# Three prerequisites for the AI-pocalypse

*13 Sept 2026* · Stella Ellervee

Setting up ridiculously powerful systems is almost trivially simple. If only you had 3 competencies.

---

Setting up ridiculously powerful systems that solve decades-old problems which
burdened people daily is doable in minutes => almost trivially simple. But it's
not trivially simple for every end-user. I think there are 3 prerequisites for
remaining adaptable in the current & coming AI winds of change.

## 1. Computer — become tech-savvy

> **Why you're doomed without this:** you won't be fast enough in the
> cooperation tasks. Without decent speed in this realm you won't win in time —
> the tools will provide too much friction & pen & paper is more worthwhile.

### What you need

It's not about mastery of one tool but comfort across the category. I don't
mean being a world-championship-level expert in Excel or any one piece of
software. I mean a solid baseline across different *types* of software:

- some sort of social media
- a spreadsheet — baseline Excel or Google Sheets
- documentation tools — Google Docs, Confluence, Jira, Notion, Obsidian; the
  basic ins and outs
- communication tools — email, Slack, Discord, YouTube, podcasts, TikTok

Plus: find a folder on your own computer, type fast (two fingers — forget it
for now; dictate instead, but you'll need to be able to find the dictation
button), probably use the terminal, debug basic stuff when it breaks. Where
exactly "basic" ends is hard to draw, but that's the shape of it. **You just
have to be online and tech-savvy, at least to a baseline.**

## 2. Critical thinking — be a domain expert with STEM-adjacent interests

> **Why you're doomed without this:** you won't give any added value to the AI
> tool, or you'll become a slop cannon.

### What you need

Baseline knowledge in coding, data analysis, engineering — and in the domain of
the task itself. Without these you'll be lost in unknown unknowns, unable to
apply critical thinking or spot problems.

Most adults already have domain knowledge in whatever field they work in. The
twist is that AI expands what your job needs way past what your job description
used to say. You end up taking tours into domains you weren't familiar with
before — the two biggest being data and coding, but also marketing, creative
writing, UI design, whatever the task pulls in.

On those tours you need some context about the domain, because otherwise you
cannot be the human judge. You won't know whether the output is sensible.
That's how absolute AI slop garbage and hallucinated facts get sent into the
world: the person sending them didn't have the knowledge to detect the errors
or be a sensible human judge. A marketing-only person who's never done raw big
data analysis doesn't know where to start, and doesn't know where to stop the
AI from heading in a very wrong direction.

**If you can't judge the output, that's a good indicator.** Either don't solve
that task yourself yet, or build the extra knowledge first.

This isn't an argument for specialization, and "specialization is a trap" isn't
really a counterargument to it. I'm not saying get a master's in computer
science — that level of depth has diminishing returns for this. Baseline is the
operative word. But baseline in coding and data analysis specifically isn't
optional, because as of 2026 those are still the two domains these tools are
strongest in, and they're exactly the direction most jobs keep expanding into.
You can't do good work on a tour into a domain you have zero foothold in — a
floor of literacy in coding and data isn't specialization, it's what lets you
take that tour at all.

## 3. LLMs — stay curious

Change is the only constant now. Get used to it. Adapt and overcome.

> **Why you're doomed without this:**
>
> - you will keep telling the clankers to `make no mistakes` and heavy-breathe
>   at your keyboard when they make mistakes
> - your colleagues who understand the tools will be annoyed by your ignorance
> - you will write a LinkedIn post about how AI is useless and a hype bubble
>   because you have to keep correcting it
> - you will be replaced by a human who knows how to use AI tools, and then, by
>   AI

### What you need

AI-specific know-how: the basics of LLM-interaction hygiene, what to expect,
how to optimise. You can bash the keyboard and get a decent result anyway. To
do intelligent work you need to know how LLMs actually work:

- how to manage a context window, and where to even keep your context
- good prompting practices, and how to get an LLM to cooperate with you
- how to write documentation so an LLM can read it
- how to work with different models

And what to expect from the interaction: how to not get pissed off when it says
*"alright, sorry, I didn't actually read that document"* after you explicitly
told it to make no mistakes. You should reach a place where that doesn't
surprise or irritate you. It's just normal LLM behaviour.

Anthropic has excellent free courses on all of this. The information is
literally at your fingertips, no barriers at all. The only barrier is that
people don't understand how important this is for the rest of their life.

I'm not confident this one stays a competitive edge for long — interacting with
LLMs keeps getting closer to interacting with a human, and everyone already
knows how to do that. But it's an edge now, because human learning is so, so
painfully slow that most people aren't diving in. And the tools change every
day, so you have to constantly keep up. That's just the way it is.

## The minimum pack

That seems to be the required minimum knowledge pack.

I do believe not everyone will adopt these tools. Even in the high-tech
Estonian medical sector there's a government mandate for clinics to keep phone
lines and live queues, because not everyone knows how to send an email, and not
everyone has a personal phone.

**The mindset package to keep afloat in this storm is a whole other matter.**