Welcome to the Van
Let's start with a simple question. When you hear "AI", what comes to mind? Robots laying bricks? A chatbot that gets everything wrong?
The truth sits somewhere in the middle. In this course, "AI" mostly means AI assistants: tools like ChatGPT, Gemini, Claude and Microsoft 365 Copilot. You type or speak a request, and they write back in plain language.
They can't hold a screwdriver. They can't smell gas or feel a hot cable. What they're good at is words: drafting, rewording, summarising and explaining. So in this first level we'll get an honest picture of where that helps on the tools and where it doesn't.
Think of It as a Keen Apprentice
Here's an analogy we'll use all course. An AI assistant is like a very keen apprentice who has read a huge pile of books but has never been on a job.
Ask them to tidy up a scruffy message to a customer and they'll do it in seconds. Ask them which cable size to use on a particular circuit and they'll give you a confident answer. It might be right. It might be badly wrong. And they'll sound just as sure either way.
You wouldn't let a first-week apprentice sign off a job. Same rule here.
Why Does It Sound So Sure When It's Wrong?
This is the most important idea in the whole level, so let's take it slowly.
An AI assistant doesn't look facts up the way you'd check a manual. It learned patterns from huge amounts of text, and it writes the words that are most likely to come next. Most of the time that gives you something sensible. But when it doesn't really know the answer, it can fill the gap with something that just sounds right.
These made-up answers are called hallucinations. In September 2025, OpenAI (the company behind ChatGPT) published research explaining that one reason they keep happening is that the way models are trained and tested rewards guessing over admitting "I'm not sure". Picture a quiz where a blank answer scores zero but a guess might score a point. Over time you learn to always guess.
So confidence tells you nothing about accuracy. A smooth, well-written answer can still have a made-up regulation number, a wrong torque setting or a manual that doesn't exist.
And this is exactly why AI never replaces the Building Regulations, BS 7671, Gas Safe rules, the manufacturer's instructions or your own competent judgement. Those are the sources. The AI is just a helper.
Three Buckets for Your Jobs
So how do you decide what to hand over? Try sorting your everyday tasks into three buckets.
1. AI can help. These are low-risk jobs where a mistake is easy to spot and costs you very little. For example, rewording a text to say you're running late, coming up with ideas for a social media post, or explaining a technical term in plain English so you can understand it better.
2. AI can draft, but you check. The AI saves you time, but every word gets read before it goes anywhere. Think of turning your rough notes into a tidy job description, drafting the wording of an invoice, or summarising a long data sheet. You check it against your notes and the original document. Your name is on it, so it's your responsibility.
3. Human only. Decisions about safety, compliance and money you commit to. Deciding whether an appliance is safe to leave on, signing a certificate, picking the final design for a circuit, or telling a customer what something will cost. The AI can't see the job, and it can't take responsibility. You can.
Quick test. Where would you put "ask AI whether this boiler flue installation complies"? That's human only. The AI might help you find which section of the manufacturer's instructions to read, but the judgement is yours.
What the Evidence Says (and How Old It Is)
So are tradespeople actually using these tools? Let's look at some real surveys, and notice the date on each one.
Jobber, a US job-management software company, published its 2026 Home Service Trends Report. It surveyed 1,050 US home service business owners in December 2025. It found 54% were using AI for quoting, 52% for invoicing and 51% for business writing like emails and proposals. HVAC, plumbing and roofing were leading the way.
A survey by Bourne AI, run between 28 May and 14 June 2026 with 1,000 skilled trade workers in the US, UK, Germany and France, found about a third already use ChatGPT, Gemini or Claude on the job.
Now compare that with a UK survey from the training provider Logic4training, published on 29 August 2025. Among the plumbers, gas engineers, electricians and maintenance workers who answered, only 4.8% said they actively used AI tools in their work.
Why are the numbers so different? The surveys asked different people, in different countries, at different times. And two of them come from companies that sell software, so they have a reason to show plenty of interest. None of these numbers is "the truth" about every trade. Each one is a snapshot.
That's the habit we want you to build. Whenever you quote a statistic, say who ran it, who they asked, and when. Today is October 2026, so the Jobber fieldwork is about ten months old, and in AI terms that's a long time.
Your Turn
Grab a notepad, or your phone. Write down ten tasks from your last working week. Then put each one in a bucket: AI can help, AI drafts, I check, or human only. Be honest. If a task touches safety, compliance or a price you're promising, it goes in the human-only bucket.
Checklist
- I've listed ten real tasks from my week
- Each task is in one of the three buckets
- Every safety or sign-off task is marked human only
- I can explain hallucinations in one sentence
- I can name one survey and say how old it is
Key Takeaways
- AI assistants are good with words: drafting, rewording, summarising and explaining. They can't see or test a job.
- They can sound confident and still be wrong, because they predict likely-sounding text and have been rewarded for guessing.
- Sort tasks into: AI can help, AI drafts and you check, and human only.
- Regulations, manufacturer instructions and your competent judgement always come first.
- Every statistic needs a source and a date, and you should ask who ran it and why.