Foundation model

A large model trained on broad data that can be adapted to many different tasks instead of being built for only one job.

Where it creates value

Best suited to work like this.

Starting new AI solutions quickly, reusing common understanding across processes, and supporting tasks with limited task-specific training data.

What it does

Reusable language, vision, audio, multimodal, scientific, or control capabilities across many downstream tasks.

Typical value

Accelerates experimentation and solution development while allowing capabilities to be reused across processes and teams.

Where it performs well: One model can support many tasks and reduce the need to train separate systems for every use case.

Value maturity · editorial assessment Scaling

Foundation models are being scaled across many enterprise tasks, but dependable regulated operating patterns are still developing.

Back

TEC-0064 · How it works in practice

Read the card type together with the use case. A component can be central in one solution and enabling in another.

01 · Inputs

Broad pretraining data plus prompts, context, retrieved knowledge, examples, or fine-tuning data for the intended task.

02 · Intelligent task

Reusable language, vision, audio, multimodal, scientific, or control capabilities across many downstream tasks.

03 · Outputs

Generated content, predictions, embeddings, structured data, actions, or task-specific model behavior.

Common application patterns

Enablers

Model selection, evaluation for the intended use, secure hosting or access, adaptation method, monitoring, and governance of inherited risks.

Where it struggles

Inherited bias and errors can spread across downstream uses; costs, opacity, vendor dependency, and unpredictable behavior can be significant.

Human role

People define the task, provide context, evaluate behavior, and set the level of autonomy and review.

Connected use cases

Examples from the current use-case library.