Generative AI is software that learns the patterns in huge amounts of text, images or code and uses them to create new content on request. Ask for a summary, a letter or a piece of code and it writes one, predicting what should come next. It produces what is likely rather than what is verified, so people should check its output.

Artificial intelligence is the whole field of computers performing tasks that need intelligence; generative AI is the branch that creates new content. Most earlier business AI predicted or classified, for example scoring a credit risk or flagging an unusual payment. Generative AI writes, draws or codes. Many useful systems combine both, with a person making the decision.

Foundation models are large AI models trained on broad data that can be adapted to many tasks, such as drafting, summarising, translating and coding. Claude from Anthropic and Gemini from Google are examples, as are open-weight models you can run yourself. Businesses rarely train their own; they connect a foundation model to their own documents and systems.

Common generative AI examples at work include drafting correspondence, summarising long documents and meetings, answering staff questions from policy documents, producing first drafts of reports and creating image concepts for marketing. In each case the model drafts and a person checks. The strongest examples involve heavy reading and writing and a clear source of truth.

Yes. Microsoft 365 Copilot is a generative AI assistant built on large language models and grounded in the emails, files, chats and meetings each user can already open. Because it sees what its user can see, permissions and sharing should be cleaned up before rollout, so that it cannot surface documents people should never have been able to open.

A hallucination is a confident, fluent output that is false, such as an invented statistic, court case or reference. It happens because models generate plausible text rather than look facts up. Connecting the model to approved sources, requiring citations and having a person check anything that leaves the organisation are the main defences.

Beyond generative AI, business innovation comes from AI that predicts, detects and optimises: forecasting demand, flagging billing anomalies, extracting data from documents, planning routes and rosters, and agents that complete multi-step tasks. These techniques often run on data you already hold, and they pair well with generative AI, which explains the results in plain language.

Generative AI can be used lawfully under POPIA if personal information is protected. Use business editions with proper data terms, keep personal information out of unapproved tools, document any processing outside South Africa under section 72, and keep a person responsible for decisions that substantially affect someone. An acceptable-use policy and staff training make the rules stick.