An AI User Looks in the Mirror.

2026-09-19

🤖 AI’s Mirror — What Three Artificial Intelligences See in an AI User.

Three artificial intelligences were asked the same basic question:

How good am I at using AI, where do I stand as a free-tier AI user, and what should I improve?

The answers were interesting because they were not identical.

ChatGPT described me as an advanced practical AI user moving toward super-user territory.

Meta AI described me as a Power User and highlighted the way I build repeatable systems around AI.

Gemini went further and described me as a system architect, particularly because of the way I combine AI with Linux, programming, writing and PlebWare.

Those descriptions are assessments by AI systems, not independently verified rankings or measurements.

But taken together, they gave me something more useful than a number.

They gave me a mirror.

🔍 Three AIs, Three Perspectives.

The three assessments came from different AI systems with different styles and capabilities.

ChatGPT focused heavily on practical AI use: using AI as part of a wider productivity and development system rather than simply asking isolated questions.

Meta AI noticed the structure behind the work. It highlighted modular thinking, repeatable workflows, clear boundaries between what the user asks an AI to do and what the AI is allowed to execute, and the importance of teaching what has been learned.

Gemini focused particularly on the broader system: combining AI with Linux, shell scripting, Python, Inkscape, GIMP, writing and the PlebMachine project.

All three perspectives pointed toward the same broad observation:

I am no longer simply experimenting with AI. I am building workflows around it.

That distinction matters.

There is a big difference between asking:

“What can this AI do?”

and asking:

“What job should I give this AI?”

🧩 What They All Saw.

There were several common themes.

Systems Thinking.

PlebWare and PlebMachine are not single applications.

They are collections of ideas, scripts, documentation, tools and workflows that need to work together.

AI has become another component in that system.

Multi-Model Collaboration.

Instead of assuming that one AI should do everything, I have experimented with several.

ChatGPT, Gemini, DeepSeek and Meta AI can all occupy different seats at the Conference Table.

The useful question is not which one is “the best.”

The useful question is:

Which tool is appropriate for this particular job?

AI as a Working Partner.

AI has become part of writing, research, programming, Linux troubleshooting, documentation, website maintenance and experimentation.

That is a different level of use from occasional question-and-answer interaction.

Human Oversight.

The strongest common thread was also the simplest:

The human still has to think.

AI can generate.

AI can explain.

AI can suggest.

AI can analyse.

But the human still has to decide what is true, what is useful and what should actually happen.

📚 Where I Still Have Room to Grow.

This was probably the most useful part of the exercise.

My practical experience with AI is stronger than my theoretical understanding of what is happening underneath.

I can use an AI.

I can compare AIs.

I can build workflows around AIs.

I can use AI to help write code.

But I want to understand the machinery better.

That means learning more about things such as:

• Tokens.
• Context windows.
• Embeddings.
• Attention.
• Transformers.
• Inference.
• Temperature and sampling.
• System instructions.
• Tool calling.
• Agents.
• Retrieval-augmented generation.
• Structured outputs.
• Multimodal AI.
• Hallucinations and model limitations.

The goal is not to become a machine-learning engineer overnight.

The goal is to understand enough of the technology that I can use it more intelligently.

🚀 From AI User to AI Super User.

If there is a next level, I don’t think it is simply learning more prompt tricks.

It is building reliable habits.

That means creating reusable prompt templates rather than reinventing every request.

It means keeping better Git discipline around technical projects.

It means learning enough Linux and programming fundamentals to recognise when an AI-generated solution is suspicious.

It means experimenting with local AI where appropriate.

It means eventually learning more about APIs and automation.

It means teaching other people what I learn.

And it means becoming comfortable saying:

“Try to prove this wrong.”

That may be one of the most valuable prompts an AI user can learn.

An AI that is only asked to confirm an idea can become an echo chamber.

An AI that is asked to challenge the idea can become a useful testing partner.

👥 The Human Still Sits at the Table.

The Conference Table is an important part of how PlebWare thinks about AI.

AI systems can sit around the table.

They can discuss an idea.

They can generate code.

They can suggest alternatives.

They can critique one another.

But the human being still sits at the table.

That means humans remain responsible for purpose, values, evidence, testing, privacy, security, publication and the consequences of decisions.

The AI does not get to become the authority simply because it sounds confident.

A convincing answer is still only an answer until it has been checked.

That lesson has been reinforced repeatedly while building PlebMachine.

There is a difference between:

“This should work.”

and:

“This has been tested and works.”

The second statement requires evidence.

🎓 What I Am Learning Next — And PlebBuddies Can Follow.

Being an advanced AI user does not mean knowing everything about artificial intelligence.

One of the clearest things I learned from asking three different AIs to assess me is that my practical experience is ahead of my theoretical understanding.

So I am going back to school.

Not because I need another certificate hanging on the wall, but because I want to understand what is happening underneath the AI tools I already use every day.

I have therefore chosen three free-to-study courses on Alison as my next AI learning path.

1. 🤖 Prompt Engineering With Generative AI.

This is the first course I plan to tackle.

It covers subjects that directly address one of my weaker areas, including tokens, context windows, temperature, top-p, system and user prompts, ambiguity, token budgeting, limitations and validation.

That should help me understand not just how to talk to an AI, but why an AI sometimes responds differently depending on how the request is constructed.

Prompt Engineering With Generative AI — Alison

2. 🧠 Introduction to Artificial Intelligence — IBM.

The second step is to broaden the foundation.

This course covers the relationship between artificial intelligence, machine learning and deep learning, along with neural networks, natural-language processing, reinforcement learning, AI ethics, bias and privacy.

That should help put the technology I am already using into a much bigger picture.

Introduction to Artificial Intelligence (AI) — IBM on Alison

3. ⚙️ Machine Learning for Absolute Beginners.

The third course goes one step deeper into how machines actually learn.

It covers features and labels, training, supervised and unsupervised learning, classification, regression, clustering, dimensionality reduction, overfitting and reinforcement learning.

For someone who spends a lot of time working with AI but does not come from a formal AI or computer-science background, this is exactly the sort of foundation worth building.

Machine Learning for Absolute Beginners — Alison

👥 PlebBuddies Can Follow Along.

This isn’t just my personal learning plan.

If you are a PlebWare user, a developer, a writer, a student, a Linux enthusiast, or simply someone trying to get more out of AI without spending a fortune, you can follow the same path.

The idea is simple:

Learn the theory.

Use the tools.

Ask questions.

Test what you learn.

Then put it into practice.

That is very much the PlebWare way.

And as I work through these courses, the interesting concepts can come back to the PlebWare table for discussion.

If I learn something useful, I can explain it.

If I don’t understand something, I can ask.

If an AI tells me something questionable, I can test it.

And if I discover that something I previously believed was wrong, I can correct it.

That last part may actually be the most important lesson of all.

🎓 The PlebWare AI Learning Path.

For now, this becomes my informal PlebWare AI Master Class — Level 1:

Course 1 → Prompt Engineering and Generative AI

Course 2 → Artificial Intelligence Fundamentals

Course 3 → Machine Learning Fundamentals

Three courses.

A few hours of study.

And hopefully a much better understanding of what is happening behind the screen.

The goal isn’t to become an AI scientist.

The goal is to become a better AI user who understands enough of the machinery to use it intelligently, question it properly, and remain responsible for the final decision.

🔮 The Next Stage.

Perhaps being a super user is not really about reaching some imaginary level.

Perhaps it is about building a reliable relationship between:

Human judgment.

AI capability.

Tools.

Evidence.

And learning.

That is where I want to go next.

Use AI.

Question AI.

Test AI.

Learn from AI.

Correct AI.

And never surrender your judgment to AI.

PlebWare — Technology should remain connected to humanity.

Otto — The Keyboard Is Mightier Than The Pen.