Will agents replace integrations?

Steven Forth is CEO of Ibbaka. Connect on LinkedIn

Integrations are central to B2B software. Once upon a time, professional services teams charged large amounts of money to integrate applications and then maintain those integrations. Today, most vendors include the most important integrations as part of the subscription. For Ibbaka’s Customer Value Management Platform, we include integrations with Hubspot and Salesforce and are planning to integrate with billing and subscription management platforms. This is part of the basic support for most of our packages and not something we charge for.

This has been made possible by two things:

• Better APIs - Application Program Interfaces that are stable and well documented

• Integration Applications - Zapier and its alternatives provide many prebuilt integrations and tools to manage these integrations

But the world is changing again.

AI agents are emerging that can replace conventional integrations and add a lot more value, thereby justifying higher prices.

Zapier pricing

The reference value for integrations has become Zapier. Most buyers will also have it as a reference when they start looking at integration agents. So lets begin by looking at Zapier’s pricing.

Zapier prices per task. “A task is an action your automated workflow successfully completes. For example, if your Zap has an action to create new Google Contacts, each contact Zapier creates will count as one task.” (Of course it is a bit more complicated than this, check out How is task usage measured in Zapier?).

How does price scale for the Professional and Team plans based on the number of tasks?

Note that the X-axis here is the number of tasks and is on a log scale. It is interesting to note that the price per task converges for the two packages.

So the price for a Zapier Zap (as they call their tasks) is quite small. Even if you are only running 2,000 taks per month it is just over 3 cents for Team and just over 2 cents for the Professional package. That is a pretty low price and Zapier is not even the cheapest alternative.

Can Agent Integration Offer More Value than Workflow Automation?

Given this low price point and the efficiency of existing integration services is there room (or a need) for agentic AI here?

Let’s look at one of the emerging multi-agent platforms, CrewAI. Its positioning is not dissimilar to Zapier.

CrewAI

Streamline workflows across industries with powerful AI agents. Build and deploy automated workflows using any LLM and cloud platform.

Zapier

Automate without limits.

I used Perplexity to compare the two applications and approaches.

Approach to Automation

Zapier:

• Uses a trigger-action model where specific events in one app trigger actions in another

• Focuses on direct integrations between existing apps and services

• Offers a vast library of pre-built integrations with over 7,000 apps

CrewAI:

• Utilizes AI agents with specific roles and expertise to perform tasks

• Focuses on creating autonomous AI teams that can handle complex, multi-step processes

• Allows for more dynamic and intelligent decision-making within workflows

Core Technology

Zapier:

• Primarily uses API connections and webhooks to transfer data between applications

• Relies on predefined integration points provided by various apps and services

CrewAI:

• Built on top of LangChain, leveraging large language models (LLMs) for AI agent functionality

• Incorporates advanced AI and machine learning techniques to enable autonomous agent behavior

Flexibility and Customization

Zapier:

• Offers a wide range of pre-built integrations and templates

• Allows for some customization through multi-step Zaps and custom webhooks

CrewAI:

• Provides greater flexibility in defining custom AI agents with specific roles and expertise

• Allows for the creation of more complex, intelligent workflows that can adapt to changing conditions

Use Cases

Zapier:

• Excels at automating routine tasks and data transfers between popular business applications

• Ideal for straightforward, rule-based integrations across a wide variety of apps

CrewAI:

• Better suited for complex, knowledge-based tasks that require decision-making and problem-solving

• Designed for scenarios where AI agents need to collaborate, reason, and adapt to achieve goals

Development and Extensibility

Zapier:

• Primarily focused on connecting existing apps and services

• Offers some development options through custom webhooks and Zapier Platform CLI

CrewAI:

• Provides a more extensible framework for developing custom AI agents and tools

• Allows for integration with external APIs and custom code execution within a safe environment

Boiling this down, Zapier is a fairly simple automation tool that stitches together APIs. This is the core value proposition of application integration and Zapier does it well.

CrewAI promises a lot more. It will let you create teams of agents that will execute complex work processes that include decision making and problem solving. In fact, there are even agents that generate APIs.

The higher level functions promised by CrewAI can be a lot more valuable than simple integration. If CrewAI can find ways to quantify that value (using a value model of course) it has a path to much higher pricing than Zapier. And it will need to get that higher pricing as well, as it will be relying on its own API calls to LLMs (Large Language Models) and these are not cheap. See Countless.dev for a quick way to cost out and compare many popular models.

Value drivers for integration with Agentic AI

Agenetic AI application integration offers a number of value drivers above and beyond that of conventional API integration.

Revenue Impact:

Both approaches can increase revenue, but agentic AI potentially offers faster time-to-market and more diverse integration possibilities.

Agentic AI may provide more upsell opportunities due to enhanced product capabilities.

Cost Reduction:

While both reduce costs, agentic AI potentially offers greater savings in development time and specialized labor.

Conventional integration may require ongoing maintenance, while agentic AI could offer self-healing capabilities, reducing long-term costs.

Risk Mitigation:

Both approaches reduce certain risks, but agentic AI may offer superior adaptability to changing standards and enhanced security through intelligent monitoring.

Conventional integration might be more predictable and easier to audit, potentially reducing certain compliance risks.

Optionality:

Agentic AI offers significantly more flexibility and adaptability, potentially enabling more complex integrations and novel use cases.

Conventional integration provides a solid foundation for modular architecture but may be less agile in adapting to new opportunities.

Efficiency Improvements:

Both improve efficiency, but agentic AI can potentially offer continuous optimization and intelligent resource allocation.

Conventional integration provides predictable efficiency gains, while agentic AI might offer more dynamic improvements.

Probable Outcomes

APIs will be central to the agentic economy. This economy will be powered by data and APIs are the valves that control access and flow. There is room for a lot of players and approaches.

A few of the many players and four of the niches are shown in the below sketch.

Starting at the bottom and working up …

At the base of the ecosystem are the APIs themselves (and below that, not shown, are the millions or billions of agents and applications generating and consuming data through APIs). There are many applications available to organize and manage collections of APIs.

API integration and automation will remain important for many years to come. It is part of the connective tissue of the B2B SaaS ecology and sometimes one wants the simplest, most robust and least expensive alternative. Prices for these services are likely to decline sharply over the next decade as they become commoditized and Moore’s law drives costs down.

The value will be added higher in the chain. One set of agents, like those from AgentX and its competitors, will leverage the API integration layer and build value added agents on top, providing insight and action on the data. These agents will be priced from 3X to 1,000X higher than that of the underlying API depending on the value they offer.

A different approach will be to bypass the API integration layer all together, and have the agents access APIs and more importantly create APIs where they find none. This is the approach being taken by CrewAI and SmythOS. It has the potential to create a much more adaptive API ecosystem, where APIs are generated as needed. There will be a lot of money made here as the more adaptive approach will let these companies move at the speed of the generative AI economy.