Uses and limits
🏠The everyday picture
A brilliant new assistant who has read every book in the library, works instantly and never tires, but sometimes invents facts with total confidence and has never actually met your customers. You'd give them work, and check it.
📘What it really is
- Where AI helps well:
- drafting and rewriting (emails, documentation, reports) for a person to check;
- summarising long material;
- answering questions from documents you provide (Module 86);
- classifying and extracting (sorting enquiries, pulling fields from invoices);
- writing and explaining code, with tests and review (as in Module 78);
-
translation and plain-language explanation.
-
Where it's weak or risky:
- facts without sources: it can make things up (Module 89);
- exact calculation and counting: better done by ordinary code;
- recent events after its training, unless given the information;
- your private context (your customers, policies, prices) unless you provide it;
- high-stakes decisions (medical, legal, financial, hiring) without human review;
-
consistency: it may answer differently each time.
-
Not "just prompting": reliable AI applications are built from parts you control:
user → your backend (login, permissions, context) → model
→ retrieval from approved knowledge → controlled tools with human approval
→ a checked answer or an authorised action
This level teaches each part: models (82), instructions (83), APIs (84), search over your knowledge (85, 86), tools (87), automation (88), and evaluation and safety (89).
- The human stays responsible. AI assists; people decide, especially when actions affect customers, money or data.
🖱️Try it
List five tasks in your CRM business where AI could help (for example: draft replies to enquiries, summarise a customer's history, answer staff questions about policies). For each, write: the benefit, what could go wrong, and what a person must check before it's used. Choose the one you'll build in this level: a staff assistant answering from approved company documents.
✅Quick check
Name two tasks where AI is weak without extra help.
Show the answer
Any two: facts without sources, exact calculations, recent events, private business context, consistent answers, high-stakes decisions without review.
In an AI application, who is responsible for decisions?
Show the answer
The people and organisation using it; AI assists, humans decide.