The era of repetitive prompting might finally be drawing to a close. We have long needed a reliable method to standardize AI behavior so it acts exactly how we want, without needing a fresh explanation every single time. I just encountered a fascinating breakdown from an AI expert regarding a new feature in Manus AI that addresses this exact friction. The original poster claims this is the best update of the year, which is a bold statement considering we are still in January.
This update centers on the ability to create “Skills” merely by chatting with the interface. To demonstrate the power of this feature, the creator built a specific Skill designed to convert standard slides and documents into 9:16 vertical carousels, utilizing a tool called Nano Banana Pro. The core promise here is moving away from random, one-off interactions toward structured, reliable engineering that anyone can build using natural language.
⚙️ Standardization is the New Efficiency
The most significant takeaway from this analysis is the shift from variability to consistency. Anyone who has worked extensively with Large Language Models knows the frustration of inconsistency; you might ask for a summary today and get a perfect result, only to use the exact same prompt tomorrow and get something completely different. The expert points out that a Skill essentially functions as a saved workflow that solves this specific problem.
By creating a Skill, you are defining a standard procedure or an optimal approach for handling hard problems and repetitive tasks. The post argues that this is vital because it packages the entire workflow—including the specific AI agents and models required—into a single, call-able function. This means you can rely on the output format and quality every time you deploy it. It transforms the AI from a chaotic creative partner into a disciplined employee that follows your specific handbook to the letter.
✍️ Two Intuitive Ways to Build
I found the breakdown of the creation process particularly illuminating because it offers flexibility depending on how you like to work. The author details two distinct methods for building these assets, both of which lower the barrier to entry for non-technical users. The first method is proactive and instructional. You simply tell Manus to “create a Skill for [your task]” and then proceed to list out the procedure, the necessary tools, the dependencies, and the logic flow.
The second method, which the expert highlighted, is perhaps even more powerful for rapid prototyping. You can conduct a normal work session with the AI—iterating on a task until you get it right—and then retroactively command it to “Pack this workflow as a Skill.” This allows you to experiment freely, and once you stumble upon a “golden” workflow, you can crystallize it instantly. This capability to turn a successful chat session into a permanent tool is a massive leap forward for productivity.
💡 The Human-in-the-Loop Advantage
Automation often scares people because they fear losing creative control, but this post clarifies that this is not an all-or-nothing proposition. The innovator behind this carousel Skill emphasized that you can decide exactly how much you want to be involved in the process. When they designed their workflow, they didn’t just tell the AI to “make it and finish it.”
Instead, they instructed the system to pause at critical junctures. In the example provided, the AI handles the heavy lifting of processing the document, but it stops to let the user choose the visual style and the template. This creates a hybrid workflow where the human makes the executive decisions, and the AI executes the labor. It ensures the final output matches your taste without requiring you to do the manual formatting. As the author perfectly summarized: You make decisions, and the AI executes.
The Nuance of Logic
While the tool simplifies the technical aspect of coding a workflow, the challenge shifts to your ability to articulate logic. The effectiveness of a Skill is entirely dependent on how clearly you can define the steps and dependencies. If you cannot explain your process to a colleague, you will struggle to explain it to Manus. The potential is huge, but it requires the user to think like a systems architect, even if they aren’t writing code.
This update suggests a future where we spend less time prompting and more time managing a library of custom agents. I highly recommend checking out the original post to see the vertical carousel example in action!