Build a server, host a site, and put AI to work

What “building a server” means today

There are two ways to “build” a server, and it’s worth knowing that the old picture in your head is mostly wrong.

• The physical way — choosing chips and parts, screwing them into a metal case, plugging it in. Real, but these days almost nobody does this for a business website. The data centre and host do it.
• The cloud way — you go to a hosting website, tick boxes for how much power/memory/storage you want, choose an operating system, and click Create. About sixty seconds later, a server exists and is yours to use. No screwdriver.

So “building a server” for us really means choosing the right parts and clicking Create — which is why the next lesson (what the chips do) matters: it’s how you tick the right boxes.

What the chips do — tap each one

When you build (or rent) a server, you’re really choosing these chips. Tap each for what it does and how much you’d want.

From parts to a running machine

Whether physical or cloud, the same things have to happen to turn a box of chips into a useful server:

• Install an operating system — the ground-floor software that runs the machine (for servers, almost always Linux, which is free, stable and light).
• Give it an address — a number on the internet (an “IP address”) so it can be reached.
• Lock it down — a firewall, secure logins, updates. (You met these in the last course.)
• Install the software you need — a web server, a database, your app.

In the cloud, most of this is a few clicks or is done for you. The point is simply to know the recipe, so nobody can bamboozle you with it.

Hosting a website: what it actually takes

“Hosting a website” sounds technical but it’s only three things working together:

• A server that’s always on (you’ve just built one).
• The website’s files sitting on that server.
• A way for the world to find it — an address people can type.

That’s genuinely it. Everything else is polish. The next two lessons walk the recipe and then the actual go-live steps.

The ingredients: domains, DNS and the padlock

Three words you’ll hear, demystified:

• Domain name — the memorable address you rent, like yoursite.com. You register it once a year through a “registrar.”
• DNS (Domain Name System) — the internet’s phone book. Computers actually find each other by number, not name, so DNS quietly translates “yoursite.com” into your server’s number. When people say “point the domain at the server,” they mean updating this phone-book entry.
• HTTPS / SSL certificate — the padlock in the browser bar. It scrambles the connection so nobody can snoop between the visitor and your server. Essential, and nowadays free (via “Let’s Encrypt”).

Taking a website live — the steps in order

Here’s the whole journey from nothing to a live site. Press Next step to walk it.

1. 1
   Register a domain — rent the name (e.g. yoursite.com) from a registrar.
2. 2
   Get a server — spin one up in the cloud with the specs you need.
3. 3
   Install the web server software — the “receptionist” that answers visitors, plus a database if needed.
4. 4
   Upload the website’s files — put the actual pages and app onto the server.
5. 5
   Point the domain at the server (DNS) — update the phone book so the name leads to your server.
6. 6
   Add the padlock (HTTPS) & test — switch on encryption, then check every page works. Live!

Using AI to your advantage — where it actually helps

Set aside the hype. AI is genuinely brilliant at a specific set of jobs, and mediocre-to-dangerous at others. For a firm like yours, the wins are real:

• Drafting — first drafts of letters, minutes, resolutions, emails. You edit; it saves the blank-page time.
• Summarising — boil a long document or thread down to the key points.
• Answering from your own knowledge — ask a question and get an answer drawn from your own vetted material — not the whole internet.
• Finding & checking — “which of these records is missing an email address?” — spotting things across lots of records.
• Automating the dull — classifying, tagging, filling repetitive forms.

Where it’s not to be trusted blindly: anything where being confidently wrong is costly — which is most legal and compliance work. Hence the golden rules, next.

The four golden rules of using AI safely

• 1. Verify. AI can be confidently wrong (“hallucinate”). Treat its answers like a keen junior’s draft — useful, but checked before it goes out.
• 2. Prefer grounded answers. An AI answering from your own vetted documents (and citing them) is far safer than one answering from its general memory. Ungrounded, it’ll happily invent a plausible-sounding rule.
• 3. Guard privacy. Never paste confidential client data into a public AI — it may be stored and seen. This is the reason serious tools run their AI on their own private kit.
• 4. Keep a human in the loop for actions. Let AI suggest; a person approves anything consequential — sending, filing, deleting, paying.

Connecting your AI to the world — giving it hands and eyes

A plain AI is clever but cut off: it only knows what it was trained on, up to a cut-off date, and it can’t do anything. It’s a brain in a jar. You make it genuinely useful by connecting it to tools. Tap each to see what it unlocks.

…but let it loose carefully

Connecting an AI to the internet and to actions is powerful — and precisely because it’s powerful, it needs guardrails. Three risks to know:

• It can be wrong. More reach doesn’t make it more truthful. Ground it and check it (Lesson 8).
• It can be tricked. If it reads a web page or document, hidden instructions in that page can try to hijack it (“prompt injection”). So an AN that browses must treat what it reads as information, not orders.
• Actions can’t always be undone. Reading is safe; sending an email, filing a form, deleting a record, moving money is not. Those must pause and ask a human first.

Done well, this is exactly how a serious tool is built: the AI can see and suggest freely, but doing anything irreversible needs a person’s nod.

Bringing it home: doing it properly

Put the last few lessons together and here’s what a well-built business AI looks like:

• A server with a GPU — the engine that runs the AI.
• An AI grounded in your own vetted knowledge, that cites its source or says “I don’t know” — never inventing facts.
• Connected to your own data, so it can actually help with real work — not just chat.
• Running on your own private kit, so confidential data never leaves — the privacy rule, honoured.
• With a human approving anything that gets sent, filed or changed — the action rule, honoured.

A quick six-question quiz

Instant feedback on each.