Generative AI vs Agentic AI: what's the difference, and which does your business need?

'Generative AI' and 'agentic AI' get used interchangeably in pitch decks, but they're genuinely different things with different value and risk profiles. Knowing which one a problem calls for is one of the most useful distinctions a business can make right now — and one of the most commonly fudged.

Generative AI: produce on request

Generative AI takes a prompt and produces an output — text, an image, code, a summary. It's reactive: a human asks, the model responds, the human uses the result. The model doesn't take actions in the world or decide its own next step. The great majority of value being captured today lives here.

• Business use cases — drafting marketing copy and emails, summarising documents and calls, answering questions over a knowledge base (RAG), generating and reviewing code, first drafts of nearly any content.
• Its shape — human in the loop, one request at a time, and a small blast radius when it's wrong.

Agentic AI: pursue a goal

An agent is handed a goal and a set of tools, and it decides the steps, takes actions, observes the results, and loops until it's done — with far less human involvement per step. It doesn't just tell you what to do; it does it, using tools and data along the way.

• Business use cases — resolving a support ticket end-to-end, triaging and routing inbound work, multi-step research and report generation, back-office processes that span several systems, coding agents that implement and test a change.
• Its shape — autonomous, multi-step, and tool-using: higher potential value, and a correspondingly larger blast radius.

Which does your business need?

Start with generative AI. It captures enormous value at modest risk, and most 'we need AI' opportunities are really content-generation opportunities in disguise. Reach for agents when the task is genuinely multi-step, spans systems, and is worth the added cost, non-determinism, and guardrail work. In practice, many production systems are a spectrum: a generative core wrapped in just enough agency to be useful.

• Choose generative when — the job is 'produce this from that' and a human will use the output.
• Choose agentic when — the job is 'accomplish this goal' across multiple steps and tools, and the autonomy is worth managing.

Generative AI answers a question. Agentic AI finishes a job. Most businesses need a lot of the first, and a careful little of the second.