Generative AI
Generative AI is AI that produces new content, such as text, images, code, audio or video, in response to a prompt. ChatGPT writing an email, a model drawing an image from a description and a coding agent writing a function are all generative AI.
Why it matters
Older AI mostly sorted and predicted: is this email spam, will this customer churn, what will sales be next quarter. Generative AI makes things, which put it in front of almost every knowledge worker at once. Drafting, summarizing, translating, analyzing a spreadsheet and writing code are now things a model can do a first pass on in seconds.
It also brought AI out of the data science team. You don’t need a trained model for each task. One general model, given the right instructions and context, can handle thousands of different jobs.
How it works, briefly
Text models are trained on huge amounts of writing and learn to predict what comes next, one token at a time. Ask a question and the model generates the most plausible continuation. That’s why the output reads fluently, and also why it can be confidently wrong: the model is producing likely text, not checking facts. Image and audio models work on a similar idea with different kinds of data.
Where it goes wrong
Generative AI makes things up when it lacks information, a problem called hallucination. It knows nothing about your company unless you give it that context. And output quality depends heavily on the instructions and material it’s handed. Most of the work of making generative AI useful at a company is about those last two points: getting the model the right context and checking what comes back.
Generative AI and agents
Agentic AI is generative AI given tools and a loop. Instead of only writing an answer, the model decides on actions, takes them and checks the results. Much of the work on AI products has since moved from chat to agents.