How I Leverage Free-Tier AI as a Pensioner Builder of PlebWare.

2026-09-22

🤖 Building PlebWare With Free AI — A Pensioner’s Working AI Stack.

Artificial intelligence is often discussed as though serious work with AI requires expensive subscriptions, powerful computers, or a professional technology background.

My experience has been rather different.

I am a pensioner in Johannesburg, South Africa.

I work extra hours to make ends meet, I use second-hand computer hardware, I prefer Free and Open Source Software where practical, and I do not currently pay for AI subscriptions.

Yet I use several AI systems as part of the daily development of PlebWare.

The important word is use.

I am not handing my life over to artificial intelligence.

I am using AI as part of a human-directed workflow.

🧠 The Basic Philosophy — AI Is a Collaboration.

My view is simple.

AI is not a crutch.

AI is not a replacement for thinking.

AI is a collaborator.

Roughly speaking, about 80 percent of the content and direction in my work comes from my own ideas, thoughts, decisions and input.

The AI helps me process those ideas.

It researches.

It explains.

It corrects.

It generates alternatives.

It writes code.

It helps organise information.

It can save me hours of work.

But the human remains the architect.

At the PlebWare Conference Table, the AI is the worker sitting around the virtual table with me.

The human decides what needs to be built.

The AI helps do the work.

That distinction is important.

The problem is not necessarily artificial intelligence.

The problem can be the human being behind the way it is used.

📱 My Free-Tier AI Stack.

I currently work with several free AI services, each fitting into the workflow in a slightly different way.

I deliberately do not expect one AI to do everything.

That would defeat the purpose of having a collaborative system.

🟦 Meta AI — My Mobile First Responder.

Meta AI is particularly useful because I live much of my mobile life inside WhatsApp.

I use Meta AI both through WhatsApp and through the standalone Meta AI application on my phone.

It is useful for:

• Quick research.
• Dictation.
• Correcting grammar and flow.
• Developing ideas while away from the computer.
• Daily Christian devotions.
• Capturing thoughts before I forget them.

I have been writing daily Christian devotions for several years, so having an AI assistant available while I am working from my phone is particularly useful.

There is one amusing limitation.

Speech-to-text does not always understand PlebWare terminology.

Words such as PlebWare, PlebMachine and PlebVox can be transcribed incorrectly.

Names can be even worse.

That means I still have to proofread what the machine heard.

And that is perfectly acceptable.

AI does not remove the need for checking.

It creates another stage in the workflow.

💻 ChatGPT — My Heavy Workshop.

ChatGPT is where I tend to do more extensive writing, editing, research and technical work.

I use it for things such as:

• Writing and rewriting.
• Grammar correction.
• Research.
• Bash scripting.
• Python.
• Lua.
• Linux troubleshooting.
• Documentation.
• PlebWare website work.
• PlebMachine development.
• Structuring complicated ideas.

I don’t actually need AI to teach me how to code.

I have been experimenting with computers since 1982, beginning with machines such as the Commodore VIC-20, and I have spent decades learning programming and web technologies.

What AI gives me is speed.

Something that might take me a day or two to debug manually can sometimes be narrowed down in seconds.

More importantly, I can recognise the code.

I am not simply copying something I don’t understand.

I can look at the generated code, follow the logic and learn from it.

That suits the way I learn.

I do not learn particularly well by rewriting the same thing a thousand times.

I learn by looking.

By listening.

By paying attention.

And by understanding what I see.

🔬 Gemini — The Strategic Researcher.

Gemini has occupied a different position in my workflow.

Over the past year, I have used Gemini extensively for architecture, research, planning and contextual analysis.

It has helped me work on Linux-based systems such as PlebMachine, including technical documentation, shell scripts, Python specifications and automation workflows.

Gemini has also been useful for context-grounded research.

I have used notebooks and structured information to examine technical documentation, historical material and research notes.

It has also acted as a creative sounding board for science-fiction development, character ideas, plot logic and universe building.

Another area has been graphics.

I have used AI assistance to turn ideas involving surrealism, cyberpunk aesthetics and artistic concepts into detailed prompts, SVG concepts and instructions that can then be taken into tools such as Inkscape and GIMP.

In my current workflow, I often treat Gemini as a structural planner and technical researcher.

The information can then move across the Conference Table for further development, editing and verification.

💬 And Then There Is DeepSeek.

I also use DeepSeek, but I am deliberately not making it a major part of this particular comparison.

My experience so far has been more strongly centred on conversational and chat-based work.

One thing I particularly appreciate is that I have not yet encountered the kind of practical conversation limit that sometimes affects other free services.

That makes it useful when I want to keep a long technical or conceptual conversation going.

But this article is primarily about the AI tools I use as part of the broader PlebWare production workflow, so I am concentrating on Meta AI, ChatGPT and Gemini here.

Different tools can have different jobs.

That is one of the main lessons I have learned from working with multiple AI systems.

🏗️ What Am I Actually Building With All This AI?

The answer is PlebWare.

PlebWare is a publishing, learning and creative technology ecosystem built around a simple philosophy:

Technology should adapt to people, rather than forcing people to adapt to technology.

The website is one part of that ecosystem.

PlebMachine is another.

Pleb-Tuition is another.

The community groups are another.

The AI systems are tools that help me build and maintain the whole thing.

🐧 PlebMachine.

PlebMachine is my Linux productivity environment and one of the most technically demanding parts of the project.

It began around the XFCE desktop and continues to explore practical Linux environments and distributions.

The basic idea is to provide a structured working environment in which different workspaces and modes help separate different types of activity.

Writing is different from graphics.

Research is different from programming.

Content creation is different from devotional work.

The computer should help make those transitions easier.

And it is being developed on ordinary hardware.

That matters to me.

I don’t believe useful technology should automatically require expensive hardware.

🌐 PlebWare Is Bigger Than PlebMachine.

The PlebWare website is also a publishing and learning platform.

It contains writing, research, technical material, tutorials, AI-related information, Christian content and creative work.

Pleb-Tuition is intended to help ordinary people learn technology without assuming that they already understand the jargon.

The community side is important too.

I use WhatsApp communities covering areas such as writing, technology and Christian devotion as places where ideas can be discussed and feedback can be gathered.

That feedback can then return to the PlebWare development process.

The cycle becomes:

Idea → AI assistance → Human review → Testing → Publishing → Human feedback → Improvement.

That is much closer to a workshop than a chatbot.

🔑 How I Keep Free AI Useful Instead of Letting It Take Over.

Having several AI systems available does not automatically create a good workflow.

Without discipline, it can actually create chaos.

So I have developed some boundaries.

1. Define the Job Before Giving It to AI.

When working on a technical component, I try to define three things:

Input Contract.

What information, files, variables or state does the component receive?

Execution Boundary.

What is the AI actually allowed to change or do?

Output Contract.

What should the finished result look like?

I also specify what must remain unchanged.

That is particularly important when working on PlebMachine.

Change one component at a time.

Preserve interfaces.

Test the result.

Then move on.

2. Use Four Levels of AI Permission.

I use a simple four-level model.

Level 1 — Observe.

The AI answers questions or explains something.

Level 2 — Assist.

The AI drafts, plans, researches or proposes solutions.

Level 3 — Operate.

The AI may perform an action, but only with explicit permission and within a controlled environment.

Level 4 — Delegate.

The AI may handle a larger workflow only after its reliability has been demonstrated.

Even then, I do not treat sensitive information, finances, account control or important publishing decisions as something to casually hand over.

The more powerful the AI becomes, the more important the boundaries become.

🎙️ My Dictation Workflow.

Mobile AI has also changed the way I write.

I can speak an idea instead of typing the entire thing.

The general process is:

I speak.

The AI transcribes.

The AI corrects grammar and flow.

I read it back.

I correct the mistakes.

Then the material becomes part of the writing process.

That last step matters.

Speech recognition is not perfect.

It can misunderstand technical terms, names and PlebWare terminology.

Therefore, transcription is not publication.

It is raw material.

The human still checks the result.

💰 Why Free-Tier AI Matters to Me.

This isn’t an academic experiment.

It has a practical financial dimension.

I am a pensioner.

Money is tight.

I would love to pay for more AI services and expand what I can do, but paid subscriptions are not always realistic within my budget.

That means the free tiers matter.

They allow me to continue learning and building.

With free AI assistance, I can:

• Write and publish.
• Maintain the PlebWare website.
• Develop PlebMachine.
• Research technical subjects.
• Produce daily devotional material.
• Experiment with graphics.
• Learn programming concepts.
• Teach other people what I discover.

If free AI disappeared completely, I would not stop building.

I would simply go back to doing more of the work the old-fashioned way.

I was coding before today’s AI systems existed.

AI gives me something different.

It gives me leverage.

🧠 Does AI Make People Dumb?

I don’t think the existence of a tool determines whether the person using it becomes smarter or less capable.

A calculator did not make mathematics disappear.

A spellchecker did not make language disappear.

A search engine did not make knowledge disappear.

The important question is whether the person using the tool is still paying attention.

My own grammar has improved through using AI.

When an AI corrects a sentence, I can see what was changed.

When it restructures a paragraph, I can compare the two versions.

When it produces code, I can study the code.

That means the tool can become part of the learning process.

But only if I actually pay attention.

Don’t just accept the answer.

Look at the answer.

Question the answer.

Learn from the answer.

That is the difference between using AI and surrendering your thinking to AI.

👤 The Human Still Owns the Work.

This is probably the most important principle in the entire workflow.

AI can help produce words.

It can help produce code.

It can help produce research.

It can help generate ideas.

But I remain responsible for what I publish.

If something is wrong, saying that an AI generated it does not make the mistake disappear.

If the code fails, the AI does not have to sit at my keyboard fixing it.

If an article contains an error, the reader sees my published article.

The responsibility remains human.

That is why I regard AI as a worker around the virtual Conference Table rather than the person in charge of the meeting.

The human remains the architect.

🚀 What Free AI Has Allowed Me to Do.

When I look at the amount of work I can now accomplish with an ordinary computer, a phone, free software and free AI services, the interesting thing isn’t the artificial intelligence itself.

It is the combination.

Human experience.

Second-hand hardware.

Free and Open Source Software.

Free AI services.

Internet research.

Persistence.

And a willingness to learn.

That combination allows a pensioner working with limited resources to build something far more ambitious than his circumstances might initially suggest.

PlebWare is still evolving.

PlebMachine is still evolving.

My understanding of AI is still evolving.

And that is fine.

A workshop does not have to be finished before you can build something useful inside it.

🔮 The Real Super User Test.

I recently asked several AI systems whether they considered me a super user.

The answers were interesting.

But the more I think about it, the less interested I am in the label.

I am more interested in the method.

A useful AI user does not necessarily need the biggest subscription.

They need to understand the job they are trying to accomplish.

They need to know what information the AI needs.

They need to know what the AI is allowed to do.

They need to check the result.

And they need to learn from the process.

For me, that is the real goal.

Not becoming dependent on AI.

Not proving that I am smarter than an AI.

Not finding one magical prompt.

Instead:

Use AI.

Question AI.

Test AI.

Learn from AI.

Correct AI.

And keep the human at the Conference Table.

PlebWare — Technology should remain connected to humanity.

Otto Brinkmeier — The Keyboard Is Mightier Than The Pen.