Top 4 Integration Use Cases for AI Products

Integrations bring value by allowing your AI product to use external data and automate work in other platforms in your users’ software ecosystem. It’s how products fits seamlessly within your users’ workflows.

For example, an AI meeting assistant can integrate with their users’ Zoom to grab meeting recordings and transcripts, their Slack/Teams to notify teammates when summaries and notes are available, their Google Drive/Notion to create summaries and notes as files, and so on.

Across the hundreds of AI companies we've worked with, we've seen 4 main use cases consistently come up:

1. Retrieval-Augmented Generation (RAG) with 3rd-party integrated data

2. AI Agent Tool Calling using 3rd-party APIs

3. AI Workflow Builders with 3rd-party actions

4. Automation Agents with 3rd-party events

RAG Use Case

The more context your LLM has on your user, the more relevant and specific your application will be. This process of retrieving context to answer prompts is called RAG, and integrations are the first step to giving your AI application access to the necessary external knowledge. This can look different for end-users depending on the implementation, from enterprise search applications like You.com all the way to customer support applications like Intercom.

• You.com allows users to select files from their own Google Drive for chat and analysis.

• Intercom allows their AI agent, Fin, to retrieve knwoledge from their users’ Shopify, Stripe, or Statuspage accounts to answer queries

There are two main ways AI applications perform RAG. The first is data ingestion to vector stores. In this method:

1. 3rd-party data is first extracted from your users' 3rd-party applications and indexed into a vector database (this is done in the background so the vector data can be used at prompt-time)

2. When users prompt your AI application, your application would perform a vector search of the database to find relevant context

3. The context retrieved from the vector search is used by the LLM to better answer the prompt

We have a tutorial walking through how to ingest 3rd-party data from Google Drive, Slack, and Notion if you’d like more details on this method.

The second method of performing RAG is by enabling your AI agent to call the 3rd-party API. In this method:

1. When a user prompts, your RAG-enabled application would call the 3rd-party API at prompt-time (essentially going to the external source of truth whenever a user asks for it)

2. The LLM could then use the data returned from the API call in its response

Whatever implementation you choose, the RAG use case allows your AI product to break down data silos and have awareness to data in 3rd-party platforms in your users’ ecosystem.

Tool Calling Use Case

Tool calling is how AI applications go from Q&A chatbots to action-enabled agents. Tools enable AI applications to run code and call APIs so they can do work on a user’s behalf, like query from Salesforce, send a message in Slack, create a task in Asana, etc.

In the example below, we’re using the Paragon MCP to provide Asana tools directly to Cursor.

Here’s how tools work.

1. Your AI engineering team provides metadata to your agent describing when a tool should be used and what inputs they require (look at the Parameters in the Cursor screenshot above)

2. Your engineering team then provides a function that will execute whenever your AI application decides a tool should be called (this is the step where the Asana API was called)

Here's an example from a customer we work with, tl;dv. Their meeting assistant uses tools in their AI product, where their AI agent goes through a meeting transcript and identifies if a task should be created in a 3rd-party integration like Jira, Trella, or Asana.

Notice that tl;dv’s implementation involves a human-in-the-loop experience where tasks are only created with user approval. This is not only a useful practice for create, update, and delete operations where actions can be irreversible, but also a reason why workflow features are useful for AI products.

AI Workflow Actions Use Case

Workflow builders complement AI agents as they allow your users to visualize and define deterministic processes to non-deterministic agent behavior. For example, imagine that whenever a new contact is created in Salesforce, your user wants your AI agent to perform a series of actions:

1. Pull the contact’s data from Salesforce

2. Notify their team in Slack that a new contact was created

3. Send a follow-up email to the new contact

Instead of having your users prompt your AI agent multiple times, you can allow users to build repeatable workflows in your application that can be used in your application.

If you’d like to learn how to easily build this type of workflow feature - with workflow UI 3rd-party actions - check out our tutorial with steps and code repo.

AI companies like Copy.ai and Unify GTM utilize workflows that mix AI steps (for example, scrapes a company page for insights) with 3rd-party actions to send that data to CRMs and messaging platforms.

Notice that in Copy.ai’s product, workflows can be triggered by Lead Created. This type of trigger allows agents to operate autonomously, without the need for user input and prompts - truly automating work with AI.

Automated Agents Use Case

Some use cases require an agent to only perform actions when a user manually prompts. However, automated agents can perform RAG, use tools, and trigger workflows whenever an event occurs in your users’ integrated 3rd-party platforms.

To enable agents to work autonomously, they need the ability to listen for 3rd-party webhooks from integrations in your AI product. In the Copy.ai example above, their AI agents will perform research and send notifications whenever a lead is created in their users’ CRMs. Copy.ai also has a variety of different webhooks - from different integration providers like Salesforce, Hubspot, Gong - that trigger an agent to automate work.

Another example is Scratchpad, Salesforce tool that automates updates, analyzes data, and enforces hygiene. One way Salesforce uses automated agents is in their meeting tool that summarizes transcripts, take notes, and updates Salesforce autonomously whenever a meeting event has ended.

As you may have noticed, many of these integration use cases work synergistically and build on top of one another. Your AI application can start as a RAG-enabled chatbot or tool calling agent, but end up as an automated agent that can perform RAG, tool calling, and user-defined workflows.

Wrapping Up

AI product integrations aren’t just nice-to-haves. They should open the door for a variety of integration use cases:

1. RAG

2. Tool calling

3. AI workflow actions

4. Automated agents

Data and work shouldn’t be separated by platform. They are all a part of your users’ organizational software ecosystem, which your AI product can integrate into.

Paragon’s embedded integration platform can power all of your integration uses cases for AI, and is trusted by leading enterprise AI products including Copy.ai, tl;dv, and You.com. If you’d like to learn more, peruse our catalog of use cases and integrations, or book a demo to talk with our team.