A large language model is one component: a trained system that predicts text. Generative AI is the broader category of systems that produce new content, including text, images, audio and code. This article explains the practical difference, what sits around a model in a real product, and why the parts you design tend to matter more than the model you pick. It is written for product, design and engineering teams making their first serious AI investment.
Someone asks which model we should use, and the discussion runs for an hour before anyone has described what the user is trying to do.
I understand the instinct. The models are the visible part, they change monthly, and choosing one feels like the consequential decision. In the products my team has designed, the model has consistently been the smallest contributor to whether the thing worked.
What is a large language model?
A large language model is a system trained on very large volumes of text to predict what comes next in a sequence. That prediction ability, at sufficient scale, produces summarising, translating, answering and reasoning as by-products rather than as separately built capabilities.
LLM technology has a few properties worth holding on to as a product person. A model has no memory between requests unless you give it one. It has no access to your data unless you provide it. It produces a plausible answer whether or not it has the information to produce a correct one. Each of those is a design problem before it is an engineering one.
What is generative AI?
Generative AI is the category of systems that create new content rather than classifying or predicting from a fixed set of options. Language models are one member of that family, alongside image, audio, video and code generation models.
The category label is useful for strategy conversations and almost useless for product decisions. Knowing that you are building with generative AI tells you nothing about what to design. Knowing that you are building with a text model that has no memory and no access to your catalogue tells you a great deal.
Generative AI vs LLM: the practical difference
The distinction that matters for product teams is scope. Generative AI describes what a class of systems does, and an LLM is a specific component you integrate, budget for and swap out.
When people ask about LLM vs generative AI in a planning meeting, the useful reframe is usually a third question: what is the system around the model going to do? That is where the work sits and where the differentiation comes from.
| Generative AI | Large language model | |
|---|---|---|
| What it is | A category of systems that produce new content | A specific trained model that predicts text |
| Scope | Text, image, audio, video, code | Text and code |
| What you decide | Whether generation suits the problem at all | Which model, at what cost and latency |
| What you buy | Nothing directly, it is a category | API access or hosted weights |
| Where product value sits | In the workflow you design around it | In the choices you make about context and control |
LLM applications: what actually sits around the model
An LLM application is the model plus four things you have to design:
- The context you feed it
- The tools it can use
- The evaluation that tells you whether it is working
- The interface where a person accepts or rejects the output
That fourth element is the one design teams own and the one that most often decides adoption. A model that is right 85% of the time is unusable in a workflow with no review step and genuinely valuable in a workflow with a good one.
How I would frame the decision
Start with the workflow, not the model. Describe the task a person is doing today, where it costs them time, and what a correct output looks like. Then work out which parts of that a generative system could take on and where a human still needs to be in the loop.
Choose the model last, and choose it as a replaceable component. Keep the context layer, the evaluation set and the review interface under your own control, and swapping the model becomes a configuration change rather than a rebuild.
Key takeaways
- A large language model is a specific component that predicts text, and generative AI is the wider category of content-producing systems it belongs to.
- The category label guides strategy conversations, while the component properties, no memory, no data access and plausible-sounding errors, guide product design.
- An LLM application is the model plus the context you supply, the tools it can call, the evaluation that measures it and the interface where a person reviews output.
- The review interface is usually what determines adoption, because it decides whether a model that is right most of the time is useful or unusable.
- Design the workflow first and treat the model as a replaceable component, so a better one can be adopted without rebuilding the product.
Designing your first AI-enabled product?
We help product and design teams work out where generative AI belongs in a workflow, and then build the system around it. Talk to our generative AI team about what you are planning.
__________________________________________________________
The Groundwork Advantage
Between 60 and 95 per cent of enterprise AI initiatives produce no measurable business value. Our free AI handbook covers what has to be true inside an organisation before AI produces anything: clear business objectives, data you can trust, governance built in from the start, and infrastructure that keeps your options open.
Read the handbook | Take the five-question self-assessment
__________________________________________________________
