fbpx
DigitalFeaturedMarketingOpinion

Stop prompting, start engineering: Why Yann LeCun is right and your agency is wrong

Katch International's Zara Kennedy makes the case for creative technologists who understand pipelines, nodes and model weights. "Agencies must stop treating AI like a magic wand; start treating it as an engineering discipline."

By Zara Kennedy, Vice President of Rotaract Club of Dubai Cosmopolitan, and Junior Account Manager, Katch International on prompting and engineering AIBy Zara Kennedy, Vice President of Rotaract Club of Dubai Cosmopolitan, and Junior Account Manager, Katch International

There is a specific kind of exhaustion that every creative in the Middle East is currently feeling. It’s the fatigue of the ‘re-roll.’

You know the scenario. You type a prompt into Midjourney for a client pitch: “A futuristic Riyadh skyline, hyper realistic, golden hour.” The result is breathtaking. But then the client offers feedback: “Love it. Just add more trees to the streets to match the Saudi Green Initiative.”

You change one word in the prompt. You hit enter. Suddenly, the trees are there, but the lighting is flat, the composition has shifted, the skyline now looks more like Tokyo than Riyadh, and you have lost the magic.

For the last two years, we have been sold a lie. We were told that prompt engineering was the skill of the future. In reality, we’ve just been pulling the lever on a very expensive slot machine and hoping for a jackpot.

This week, Yann LeCun, Meta’s Chief AI Scientist and a Turing Award winner, validated what many of us have secretly suspected. He argued that current large language models (LLMs) are essentially a dead end for true intelligence.

His reasoning? LLMs don’t understand the physical world. They don’t understand gravity, permanence, or logic. They are probabilistic engines predicting the next pixel or word. As LeCun famously put it, they are “dumber than a house cat.”

And that is exactly why your agency is struggling to monetise AI. You cannot build a consistent brand identity on a probability machine.

McKinsey’s State of AI 2025 report highlights a massive profit gap. While 88 per cent of organisations claim to be using AI, only about 30 per cent have successfully scaled it to generate actual enterprise value.

The Middle East is full of agencies in that 88 per cent bracket; experimenting, playing, and posting viral images on LinkedIn. But few are in the 30 cent bracket that trusts these tools enough to run a global campaign on them. Why? Because clients don’t pay for cool images. They pay for consistency, control, and brand safety.

So, how do we fix it? We don’t abandon prompting, but we acknowledge that prompting is not enough. We must supplement it with engineering.

While the mainstream media obsesses over chatbots, the real revolution is happening in the developer trenches with tools like ComfyUI, Stable Diffusion (SDXL), and ControlNet. I’ve spent recent months diving into these node-based workflows, and let me tell you; this is visual architecture.

In a ComfyUI workflow, we don’t ask the AI to imagine a product placement. We use ControlNet to mathematically map the geometry of the product and force the model to render pixels only within those constraints. We don’t hope for the right brand colours; we use IP Adapters to inject specific assets directly into the neural network.

We are moving from an era of generative AI (guessing) to agentic AI (reasoning).

Just look at Google. This week’s release of Gemini 3 and the Google Antigravity platform proves that the big players are done with simple chatbots. Antigravity allows for agents that can reason, test their own output, and correct errors before a human ever sees them. It’s a shift from ‘make this’ to ‘solve this.’

For the creative industry in the region, the writing is on the wall. The era of the prompt engineer, the person who is good at whispering to a chatbot, is already obsolete.

To lead in this market, agencies must stop treating AI like a magic wand; start treating it as an engineering discipline. We need creative technologists who understand pipelines, nodes and model weights.

  • Can your workflow generate the same character in 50 different poses without hallucinating?
  • Can you localise a campaign for KSA, UAE, and Egypt instantly while keeping the product strict to brand guidelines?

If the answer is no, you are still playing slots.

Yann LeCun is right. The chat era is hitting a wall. But that doesn’t mean we stop building. It means we stop gambling with our creative output and start engineering the architecture that controls it.

By Zara Kennedy, Vice President of Rotaract Club of Dubai Cosmopolitan, and Junior Account Manager, Katch International