We might be looking at the absolute death of manual prompt engineering for repetitive tasks. Usually, getting consistent results from an AI requires pasting the same long instructions every single time, hoping the model doesn’t hallucinate a new method. But I just saw this incredible post from an AI professional that highlights a massive shift in how we interact with autonomous agents. This expert demonstrated a new feature in Manus AI called “Skills,” and it essentially lets you program complex, reusable software agents just by having a conversation.
What caught my attention immediately was the specific use case the creator shared. They built a custom Skill designed to take standard presentation slides or documents and automatically convert them into 9:16 vertical carousels. This isn’t just a simple summary; the workflow utilizes a tool called Nano Banana Pro to format the output visually. The most impressive part is that the author didn’t write a single line of code to build this automation. They simply chatted with the interface, defined the rules, and the system “learned” the behavior. It represents a move away from static prompts toward dynamic, repeatable workflows.
⚙️ The Mechanics: Converting Chat into Reusable Software
A “Skill” in this context is basically a crystallized workflow. Think of it as a Standard Operating Procedure (SOP) that you teach the AI once, and it remembers forever. The original poster explains that this differs significantly from a standard chat session where the context is often lost once you close the window. A Skill packages the entire workflow—including the specific procedure, the necessary tools, the dependencies between steps, and the logic flow—into a singular, reusable unit.
In the example provided, the creator built a Skill that knows exactly how to handle the specific request of “document to carousel” with zero friction. Instead of guiding the AI step-by-step through the conversion process every single time (e.g., “read the doc,” “now summarize it,” “now format it for mobile”), the user defined the logic once. Now, the AI accesses that “package” and executes the entire chain of thought autonomously. This functionality suggests that we are moving into an era where every user becomes a software engineer, using natural language as their coding syntax to build robust internal tools.
📌 Natural Language Creation: Two Paths to Automation
The most daunting part of building automations is usually the setup, but this LinkedIn creator highlighted how surprisingly intuitive this process has become. They outlined two distinct methods for building these Skills, and both are accessible to anyone who can write a sentence. The first method involves proactive planning: you simply tell the AI to “create a Skill for [your task]” and then list out the requirements, tools, and logic explicitly. It’s like dictating a recipe to a chef who instantly memorizes it.
However, the second method the author described is what I find truly fascinating. You can perform a task manually in a chat session—experimenting, tweaking, and getting the result just right—and then, at the very end, simply tell the AI: “Pack this workflow as a Skill.” This “Save Game” approach is brilliant because it allows you to turn a successful experiment into a permanent tool without needing to scope everything out beforehand. It effectively lowers the barrier to entry for creating complex logic chains, meaning you can build your toolkit organically as you work.
✅ Solving the Consistency Crisis
One of the biggest headaches with current Large Language Models is variance. You can give an AI the exact same prompt on Tuesday that you gave it on Monday, yet receive a completely different format or tone in response. The expert points out that Skills are the solution to this problem. By defining a “standard procedure” or an “optimal approach,” you are locking in the methodology.
This matters immensely for business processes. If you are generating monthly reports, client emails, or social media content, you cannot afford to have the AI improvise on the structure every time. The author emphasizes that by packaging the workflow, you ensure that the AI calls upon the exact same logic every time you invoke that Skill. It turns the AI from a creative brainstorming partner into a reliable execution engine. This reliability is what will allow businesses to actually trust AI agents with critical, repetitive operations rather than just using them for one-off creative tasks.
💡 The Human-in-the-Loop Architecture
Perhaps the most sophisticated aspect of this discovery is the ability to define exactly where the automation stops and the human takes over. The post’s author explained that total automation isn’t always the goal; sometimes you need steering power. In their carousel generator example, they didn’t want the AI to just guess the visual aesthetic.
Instead, the creator explicitly instructed the Skill to pause and ask for user input regarding the style and template before proceeding to the final generation. This implies a “manager-worker” relationship between the user and the agent. You are the manager making the high-level executive decisions (style, tone, strategy), and the AI is the worker executing the labor (formatting, drafting, resizing). The author summarized this perfectly: “You make decisions. AI executes.” This hybrid model prevents the AI from running wild and ensures the final output actually aligns with your vision, combining the speed of machines with the taste and judgment of a human.
Potential Challenges to Consider
While this is a significant leap forward, it is important to remember that the output of a Skill is only as good as the logic you provide. If you “pack” a flawed workflow, you are simply automating a mistake. Users will need to be diligent about testing their workflows before relying on them for critical tasks. Additionally, as you build more Skills, managing them and remembering which specific “trigger phrases” activate them could become a new form of administrative overhead.
I am genuinely excited to see how this evolves. The ability to spin up custom tools in seconds without code is huge!
If you want to see the original breakdown of how this carousel generator works, you should definitely check out the full post on LinkedIn.