Rob Bredow’s TED Talk about artist-driven innovation at ILM is useful because it does not start with general claims about technological change. It starts with production history. Motion-control cameras. CG dinosaurs. De-aging. LED volumes. Generative AI tests. Actual jobs where a tool ran into a story problem, a schedule, a crew, and a director who needed the shot to work.
Tippett is remembered in this conversation because of the line from Jurassic Park, usually paraphrased as “I think I’m extinct.” In Bredow’s telling, the real moment came after an early CG dinosaur test changed the direction of the film. Tippett, who was known for stop-motion and go-motion creature work, saw where things were headed. Spielberg liked the line enough to give a version of it to Dr. Grant.
The familiar retelling makes this sound like the end of practical creature-animation expertise. Yet Tippett’s knowledge of creature performance remained useful when the production adopted different tools.
Applying stop-motion experience to CG
Jurassic Park was originally built around a blend of approaches. Stan Winston’s team created full-scale animatronic dinosaurs for the close physical work. For shots where a dinosaur had to run, leap, or move with more freedom, the production explored stop-motion and go-motion techniques with Tippett’s team.
Then ILM’s CG tests changed the plan.
That part is real. It was a shock. Computer graphics could suddenly do something that had not seemed practical at that scale: put living, breathing synthetic animals into live-action plates with a freedom of movement the older approach could not easily match.
But Tippett’s value was never only “the guy who knows how to move a puppet one frame at a time.”
His value was creature performance: movement, weight, timing, anatomy, rhythm, animal behavior, and the hard-won ability to know when something imaginary feels alive. A tool can interpolate motion. It cannot automatically know why a predator pauses before striking, why a heavy animal carries momentum through a turn, why a head twitch feels alert instead of random, or why one extra beat before the attack makes the audience lean forward.
When the film moved toward CG dinosaurs, that knowledge did not become useless. It became necessary in a new form.
As documented in vfxblog’s oral history of the Dinosaur Input Device, ILM and Tippett Studio developed a way to use stop-motion techniques to create digital animation. The Dinosaur Input Device was essentially a stop-motion armature fitted with sensors so traditional animators could perform dinosaur motion physically and translate that motion into digital data. ILM could then apply the animation to CG models, light it, render it, and integrate it into the final shots.
The final image was digital, but the performance did not come from the software by itself. The dinosaurs worked because digital tools were guided by people who understood physicality.
Tool skill and craft judgment
A lot of people confuse tool skill with craft judgment because, in a stable era, they can look like the same thing.
Tool skill is knowing the software package. It is knowing the camera menu. It is knowing the edit system, the plugin, the render settings, the model, the API, the prompt pattern, the shortcut, the button. Tool skill matters. Nobody wants to sit through a session with someone who cannot operate the machine.
But tool skill is not the whole job.
Craft judgment is knowing what good looks like. It is knowing why something feels wrong before you can explain it cleanly. It is movement, rhythm, story, performance, taste, proportion, timing, audience, cost, feasibility, and context. It is knowing what to ignore. It is knowing when to push back. It is knowing what the tool cannot see.
That distinction becomes important when tools change.
If your value is only that you know how to use an older interface, you are vulnerable. That vulnerability comes from changes in demand, rather than any failure of character. Interfaces change. Workflows collapse. Vendors bundle features. Models absorb steps. A thing that used to require a specialist can become a dropdown, a template, or a prompt.
But if your value is judgment, taste, context, relationships, production reality, and knowing what good is, you have experience to apply. Applying it still takes effort. You still have to learn enough of the new tool to understand what it changes. But you are not starting from zero, because your experience includes more than operating the tool.
AI changes the toolchain
Discussions of AI often overlook that distinction.
Claims that nothing will change and claims that all human work will disappear are both too broad to help someone assess their own position.
AI will replace some tasks. AI will compress some workflows. AI will make some tool-specific skills less valuable. Some people will get displaced. Some skills will lose market value. Some workflows will collapse. That is not a thought experiment. It is already part of the pressure around creative work, production work, software work, research work, and the thousand little coordination jobs that hold projects together.
Developing the ability to evaluate work is a useful response alongside learning to operate new tools.
AI can generate a shot idea. It does not know whether the shot is shootable, ownable, safe, on brand, on budget, or right for the story.
AI can generate a creature image. It does not know whether the creature’s motion will feel alive, whether the anatomy can support the action, or whether the design will still work after rigging, lighting, and editorial get involved.
AI can make a deck. It does not know whether the plan fits the schedule, budget, rights, crew, client requirements, or construction process.
AI can generate code. It does not know the product intent unless someone gives it the context, and it does not prove the behavior is correct without tests, review, and judgment.
AI can summarize notes. It does not know which sentence matters politically or operationally unless the person using it understands the room.
These are questions the people using the tool still need to answer.
Poor choices can be repeated faster and presented more convincingly when generation is cheap. Review needs to account for that.
Using generated material for exploration
The Landis Fields AI test in Bredow’s talk is a good example because Lucasfilm did not present it as a finished movie. It was closer to a moving mood board: one artist using AI tools over a short period to explore what a probe droid might see on a new Star Wars planet.
A moving mood board can help a team react earlier. It can make a tone visible. It can give producers, directors, designers, and supervisors something to reject, combine, redirect, or investigate before significant production spending begins. It can be a faster way to ask, “Is there anything here?”
But the fact that a tool can make something that looks like a scene does not mean production happened.
A real show still needs people who can answer the next questions. What is the story? What is approved? What is cleared? What needs design? What needs to be built? What can be shot? What needs VFX? Who owns the assets? What survives editorial? What does the schedule allow? What happens when the director changes the blocking? What does legal say? What does the crew need to know before call time?
That is why “text prompts alone are not a great way to make a movie” is a useful observation. Generated material still requires decisions about story, design, performance, and production.
Apply what you know to new tools
So what do you do when the tools used in your work are changing?
First, learn the new tool enough to understand what it changes. You do not need to become a zealot. You do need enough practical experience to know what is newly easy, what is newly cheap, what is newly risky, and what everyone is pretending it can do.
Second, preserve the craft underneath the old tool. If you came from photography, do not let camera menus be the only thing you know. Keep studying light, gesture, lens choice, composition, timing, subject, and story. If you came from editing, do not confuse the NLE with pacing, tension, clarity, and performance. If you came from production, do not confuse software with judgment about people, time, money, risk, and approvals.
Third, study why outputs work, not just how they were made. Compare versions. Keep references. Save source files. Look at failures. Ask what changed between the bad version and the good one. Comparison and critique help you explain why one version works better than another.
Fourth, learn enough production reality to know what breaks. A beautiful image can be unshootable. A gorgeous creature can be unriggable. A clever workflow can be impossible under a union agreement, a security policy, a client approval path, or a delivery spec. The person who knows the tool’s practical limits is valuable.
Fifth, use AI to remove bad friction, not to skip judgment. Let it draft, search, summarize, compare, rough out, clean up, and generate disposable options. Then add the human review layer needed to check quality, rights, safety, budget, and story.
Become the person who can evaluate the output, not just generate it.
Judge the method by the work
Familiarity with a tool can become a source of professional pride. So can being among the first to adopt its replacement. Neither establishes that the method suits a particular job.
Tippett’s experience is useful because it shows specific knowledge transferring to a new process: anatomy, weight, timing, and performance. That did not happen without learning and collaboration, and it does not guarantee that every worker will avoid displacement.
The practical question is which parts of your experience remain useful, and what you need to learn to apply them with a different tool.