Directing a product review shoot for Computer Jagat is not something most people would file under prompt engineering. There is a camera, a product, lighting, a presenter reading through talking points. It looks like a media production task, not an AI task. But the discipline underneath it is the exact same discipline that makes a system prompt work or fail, and realizing that changed how I write scripts for everything else.
Two Very Different Jobs Wearing The Same Coat
On paper, writing a video script and writing a system prompt look nothing alike. One is meant to be spoken out loud by a person in front of a camera, the other is meant to be read and executed by a model. But both are attempts to compress a large amount of intent into a sequence that a downstream actor, whether human or model, can follow reliably without losing the point halfway through.
A product review script has to hit a very specific structure to work on camera. Open with something that earns attention in the first few seconds, establish what the product actually is without assuming prior context, walk through features in an order that builds understanding rather than just listing specs, and land on a clear point of view instead of a vague summary. Miss that structure and even a technically accurate script produces a flat, forgettable video, the same way a technically correct but poorly sequenced system prompt produces a technically correct but forgettable model output.
Directing on set adds a layer that prompting work does not usually have, which is watching the script fail in real time and needing to fix it live. A line that read fine on paper sometimes does not survive being spoken out loud, either because the phrasing is awkward, the pacing is wrong, or it assumes a level of viewer knowledge that the earlier part of the video never actually established. That immediate feedback loop, seeing exactly where a written instruction breaks down when a real human tries to execute it, turned out to be the most useful debugging practice I never got from writing prompts alone.
Why Immediate Feedback Changes How You Write
Prompt engineering usually has a slower feedback loop. You write an instruction, run it, read the output, and infer where the instruction was ambiguous. On a shoot, the feedback is instant and unforgiving. A presenter stumbles on a sentence, and you know within seconds that the sentence itself is the problem, not the delivery. There is no room to assume the reader will fill in a gap the way a capable model sometimes will.
That forced a habit I now carry directly into system prompt writing, which is reading instructions out loud before finalizing them, almost as if a person had to perform them. An instruction that sounds clear and complete in your head often reveals a gap the moment you try to say it in a natural sequence. Scripts that had to survive an actual human presenter, in a single take, under time pressure, could not tolerate the same kind of implicit assumptions that a written prompt can sometimes get away with.
Directing also forces a discipline around economy of instruction that a lot of prompting work skips. On set, every extra beat costs time, costs another take, costs the presenter's patience. That pressure to cut anything that is not earning its place in the sequence is exactly the same pressure that produces a tighter, more reliable system prompt, one where every instruction is there because removing it changes the outcome, not because it felt thorough to include.
What Actually Transferred Back
The clearest lesson that came back from directing into prompting work was about ordering. On a shoot, the order information is presented in determines whether the viewer can follow the argument at all. The same is true in a system prompt, where a model reading instructions in the wrong sequence can arrive at a technically valid but practically wrong interpretation of the task, simply because critical context arrived after the point where it was needed.
The other lesson was about earning trust early. A product review that opens weakly loses the viewer before the actual content starts. A system prompt that opens with vague framing before establishing hard constraints gives a model room to drift before the real boundaries are even introduced. Both problems come from the same root cause, treating the opening as a formality instead of as the moment that sets the frame for everything after it.
The Actual Takeaway
Prompting and directing are the same underlying skill wearing different clothes, which is compressing intent into a sequence that a downstream executor, whether that executor has a face and a camera or a context window, can follow without losing the thread. Once that clicked, script writing got sharper because it borrowed rigor from prompting, and prompting got sharper because it borrowed the immediate, unforgiving feedback loop of watching a script fail live on set.
Specific shoot details, client names, and unreleased content remain confidential given ongoing production work. Happy to discuss the general approach to script and content structuring with anyone working across both media production and AI systems through the proper channel.
Written by Mohammad Farhan Habib Faraz
Senior Prompt Engineer and Prompt Team Lead at PowerinAI
www.powerinai.com
