Model

An AI model turns an input into a prediction, label, or generated result. That result is one part of delivery, not the whole delivery system. In July 2026, one production agent system I run used models across more than 2,000 tasks. The median active conversation cost about eight cents. The most expensive task cost $16.12. Moving routine work off an oversized model cut the bill by about 20 times without changing the accepted result. A service firm should choose a model inside a real workflow, then measure the finished work. The surrounding system still needs current client data, permissions, checks, and an owner for exceptions. A stronger model cannot supply a missing record or decide who owns a failure.

How does an AI model produce an output?

A model turns an input into an output through a structure and learned values. NIST defines an AI model as a component of an information system that uses computational, statistical, or machine-learning techniques to produce outputs from inputs.

Most modern models learn those values during training. The process shows the model data, compares its output with a target or objective, and adjusts the model. Google’s machine-learning guide describes the learned result as a complex collection of numbers that captures relationships between input patterns and output values.

Using a trained model is called inference. A spam model receives an email and returns a probability that it belongs in the spam folder. A demand model receives past sales and current conditions and predicts a number. A language model receives tokens and generates more tokens. Each model returns the kind of output it was built to produce.

The model does not keep one neat row for every fact in its training data. Its learned values capture relationships between patterns. A generated answer can resemble a source without retrieving that source. Current client records must reach the system another way.

What is the difference between a model and an AI system?

An AI system combines one or more models with the parts needed to use them for a purpose.

The OECD’s March 2024 explanatory memorandum draws this line. It calls a model a core component used to make inferences. It describes a system as models and other parts built around an objective.

Part	What it supplies	Example in a service firm
Model	A prediction, classification, or generated result	Drafts a project update from supplied notes
Context	The current information for this run	Client scope, project status, and approved terminology
Tool	A controlled way to read data or take action	Reads the project record or saves an approved draft
Workflow	The order, rules, and stopping points	Routes missing facts to the project owner
Eval and review	A repeatable test of the result	Checks dates, scope, unsupported claims, and tone

ChatGPT is an application. The model inside supplies language ability. The application adds the interface, conversation history, file handling, web access, tool connections, and product controls. Two applications can use the same base model and still behave differently because the surrounding parts differ.

This difference helps you find the real problem. If a report uses an old scope, a stronger model may write the wrong report more clearly. The missing current scope is a system problem. If the model sees the right records and fails the same reasoning test, model choice may be the constraint.

Is every AI model a large language model?

Large language models are one family of AI models. They learn broad language patterns and generate text or other token sequences. Other models predict numbers, classify records, rank options, detect objects in images, transcribe audio, or generate images.

A large language model starts broad, so it can handle many language tasks. Stanford describes a foundation model as a large model trained on broad data that can be adapted to many downstream tasks and applications. Broad training gives you more possible uses. It does not make the model right for every step.

Exact tax calculation belongs in ordinary code. A stable yes-or-no routing rule may need a small classifier. A client interview summary fits a language model because the language changes from call to call. One workflow can use all three and still present one finished result to the team.

What changes the model, and what changes only the request?

Training changes the model’s learned values. Fine-tuning continues that process with additional examples or feedback to shape behavior for a narrower need.

A prompt usually does not change the base model. It gives the current run instructions, examples, source material, and a requested format. The answer may improve because the model received better input. Its learned values stay the same.

Retrieval and memory work outside the model as well. They select information and place it into the current context. This matters for private facts. A firm that needs the current contract amount should retrieve the approved contract. Training that amount into a model would make the fact harder to update, prove, and remove.

How should a service firm choose a model?

Choose the smallest model that passes the firm’s test for one named task. “Best model” is incomplete without a task, source set, failure cost, speed requirement, and budget.

Start with real examples from the workflow. Include ordinary work, edge cases, incomplete inputs, and cases that should stop for a person. Write the acceptance rule before you compare models. Run the same set through each candidate. Record output quality, response time, model cost, and human review time.

The cost spread can be large inside one system. Across more than 2,000 agent tasks I measured in July 2026, the median active conversation cost about $0.08. One complex task cost $16.12. I had put an oversized model on routine work. Changing that choice cut the bill by about 20 times with the same accepted result. The full numbers and their limits are in the agent rate-card analysis.

A cheap model that creates ten more minutes of review can cost more than an expensive model that passes cleanly. A capable model used for every small classification can waste money without improving the finished work. Compare model cost and review cost together.

Keep the model connection replaceable. Store the prompt, test cases, output contract, and workflow rules outside provider-specific code where practical. A model upgrade should rerun the test set. It should not make the firm rediscover how the work gets done.

What must exist around the model before client work?

The surrounding system needs a named task and a source of truth. It also needs the right context, limited permissions, a completion check, and an owner for exceptions.

Suppose a model drafts a client status report. The system has to fetch the right project and attach the current scope. It has to separate completed work from planned work, block unsupported promises, and send missing information to someone who can decide. The model supplies the draft. The firm supplies the operating truth.

Write that process down before you automate it. A model cannot resolve a rule that changes by manager and lives only in memory. Document the process first so every run receives the same source, definition of done, and exception path.

Set permissions by what can go wrong. Research and drafting can tolerate a cheap retry. A client promise or deleted record needs a harder boundary. The decision to delegate by how easily work can be checked and undone still applies when a newer model arrives.

Where does a model sit on the Delivery Model Ladder?

Direct model use sits mainly at Stage 1, Enhanced, on the Delivery Model Ladder. An employee opens an AI application, supplies the input, checks the output, and carries it into the next system. The person moves faster while the delivery path stays mostly intact.

At Stage 2, Augmented, a workflow calls one or more models for bounded steps. The workflow supplies approved context, checks results, controls actions, and sends exceptions to an owner. Model choice becomes one setting inside a redesigned process.

Stage 3, AI-native, depends on shared operating context and an owned system across the firm. A larger model cannot create those conditions. Treating a model demonstration as a finished operating change is one way firms enter the Production Gap.

When is an AI model the wrong tool?

Use ordinary software when a rule is exact and stable. Arithmetic, database constraints, fixed eligibility rules, and irreversible account changes should not vary because a model produced another plausible answer.

A model can handle the messy language around an exact step. It can extract invoice fields from an email while code calculates tax and checks the total. It can classify a request while a permissions rule decides whether any action is allowed.

This advice assumes the firm can name the task and test the output. Work with no source of truth or acceptance rule stays manual until those parts exist.