liveBook · Manning

Contract

As explained in Effective Conversational AI

An application programming interface (API) is a defined interface through which one software system or component requests capabilities, data, or operations from another.

A chatbot can understand “I’m locked out” yet still leave the user stuck. Recognizing a password-reset request is only part of the job: the assistant also needs a way to authenticate the user and reset the password. An application programming interface (API) is a defined way for the assistant to request that work from another system. section 1.1.3 section 1.1.2

Think of an API call as a handoff in a conversation. The assistant identifies what the user wants, gathers the information needed to act, and calls the relevant service. The service returns a response the assistant can use to continue the dialogue. A developer must make the API available to the assistant and define its required inputs, response format, and error conditions. A question-answering bot may rarely need this handoff; a task-oriented bot often does. section 1.1.3 section 1.1.2

Two useful API patterns are:

  • Complete a task or reduce user effort. A backend connection can retrieve customer information or handle part of a self-service process instead of making the user supply every detail. section 8.2.5
  • Add an external capability. A conversational AI platform can call a generative AI service for text generation; platforms may call this connection an extension, integration, or fulfillment. section 2.3.1

Make a difficult phone question simpler

Suppose an insurance member ID has nine or eleven digits, and the nine-digit version begins with a letter. Asking callers to speak an alphanumeric ID is error-prone. A backend step can recognize when it has received enough digits and add the shared initial letter for the user. The assistant can then ask a simpler question, while reserving clearer instructions for a retry. That convenience must be weighed against implementation cost and new failure points; provide a next-best path when the backend step fails. section 8.2.5

Let an LLM choose; let the API check

For a logged-in customer asking, “How much money is in my son’s account?”, an assistant can give an LLM the customer’s account descriptions and ask it to select the likely account ID. The assistant still makes the final check-balance API call and formulates the reply. This division is suited to a small set of accounts; searches involving many records or strict criteria need greater care. section 10.2.2

When connecting an API, inspect its documented methods, required and optional parameters, and responses. An OpenAPI specification can describe these details; the connection also needs an implementation URL and authentication information such as an API key. Integration may be coded or configured through a visual interface, depending on the platform. Judge the result by whether it helps users finish the task with less effort—and whether errors leave them a workable alternative. section 2.3.1 section 8.2.5

How Effective Conversational AI covers it

How deep it goes
Explains API inputs, responses, errors, and platform integration steps without descending into implementation code.
Examples
Illustrates password resets, an insurance member-ID voice flow, and an LLM-assisted bank-account balance request.
What it assumes
Introduces APIs by name and explains their conversational role while assuming a developer can expose the service.
What you can do after
Apply it via an assistant platform to connect external services while planning for user effort and API failures.
How it connects
Links APIs to intent recognition, dialogue orchestration, backend processes, OpenAPI specifications, and large language models.

Also explained in

Compare how 44 books explain contract →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Effective Conversational AI. The section links point to the parts of the book it draws on.