AI Primitives

To do Business-as-Code, all work is requested and delivered by a Function using the API or SDK.

Workflows

A Workflow is a special type of Function defined as declarative Code that enables AI and Humans to work together in the pursuit of an Objective as measured by Key Results.

Functions

A Function is a composable and atomic primitive that accepts inputs and returns strongly-typed outputs. There are four types of functions: Code, Generative, Agent, and Human.

Code Functions are deterministic with objective results, making them ideal for anything that has traditionally been capable of being automated with classical computing.
Generative Functions use AI to generate content or structured data. While they are inherently non-deterministic, they may have either objective or subjective results.
Agent Functions use autonomous AI agents to perform tasks.
Human Functions use human workers to perform Tasks in prioritized Queues.

Agents

An Agent is an autonomous digital worker with Goals & Objectives and is evaluated by the Key Results it delivers to the business. It can do work at scheduled intervals or in response to Events in real-time.

Agents interact with other Agents and Humans through Functions and Workflows as Tools. Agents also have access to tens of thousands of searches and actions from external systems via integrations.

Business

Doing Business-as-Code requires clearly-defined Objectives and Key Results. Objectives are the strategic business goals that you want to achieve, and the Key Results are how you measure progress against your goals.

Data

Nouns are the types of physical or digital entities in your business, and Verbs are the potential Actions that can be performed by or to them. Resources are the data records that are instances of Nouns and are stored in the Database.

Events

Events occur when a Subject performs an Action to an Object. They can be Triggers for Workflows or Functions enabling Agents or Humans to take Actions.

Evaluations

Evaluations (i.e. Evals) are the process of measuring the performance of Functions, Workflows, Agents, and Humans against their Objectives and Key Results.

Deterministic evals are those that are based on objective metrics, while Non-deterministic evals require LLM-as-Judge, Human Feedback, and/or business KPIs to measure and score.

With AI Evals, the key variables to be evaluated in the context of inputs & expected outputs are the Prompts and the Models.

Experiments

Experiments allow you to test the performance of Functions, Workflows, Agents, and Humans against their Objectives and Key Results, automatically iterating on the best-performing models, prompts, and settings.

Integrations

Integrations allow you to connect external APIs and systems to your business processes, enabling you to leverage the power of AI and automation to improve your business processes.

Deployment

Deployments enable you to put your Business-as-Code into production on your own cloud infrastructure, with one-click deploy on Cloudflare and Vercel, and integration into the AWS, Azure, and GCP marketplaces.

Observability

Observability enables you to monitor the performance of your business processes, enabling you to identify and resolve issues quickly and efficiently. Every API Request, Function Call, Workflow Invocation, Agent Action, Human Task, and underlying Event including the inputs, outputs, and metadata is logged and available for analysis.