So, is prompting obsolete—or is its role simply changing as AI becomes more capable?
Lately, I have seen more people saying things like, “Prompting is beginner-level. The real future is Skills, MCP and AI agents that can work for you.”

Because ChatGPT was one of the first AI tools to gain widespread public attention, many people still associate it mainly with typing prompts into a chat window. As newer technologies emerge, ChatGPT is sometimes used as shorthand for an older and supposedly obsolete way of using AI.

I understand why this view is becoming popular. AI is becoming more connected and automated, and typing instructions into a chat window can begin to look old-fashioned. But my own experience has led me to a different conclusion: ChatGPT has grown far beyond a simple prompting interface, while prompting itself has not become irrelevant.

Prompting, Skills and MCP are not different levels where one replaces another. Prompting communicates the objective. A Skill provides a reusable set of instructions for completing a task consistently. Model Context Protocol, or MCP, provides a standard way for AI applications to connect with external tools, systems and information. They perform different roles, and all of them still depend on someone defining what a good result should be.

I Am Using ChatGPT for Real Work

When I first started using AI seriously, I thought of ChatGPT mainly as something that could answer questions, explain difficult topics and help with writing. That has changed.

Recently, I used ChatGPT while rebuilding www.kkecommerce.com. It guided me through nearly the whole process: thinking through the website structure, moving away from Wix, setting up Hostinger and WordPress, troubleshooting technical issues, organising my professional story, and deciding what KK Ecommerce should represent going forward.

ChatGPT could not directly perform every action inside my hosting account or WordPress editor. Often, I still had to show it a screenshot, ask what to do next, follow the instructions and return with the result. That did not make it useless. It made ChatGPT my adviser and problem-solving partner while I worked through an unfamiliar process.

Then I Tried Claude and Hit a Wall

Wanting to test the alternatives for myself, I decided to try Claude too. Connecting it to my Hostinger account was surprisingly easy, and through that connection Claude could access real information from my hosting environment — genuinely impressive, and enough that it would have been easy to conclude Claude was simply the better AI. But when I asked it to go further and edit the site directly inside Elementor, we hit a wall: not because Claude lacked the intelligence, but because, in my setup at that time, the necessary support was not there yet. (I wrote about this experience in more detail in last article.)

That experience taught me something I keep coming back to: an AI may be capable of a task and still be limited by what it can access, what it has permission to do and how much autonomy it has been given.

Where I Think People Get This Wrong

AI discussions often turn into competitions: Which AI is the smartest? Which is best for coding? Which model has the strongest benchmark? Which tool should everyone stop using?

Those questions are understandable, but I think they distract us from a more useful one: What is this AI able to reach and act on right now, for this particular task?

We often compare AI models as though intelligence alone determines what they can accomplish. In practice, the result depends on the model, the tools available to it, the systems it can access, the permissions it has and the quality of the human direction. A highly capable AI without access may be less useful for a particular task than another AI with the right connection.

My own website project is the clearest example: ChatGPT could advise, Claude could connect, but neither was enough on its own without the right access and the right software support behind it.

The limitation was not simply, “AI cannot do this.” It was closer to this: the software around AI is still catching up with what AI can potentially do.

Prompting Is Not Just for Beginners

This is why I do not agree that prompting has become a beginner skill that no longer matters once we start using Skills, MCP or agents.

Connecting AI to a system gives it reach, but reach without clear direction can produce the wrong action more efficiently. A Skill can make a process repeatable, but someone must still define what success looks like. MCP can provide access to information and software, but the human must still communicate the objective, context, constraints and boundaries.

The interface may change, and we may spend less time writing long prompts manually. But the ability to think clearly and define the intended outcome will remain important. Good prompting is not about finding clever words. It is about communicating intent.

Why I Did Not Automate the Whole Article

I could take the automation much further. I could create a Skill in ChatGPT that selects a topic, generates the argument, structures the article, polishes the language and prepares everything for publication. I might only need to review the result, copy it and paste it into my website. That would save time and reduce the cost of producing content.

But I deliberately chose not to do that.

My objective is not simply to publish as many AI-related articles as possible. I am trying to build my own judgment and capability while learning how AI can support a real business. That requires me to remain involved in the thinking, question the result, decide what I genuinely believe and recognise when a convincing argument does not match my experience.

This article did not begin with AI inventing an opinion for me. It began with something I experienced while rebuilding my website. ChatGPT helped me organise that experience, challenge my reasoning and express it more clearly, but the reflection came from the work itself.

If I automate the entire process too early, I may produce articles faster, but I could also outsource the exact ability I am trying to develop: human judgment. Efficiency matters, but it is not the only measure of progress.

Real human reflection makes my learning journey more meaningful and clearer. For me, the best use of AI is not to remove myself from the process. It is to help me think more deeply, communicate more effectively and turn real business experience into something useful for others.

That does not mean I am resisting automation. It means I want to adopt it deliberately, at the right stage of my learning, while paying attention to how quickly the platforms themselves are evolving.

ChatGPT Is Not Standing Still

Another reason I do not think prompting can be dismissed as obsolete is that the platforms around it keep changing.

OpenAI introduced GPT-6 Astra on 3 September 2026, highlighting improvements in computer use, coding, research and complex multi-step professional work. At the time of writing, access was beginning with a limited group of organizations before rolling out more broadly across eligible ChatGPT plans.

Developments like this do not necessarily make prompting obsolete. Instead, they change the kind of prompting we need. As AI becomes more capable of carrying out work, our instructions may shift from detailed step-by-step directions towards defining the goal, supplying the right context, setting boundaries and judging the outcome.

Something complicated today can quickly become a normal built-in feature. A task that currently requires several connected tools may eventually be handled within a single platform. Learning AI, therefore, is not only about chasing the newest tool. It is also about understanding where the technology is heading and deciding when a workaround is worth building versus when it makes more sense to wait for the platform to catch up.

This was the first time I chose to wait for further AI development instead of rushing to build around a temporary limitation. When someone tells me, “Prompting is old now—move on,” I think the better response is: obsolete compared with what, and for which task?

The Future Is Not ChatGPT Versus Claude

The more I experiment, the less interested I become in arguments about whether ChatGPT or Claude is better. I would rather understand how to use each of them well.

ChatGPT has become deeply useful to me for thinking, writing, planning, research, problem-solving and navigating unfamiliar situations. Claude showed me the value of connecting AI directly to a real business system. Each gave me a different kind of capability, and neither experience convinced me that I should abandon the other.

So when people tell me that prompting is merely beginner-level or already obsolete, I disagree. The more useful question is not whether prompting, Skills, MCP or automation is best.

How should I combine prompting, Skills, system access, automation and human judgment for the task I am trying to complete?

That answer will keep changing as the technology develops, and I don’t think I should pretend otherwise. What won’t change is the habit I am trying to build: waiting when a limitation appears temporary, working around it when it does not, and staying involved enough to tell the difference. For now, I am not choosing one AI and rejecting the rest. I am learning how to use the right combination—and documenting what actually happens when I apply these tools to a real business.This continues the story I started in Part 1: Building a Website with AI