performer interaction, real-time adjustment, and craft judgment that does not map to a workflow AI can absorb. A makeup artist applying a prosthetic appliance to an actor adjusts for the actor’s specific skin, the lighting on the day, the camera distance, the scene’s emotional context, and a dozen other variables that the artist resolves through experience and trial. The artist’s craft is not a transcribable sequence of steps. It is a real-time response to conditions that are different on every production.
The third reason is supply chain. Practical effects depend on a physical supply chain of specialty materials, applied makeup products, prosthetic compounds, theatrical paints, and tools that have no digital substitute. The materials exist because productions need them, and the suppliers that stock them exist because the demand has been continuous for a century.
The specialty supplier ecosystem
The supplier base for practical effects is one of the more durable specialty industries in American production. It is small by revenue standards. It is geographically concentrated in the cities with large production communities. And it operates on relationships and expertise that no efficiency intervention has been able to displace.
A Manhattan example is Abracadabra NYC, which has operated as a costume and SFX Cosmetics specialty supplier serving film, theater, and television productions for around forty years. The structural detail worth pulling out is not the brand. It is the category. Long-running specialty suppliers like this serve the practical-craft end of production as a baseline business, and the AI tools that have transformed pre-production and post-production have not changed the demand at that end of the pipeline in any noticeable way. Productions still need physical materials, and they still source them through suppliers who know the material and the production community.
What this tells us about enterprise AI adoption more broadly
The production industry is a useful case study because it has had longer than most industries to test where AI tools win and where they stop. The pattern that has emerged is consistent across the productions and the studios that have done the most aggressive adoption. AI replaces or augments digital workflows. AI does not replace or augment workflows that require physical interaction with the real world.
That sounds obvious when stated plainly. It is less obvious in practice because the enterprise AI conversation often treats workflow boundaries as temporary, as if any current limitation is a roadmap item for the next model generation. The production case suggests that some boundaries are not roadmap items. Some boundaries are structural features of the work itself, and they will persist regardless of how capable the next generation of models becomes.
For enterprises modeling their own AI adoption, the lesson worth extracting is the importance of distinguishing between workflows that are temporarily out of AI’s reach and workflows that are structurally outside the category of what AI does. The first category will move. The second will not. Treating the two the same way leads to overspending on tools that cannot solve the problem and underinvesting in the human capacity that the problem actually requires.
Production studios figured that distinction out faster than most enterprises because the costs of getting it wrong were immediate and visible. A studio that bet on generative AI to replace its practical effects team would not have a movie. Other enterprises have more room to be confused about where the boundary sits, and the cost of that confusion is showing up in AI budget reviews across the sector. The studios are not making the same mistake. The supplier ecosystem that serves them is, by all available evidence, in no danger of being disrupted by the tools that are transforming everything else around it.
About The Author
