Every tax rule ever written assumed a human would read it. A person sitting with a statute, a calculator, and years of training, working through the logic line by line. That assumption held for a century.

It doesn't hold anymore.

Today, the thing reading the rule is just as likely to be a machine. Someone opens an AI agent at 11pm, types "how much tax do I owe on £60,000 self-employed," and gets a confident number back in two seconds. No statute. No calculator. No training. Just a model predicting what a tax answer probably looks like.

And here's the uncomfortable truth I kept running into as an accountant: the machine is often wrong. Not occasionally — routinely. Wrong rates. Last year's thresholds. Deductions that don't exist in that country. All delivered with the same calm confidence as the correct ones.

The problem isn't AI. It's what we feed it.

AI models don't have a live connection to the tax code. They learned from whatever was on the internet when they were trained — forum posts, outdated blog articles, marketing pages, the occasional government PDF, a lot of confidently-wrong noise. Tax changes every year. The training data doesn't.

So when an agent answers a tax question, it isn't consulting the law. It's recalling an averaged impression of the law from a frozen moment in the past. For most questions, "roughly right" is fine. For tax, "roughly right" is a penalty notice.

We can keep telling people not to use AI for tax. They won't listen — it's free, instant, and available when no accountant is. Or we can accept the obvious: AI is going to do tax work whether the profession participates or not. The only question left is what it reads when it does.

So we gave it something correct to read.

I spent the last stretch of my career doing something no accountant is supposed to have time for: turning tax knowledge into something a machine can read correctly.

We call them skills — structured documents that lay out the actual rates, thresholds, deductions, deadlines and computation steps for a specific country and tax type, each cited back to primary legislation. An AI agent references the skill instead of guessing. It stops recalling a blurry memory of the law and starts following rules an accountant actually wrote.

Three things make this different from anything that came before it:

It's open-source. The skills are free and public. Any AI model can use them — not one company's product, not one chatbot. The point isn't to own tax knowledge. It's to make sure the version the machines read is the correct one.
It's human-verified. AI can draft a skill, but a qualified, practising accountant in that jurisdiction reviews it and signs their name to it. Their credentials sit permanently and publicly on what they verify.
It's everywhere. The library already spans hundreds of skills across the majority of the world's countries — income tax, VAT and GST, payroll, formation, cross-border. Not one market. The map.

No one had put those three things together. That's not a slogan we print on the homepage — it's just what happened when an accountant got tired of watching machines get tax wrong and decided to do something an engineer couldn't and a tech company wouldn't.

Why this had to come from an accountant

A tech company can build the plumbing. It cannot tell you whether the answer is right. That requires someone who has filed real returns, argued real positions, and carried the professional liability when they're wrong. The credibility of every skill rests on a human professional standing behind it — and that human has to be one of us.

That's also why this isn't a threat to the profession. It's the opposite. For the first time, an accountant's judgement can sit behind millions of AI answers at once. Your expertise stops being something you sell one hour at a time and becomes something that scales — while you stay the authority.

Where this goes

Tax was written for humans because, until recently, only humans could read. That era is ending. The rules that govern how every person and business on earth is taxed are about to be read, applied and acted on by machines billions of times a year.

Someone has to make sure those machines are reading the truth. We decided it should be accountants, in the open, with their names on the work.

If you're an accountant, verify your country. If you build with AI, the skills are free to use today. And if you just want your own taxes done right by an agent, start here.

The machines are already reading. Let's make sure they're reading the truth.