AI for business

We'll Get You Past the Chatbot

Almost every business is already "using AI" in the loosest sense — a chat window here, an AI search result there. Very few have gone further, not because it isn't worth it, but because getting started feels complicated and a bit scary. That's genuinely the hardest part, and it's exactly where we come in.

We're not a software house charging six figures for a developer team. We're the local computer shop that's spent the last while building real tools with AI for our own business — including the website you're reading right now, built with Claude Code. Ask us about it — we mean that.

Talk to us about AI07 3820 900007 3824 8867

There Are Three Levels of AI

Most people only ever meet the first one, assume that's the whole picture, and stop there. It isn't even close.

Level 1

Prompt & Response

A chatbot, an AI-powered search result, a tool that turns a sentence into a picture or a video. This is the level almost everyone's already on, whether they realise it or not — and used on its own, it's basically a very fast conversation with yourself. Useful, but it's the shallow end.

Level 2

Workflows & Assistants

AI sitting inside the tools you already use — Microsoft Copilot, Whisper Flow, an assistant that drafts your emails or summarises a meeting while you're in it. This is where AI starts actually doing work instead of just answering questions, and it's the level most businesses never get shown.

Level 3

Agents & Vibe Coding

This is where it gets interesting, and it's where we spend our time. AI agents that plan and carry out real, multi-step work: building a custom website, a piece of software, a management system, even the code that runs a piece of hardware. Give a man a fish, feed him for a day. Teach a man to fish — sorry, teach a man to vibe-code — and you'll never get him off the keyboard. The imagination is genuinely the only limit here, and it's this site talking.

Being stuck at Level 1 isn't a personal failing. Nobody hands you a manual for Level 2 or 3 — you either stumble onto it or someone shows you. That's the actual job here: not selling you AI, but showing you the door most people don't know exists.

What AI Actually Costs

Everybody has an opinion on whether AI is worth twenty or thirty dollars a month. Almost nobody knows what a single question costs to answer, why a long conversation costs many times more than a short one, or why the companies selling those subscriptions are losing money on them. It's worth five minutes, because once you can see it, you use the thing differently — and you stop being surprised by the bill when you move past the monthly plan.

What gets counted

It Reads and Writes in Tokens

An AI model doesn't see words. It sees tokens — chunks of roughly four characters, so about three-quarters of a word each. Common words are one token, longer ones get split up:

Six words, eight tokens.

Is that just syllables?

It's a good picture to start from, and "lap-top" certainly looks like it — but no, and the real answer is more useful. Tokens follow how common a run of letters is in writing, not how a word sounds. So an everyday word is usually a single token however long it is — "punctuation" is one or two, not one per syllable, because it's a word the thing has seen a million times. An unfamiliar one gets chopped into pieces that needn't line up with syllables at all: a brand name, a part number or a password can cost five or six on its own.

The rule of thumb that does hold: ordinary words are cheap, unusual ones are dear. Plain English is the cheapest thing you can send it.

You pay for every token it reads and every token it writes — and there is always far more reading than writing, which is the whole point of what happens next.

Why it adds up

It Has No Memory At All

This is the bit that surprises people, including plenty who've been using AI daily for a year. The model remembers nothing between messages. It only appears to, because the entire conversation so far is sent again, from the top, every single time you hit enter.

Turn 1: your question.
Turn 2: your question, its answer, and your next question.
Turn 3: all of that, plus the second answer and your third question.
…and so on, growing every time.

So a turn late in a long thread can cost twenty times what the first one did, for the same short question typed at the end of it.

But the clever ones do remember things?

They do — and it's worth knowing exactly how, because the model still isn't the part that changed. The tools built around it can write real notes into real files, keep a running summary of what's happened, and go and look something up when it turns out to matter. The agentic ones — Claude Code, the thing this site was built with — do it constantly.

But nothing sitting in a file is in the model's head. It's outside the conversation and costs nothing at all — right up until something reads it back in, and from that moment you're paying for it on every turn after, the same as anything else you typed.

Which makes a project's own instructions file the example worth understanding: it's read once when a session starts, then rides along with every single message for the rest of it. A tidy one is a rounding error. A bloated one is a tax on every question you ask all day.

The short version: the model didn't get a memory. The software around it got a filing cabinet — and fetching one folder out of it is the cheap version of never having put anything away.

Twice as long isn't twice the cost. Because each turn resends the lot, the running total for a whole conversation grows with roughly the square of its length — a twenty-turn thread costs about four times a ten-turn one, not twice. It's also why a long conversation gets slower, and why it eventually starts losing track of the beginning: there's a hard ceiling on how much can be resent at once.

What To Actually Do About It

Start a new chat for a new subject. Far and away the biggest one. Asking about next week's roster at the end of an hour-long thread about your website means dragging that whole website conversation along with it, every message.
Fold the thread up when it gets long. The better tools can now do this on command — Claude Code's /compact replaces everything said so far with a short summary of what actually mattered and bins the back-and-forth. The running total drops off a cliff and you keep going, as in the chart above.
Give it the relevant bit, not everything. Pasting a forty-page document to ask about one paragraph means paying for forty pages — on that turn and on every turn after it.
Don't send small jobs to the big model. Tidying a sentence doesn't need the most capable model available, and the gap between the cheap one and the expensive one is a good deal wider than most people assume.

Why Your Subscription Is Such Good Value

Because somebody else is wearing the difference. A monthly plan is all-you-can-eat at a fixed price, while the compute behind it is metered by the token — so a heavy user can quietly consume many times the value of what they pay. That isn't a theory: in January 2025 the chief executive of OpenAI said publicly that the company was losing money on its two-hundred-dollar-a-month plan, because people used it more than expected.

The companies building these models are, on their own numbers, spending far more than they take in — buying market share while the technology is young. Great for you today. Just don't mistake it for what the thing costs.

Which matters the moment you build something real. A business process that runs on an all-you-can-eat login is priced on somebody's marketing budget, not on its actual running cost. Build it so you know what a job costs in tokens, and you can answer the only question that matters: is this cheaper than the hour of work it replaces? Usually it is, by a mile — but you should know the number rather than hope.

And yes, we sell AI setup and we're telling you how to use less of it. Working out what this actually costs you is most of the job — a fast conversation you didn't need is still a conversation you paid for. The cheapest token is the one you never sent twice.

Pick However Much of This You Want to Own

Deliberately no pricing here — the right level of involvement is different for every business, and a rate card can't tell you that. Talk to us and we'll work out which of these actually fits.

We Build It For You

Tell us what you're trying to solve and we'll set the whole thing up — the accounts, the tools, the first working version — and hand it over ready to use.

We Teach You

One-on-one, at your own pace. We do the hard, fiddly setup work most people bounce off, then get you far enough that you're genuinely driving it yourself.

We Train Your Team

The same thing, for a whole office. Getting a team over the line together is a different job to teaching one person — we run it as one.

We Build & Host It, You Run It

A real system — not a demo — that your business owns and operates long-term, with us handling the technical setup underneath it.

The Hardware Side, Too

AI isn't only software. We sell and build the machines it actually runs on.

AI-Ready Machines

The new wave of compact AI supercomputers coming out of NVIDIA and AMD, built for running serious AI workloads on your own desk instead of someone else's server. We stock and build with them.

Turn What You've Got Into One

Already own a decent gaming PC? Often that's most of the way to a genuinely capable local AI machine already — we can tell you honestly whether an upgrade gets you there.

AI Gadgets

Real-time translators, transcription pendants, AI meeting and voice recorders — the growing range of purpose-built AI hardware, not just software. If it's AI and it's a physical thing, we're across it.

Level 3 Is How This Website Got Built

Not a slide-deck example — the site you're on, the online shop behind it, the in-house software that runs our counter and workshop, and the custom electronics that plug into it were all built here, with AI agents doing the heavy lifting alongside us. If what you actually want is a website, a custom app or a purpose-built gadget, that's its own page.

Websites, software & custom electronics →

Want a Go Yourself?

Claude Code is the AI coding agent we used to build this site, the shop behind it and the software running our counter. It lives in a terminal window, it reads and writes real files, and it is genuinely the thing that got us from Level 1 to Level 3. If you are curious enough to be reading this far, go and break it for an afternoon.

Try Claude Code yourself

Fair warning: that is our referral link. Sign up through it and we get a bit of free credit to keep building with — it costs you nothing extra, and we would have pointed you at it either way.

Where We Stand

We sell AI hardware and we build AI-powered tools — and we also think the pace this technology is moving at is genuinely worth taking seriously. Both of those are true at once. If you want to look into the actual conversation happening about AI safety rather than the hype either side of it, these are real places to start — not us, just genuinely useful reading.

Ready to see what's past Level 1?

Tell us where you're stuck and what you're trying to get done. No pitch, no pricing sheet shoved at you — just a real conversation about what would actually help.

Start the conversation