- A connector, an integration, a workflow, and an automation each answers a different question. Conflating them can be confusing when you try to lay out your integrations and what they do.
- A single connector can power many separate integrations. One might sync contacts, another creates opportunities, and another provisions accounts.
- A workflow orchestrates steps within an integration. A bidirectional integration might include one workflow that sends contacts out and another that imports updates.
- You can integrate two apps without automating anything, or automate one step of a workflow. An automation makes work happen without manual intervention; an integration makes the systems capable of working together in the first place.
- An AI agent chooses which workflow to invoke; a deterministic workflow follows defined logic. The same integration can support both, as long as the AI accesses approved tools through the MCP flow server.
In integration conversations, connectors, integrations, workflows, and automations are often used interchangeably. That may be okay until you're deciding what to build, what to reuse, and how to deliver integrations as part of a B2B SaaS product. Then, some clarity is essential.
In this post, we use connector for reusable connectivity code and integration for the larger business capability. An integration isn't necessarily an automation. A workflow describes something different from both. APIs and AI can add further complexity.
And yes, we know that these terms and definitions vary across vendors. So, everything we say here is based on how we understand and use them.
Start with the connector
A connector is reusable code that enables an integration to interact with an application, service, database, protocol, or other system.
A Salesforce connector, for example, might handle OAuth authentication and provide actions to create, update, and query Salesforce records. Rather than rewriting that authentication and API-request code for every use case, you reuse the connector.
Connectors typically provide:
- Authentication and connections
- Actions that interact with an API or other interface
- Triggers that respond to events
- Configuration options
- Handling for common API behaviors such as pagination and rate limits
Connectors aren't limited to SaaS applications. They can provide reusable functionality for databases, message queues, file systems, SFTP, and other systems.
At Prismatic, a connector is a reusable building block that communicates with an external system and exposes functionality (such as actions, triggers, connections, or data sources) for use in integrations. A component includes connectors but also reusable logic, helper functions, and other building blocks that may not connect directly to an external API. So, connectors are a subset of components.
Workflows define what happens
A workflow is a sequence of steps, mappings, and decisions that accomplishes a process. It might retrieve data, transform it, evaluate conditions, loop over records, send data, or send a notification. Within an integration, a workflow interacts with the connectors and orchestrates the business logic. An integration may consist of one or more workflows (aka flows).
An order-fulfillment workflow might look like this:
New order → Validate order → Look up customer → Check inventory → Create fulfillment request → Update CRM → Notify account manager
The integration, for example, might include one workflow that sends new contacts from your product to Salesforce and another that imports updated Salesforce contacts into your product.
The integration is the larger product capability, while workflows define the processes, mappings, and business logic that customers care about.
An integration puts connectors to work
An integration is a complete, configured capability that connects systems to solve a business need. It includes the triggers, API calls, transformations, branching, business logic, error handling, configuration, and other pieces required to make the systems work together.
Suppose your SaaS product needs to work with Salesforce. That integration might:
- Receive a closed-won opportunity from Salesforce
- Retrieve the related account and contact
- Transform the data to meet your product's format
- Create a customer record in your product and your billing system
- Notify CS via Slack
That integration might use Salesforce, Stripe, Slack, and your own application's connectors. None of those connectors is the integration. The integration is the complete, deployable capability built from them. An integration generally has one or more connectors, and one or more workflows.
A Salesforce connector isn't a Salesforce integration
Some vendors use "integration connector" to describe what we'd call an integration. Others use "connector" for nearly anything that connects two systems.
We distinguish the terms because they solve different problems.
A Salesforce connector might power many integrations: one that syncs contacts, one that creates opportunities, one that provisions accounts, and another that synchronizes several objects bidirectionally.
That distinction also changes how you evaluate integration platforms. A vendor may have 200 connectors, another 500, and another 1,000. But how many of those are the integrations your customers need?
Connector availability matters. You don't want devs rebuilding basic API functionality. But Prismatic's open-source connector library illustrates the larger point: connectors are increasingly accessible, while authentication, deployment, monitoring, security, versioning, and customer-specific configuration remain substantial challenges.
Automation streamlines specific work
An automation uses software to perform work that would otherwise require manual intervention. It can execute part or all of an integration or workflow.
Suppose that when a customer closes an opportunity, someone must copy the customer information into your application, create an account in the billing platform, and notify customer success. You can connect those systems and still require someone to start the process. Or you can automate it.
The closed-won event triggers the process, and the remaining steps run automatically. In another scenario, automation might run only the notification step when a condition is met, while the rest of the integration is handled separately.
Integration and automation are closely related, but they aren't synonyms:
- You can integrate two applications without fully automating the process between them.
- You can automate a process entirely within one application, without an integration.
- You can automate one step within a larger integration.
Integration enables systems to work together. Automation makes work happen without manual intervention.
Integrations make automations more powerful
Without integrations, automations are siloed. You might automate everything inside Salesforce, but you're stuck when the process needs data from your product or needs to update NetSuite.
Integrations provide access to events in other applications, data an automation couldn't otherwise see, and actions it couldn't otherwise perform.
For example, a closed-won opportunity can trigger an integration. A workflow can retrieve customer information, provision an account, configure billing, update the CRM, and notify the right people. Automation can run the full workflow or invoke part of it when an event or condition occurs.
Integrations don't compete with automations. Instead, they help make automation useful across the business.
APIs help everything work together
An API defines how software can interact programmatically with an application. It might let you retrieve a Salesforce contact, create a Stripe customer, update a Zendesk ticket, or post a Slack message.
Using an API directly means handling auth, webhooks, rate limits and more. A connector packages all of that into reusable functionality. An integration uses connectors to access APIs or other interfaces. Workflows arrange the resulting actions, mappings, and decisions into a process. Automation runs some or all of that process when the appropriate event, schedule, request, or condition occurs.
It isn't a rigid hierarchy. An integration can contain multiple workflows and use multiple connectors, and an automation can invoke only part of a workflow. Systems can also exchange information through webhooks, databases, queues, files, SFTP, and other mechanisms.
The model is still useful:
- An API or other interface gives you access.
- A connector makes that access reusable.
- An integration applies it to a business need.
- A workflow defines the sequence, mappings, and logic.
- Automation makes some or all of the work happen.
AI adds another layer
During development, AI coding agents can help engineers understand unfamiliar APIs, write connector code, build integration logic, generate transformations, test integrations, and troubleshoot problems.
AI can also interact with workflows once they are built.
Traditional workflows are deterministic. That's ideal for predictable processes such as updating an account after a successful payment or stopping when a required field is missing.
AI introduces reasoning into selected parts of the process. Instead of defining every path in advance, you can give an AI agent a goal, context, and approved tools (via the MCP flow server) and let it determine what should happen next.
That doesn't mean every workflow should be agentic. Predictable processes are usually better handled by deterministic software. AI is useful when a process involves ambiguity, interpretation, or decisions that don't fit neatly into if/then statements.
AI can reason, but within the constraints of approved tools
An AI model may determine that a support issue warrants a refund. Retrieving the customer's order, checking its status, creating the refund in the billing system, updating the support ticket, and notifying the customer requires access to those systems.
That's what the integration handles.
AI agents need controlled, reliable ways to retrieve data and act across applications. Existing connectors and integrations can provide those capabilities instead of requiring a separate connectivity layer for AI.
One integration can support deterministic workflows and AI-invoked flows, as long as the AI has access to approved tools.
Why the distinctions matter for B2B SaaS
You don't need to correct someone every time they call an integration a connector. But if you're responsible for your B2B SaaS product's integration strategy, the distinctions are important.
If you think connectors are integrations, you might judge a platform primarily by its connector catalog. If you think integration and automation are interchangeable, you might choose a workflow automation tool without considering whether it supports multi-tenant, customer-facing integrations. If you assume every integration is just a connector, you may underestimate the workflows, configuration, monitoring, and support that it needs to be useful to your customers. And if you assume AI eliminates the need for integration infrastructure, you're in for a tough ride.
The more important questions are:
- Can you build the business logic your customers need?
- Can you reuse connectors instead of rebuilding API functionality for every integration?
- Can you build workflows that span multiple systems?
- Can you deploy and configure integrations for hundreds or thousands of customers?
- Can customers build their own workflows when appropriate?
- Can you monitor and support integrations after deployment?
- Can you safely expose those capabilities to AI when that's the right approach?
Those questions tell you more than a connector count. As we note in our post on the embedded workflow builder, the right approach is often a combination of productized integrations, bespoke integrations, and customer-created workflows.
Connectors are the beginning, not the integration
Connectors, integrations, workflows, and automations overlap because they're parts of the same ecosystem. But they aren't different names for the same thing.
- A connector answers: "How can I interact with this system?"
- An integration answers: "How do these systems work together to solve a business need?"
- A workflow answers: "How should data and work move through the integration?"
- Automation answers: "When and how should some or all of that work happen without manual intervention?"
And an AI can interpret context, make recommendations, or choose among approved actions when the path is not fully predetermined.
Keeping these terms straight helps you understand what you're building. To see all of these in action on an integration platform, check out our free trial.
