Everyone is a tech citizen now: why the mental model of an LLM matters more than the tool
By Vishal Kumar, Founder and CEO · Last updated · 7 min read

In short: For years, the goal was creating tech jobs. The bigger goal now is tech citizens: ordinary people who know what AI can do and use it to do more. What separates them from everyone else isn't a favourite tool. It's a mental model of what a large language model (LLM) can do. Once you have that model, your first thought changes from "I don't know how to do this" to "let me describe what I want and see it built".
A couple of people recognised me at Bengaluru Tech Summit this week.
Not from a stage. From my Sunday afternoon classes.
Most weekends, a few hundred working professionals and business leaders join me online to learn AI. I started it as masterclasses on Claude. I'm now turning it into open office hours.
Here's why. I have material for about an hour and a half. After that is when the real learning starts.
Some sessions run for four hours. The questions after the slides are where people actually change how they think.
That shift is what this article is about.
What is a tech citizen?
At a dinner during the summit, I was talking with people about how countries measure progress in technology. The old measure was jobs.
"Early it was about creating tech jobs, but now it's about empowering tech citizens. It's a very different way of looking at it. How do we just help people realise that they can do a lot more?"
A tech job is a role. A tech citizen is anyone, in any role, who knows what today's technology makes possible and uses it.
The accountant who automates her month-end checks. The site manager who talks his inspection notes into an app instead of typing them at night. The teacher who builds a quiz generator for her class over a weekend.
None of them needs to be an engineer. They need to know what's possible.
Why does the mental model matter more than the tool?
I used this comparison at the dinner:
"It's almost like when electricity first came, what could you do? When a cell phone first came, what could you do? Now people are normalised. Now the idea is, what can you do? Most people just don't know."
When electricity arrived, the brand of light bulb didn't matter much. What mattered was understanding that power could now run anything you plugged in.
AI is at that stage now. People argue about which chatbot is best. That's the tempting question.
The better question: do you understand what this kind of system can do for you?
For me, the big shift is simply understanding the mental model of what LLMs and generative AI can do today.
"If you can just imagine something, it can come to life. It's a very different mindset from 'I don't know how to do this, let me search', or 'I can't do it', to a mindset of 'I just need to imagine, and it'll become real.'"
Tools change every few months. A good mental model carries over from one tool to the next.
What is the mental model of an LLM, in plain words?
Here is the version I find most useful. It isn't the technical one.
An LLM is a system trained on a very large amount of text. It's very good at working with language: reading, writing, summarising, comparing, translating, and turning a description into code or a structured document.
From that, a few practical rules follow.
It works from what you give it. It doesn't know your business, your customers or your files unless you share them. Better context in → better work out.
It drafts fast and checks poorly. It will produce a confident answer whether or not it's right. You stay the judge.
Describing is the new skill. If you can explain clearly what you want, who it's for and what good looks like, it can build a first version.
Small pieces add up. One prompt is a draft. A few prompts chained together can be a working tool.
Put together, the loop looks like this:
- Imagine the outcome
- describe it in plain words
- let the AI draft it
- check it against what you know
- refine and repeat
Researchers at Microsoft Research and Carnegie Mellon University found something similar when they surveyed 319 knowledge workers. With generative AI, critical thinking shifts towards checking information, integrating the AI's output and stewarding the task. (Lee et al., CHI 2025) That's the tech citizen's job description in one line.
- 319 workers
- surveyed: with generative AI, critical thinking shifts towards checking, integrating and stewarding (self-reported)
- Source: Lee et al., CHI 2025
| Part | What it means |
|---|---|
| Prompt | What you want, who it is for, what good looks like |
| Context | The real template, email or rules you give it |
| Model | Good at: reading, writing, summarising, comparing, translating, turning a description into code or a document |
| Answer | A fast first draft |
| Check | Sounds confident whether or not it is right; Does not know your business unless you share it; Numbers, names, dates and claims |
What's the difference between users and believers?
At the dinner I put it this way: some people are believers, and some are just users.
A user opens a chatbot when they're stuck, asks one question and closes it.
A believer starts from the assumption that most problems now have a fast path, and goes looking for it.
A nation of believers grows a completely different culture and ecosystem for the next generation. Most countries are not thinking at that level yet.
The same is true inside a company. A team of believers spots the waiting, the retyping and the copy-paste, and fixes it themselves. A team of users waits for IT.
One caution. Believing in what AI can do doesn't mean handing it your thinking. I wrote about that risk separately in Is AI making you think less? The best tech citizens can think with AI and without it.
Why does simplicity decide who becomes a tech citizen?
Most people won't learn a complicated tool. They'll use a simple one.
Someone at the dinner described sitting through a 40-minute talk they didn't understand. My reply was that it's easy to explain complicated things in a complicated way. It's harder to explain a complicated thing in a simple way.
That's also how we build at Quantana. Here's how I described it to someone at the summit:
"Everyone's doing something in AI. We make AI look good. So very strong focus on design. Everyone can build some sort of software, but what we try to do is figure out how do we take a very complex process and make it as simple as possible."
One example: people don't like typing, but they like to talk. So we use speech to text to get data in simply, and then connect the whole workflow behind it. In Australia that's mining compliance. In the US it's insurance.
When the interface is that simple, people who would never call themselves "technical" start using AI every day. That's how a team becomes a team of tech citizens.
How do you build your own LLM mental model?
You don't need a course to start. You need a few honest attempts.
- Pick one annoying task. Something you do every week that involves reading, writing or comparing.
- Describe it out loud first. What goes in, what should come out, who it's for.
- Give the AI real context. Paste the actual template, the actual email, the actual rules.
- Ask for a first version, not a final one. Then tell it what's wrong.
- Check everything that matters. Numbers, names, dates, claims.
- Write down what worked. That note becomes your next prompt, and your mental model grows.
Do this with three different tasks and you'll notice your first thought starts to change.
Try it this week
Pick one task you'd never have thought to give an AI. Describe it in three sentences, and see what comes back.
What's one thing you'd build this month if you knew it could come to life?
Keep reading: AI training for teams · Q Labs · AGI and the agentic operating system
Frequently asked questions
What does "tech citizen" mean?
Anyone, in any role, who understands what current technology, especially AI, can do and uses it to get more done. It's a shift from creating tech jobs to building tech capability in everyone.
What is a mental model of an LLM?
A simple working picture of what a large language model is good at (language: reading, writing, summarising, comparing, drafting code), what it needs (your context) and where it fails (it can be confidently wrong).
Do I need to learn to code to become a tech citizen?
No. Describing clearly what you want is now the more important skill. Code helps, but it isn't the entry ticket.
Which AI tool should I learn first?
Any mainstream assistant is fine. The tool matters less than the habit of describing, checking and refining. That habit carries over when tools change.
How do I get my team to become "believers"?
Start with one real task each, share what worked, and make the tools simple to reach. Small visible wins spread faster than training slides.
Isn't relying on AI risky for our thinking?
It can be. Pair AI use with habits that keep you thinking on your own. See Is AI making you think less?





