How artificial intelligence is shaping the future of API development

Microsoft predicts that 100 million new applications will be created per year in 2020, leading us to develop more applications in five years than we have in the last 40 years. This prediction was based on the capabilities of low-code and code-free software. LLMs have improved dramatically since then, and the number of applications created each year will continue to grow.

That's why companies need a robust strategy to bring these applications together, and APIs are the most appropriate tool. Traditionally, developing, documenting and implementing APIs has been time-consuming and technically challenging. However, I believe that LLMs can simplify API development. In this article, I will share practical tips that can simplify tasks such as extracting data, creating documentation, and building code using LLMs.

Getting started with API design

Before implementing an API, complex business requirements need to be analyzed: what actions should be performed, what data should be exposed, and what data management criteria should be taken into account. LLMs can help analyze the requirements specifications and define use cases and workflow patterns by creating a first draft of the initial API design.

Practitioners can inject technical specifications and business requirements into the model so that a linked LLM can create a high-level offering of your API based on the common patterns identified. The more information you add to the LLM over time, the better the results will be. This does not negate the need to understand the technical requirements. However, it will help jumpstart your API design by giving you and your team a framework for discussing and agreeing on technical specifications.

Creating code scaffolds and generating APIs

The concept of creating scaffolding for code is not new. For example, while artificial intelligence can automatically generate API endpoints based on a data model in different languages, LLMs take it a step further. By analyzing API specifications created from the same model, these designs can be pushed back to LLMs to create skeletal applications that include specific routes and connect to underlying data sources. Instead of manually implementing the logic for input validation, connection to subsystems, error handling, authentication and authorization, these tools help avoid repetitive tasks and streamline development. This fundamental work allows developers to focus on refining business logic, handling business-related exceptions and edge cases, or improving code efficiency for scaling purposes.

Automating data display using artificial intelligence

Data matching is a time-consuming and error-prone task. For example, developers typically struggle with complex JSON structures or native XML with nested arrays - data structures with hundreds of nodes - when creating APIs to provide medical records data, complex banking information, as in FinTech or other highly regulated industries.

Developers can use LLM-based tools to simplify this task by automatically pre-matching data. By analyzing the technical specifications of all systems involved, repetitive tasks of matching data fields can be avoided. This automation speeds up the process and ensures error-free data matching.

Simplifying complex data formats

Over time, LLMs learn and become experts in industry-specific complex data formats such as legacy Electronic Data Interchange (EDI) for supply chain integration, HL7 commonly used in healthcare, SWIFT in banking, or any other non-public specific industry segment requirements.

The LLM's ability to handle the various data formats used within an organization also allows it to contribute to the creation of a canonical data format - a requirement often desired by the business but never fulfilled by IT due to the complexity of the various data elements, nuances and relationships. An LLM can easily analyze multiple formats and propose a common data language for use across the organization. Businesses can use AI to quickly decompose documents and convert them into a more modern or widely used format, which helps in data management and integrity.

From data models to API documentation

A critical step in the API lifecycle that is often overlooked is documentation. Artificial intelligence tools can automatically generate and compose initial API specifications based on open standards such as OpenAPI using existing data models.

Due to time constraints, developers often neglect documentation, which becomes inconsistent with code and reality, rendering initial efforts useless. AI, along with LLM, can automatically review and update documentation and point out inconsistencies. Good documentation makes API endpoints more accessible to consumers and makes it easier for developers to update and maintain.

Train your team and model

The LLM is a very powerful tool, but like any other tool, it requires training to make the most of it. Companies must provide training and documentation so that their teams know best practices for optimal results. And because LLM can be trained on your data set, companies can specifically train their models based on data, code, documentation, and company guidelines, ensuring that results are better aligned with company standards.

Limitations

While LLMs have significant potential for the design and development of APIs, they also have a number of limitations. AI approaches are highly dependent on the quality and variety of training data, which may not always reflect the full range of real-world usage scenarios or boundary situations. This can lead to errors in API design proposals, generated mappings or their implementation. Artificial intelligence typically requires assistance in understanding complex business requirements and creating accurate code, which can lead to the development and implementation of APIs that do not fully meet the needs of end users.

Conclusion

As the range of applications expands, companies must become increasingly confident in creating and utilizing APIs. LLMs can play a key role in solving the problem of scale for the engineering team by offering tools that accelerate the development process and improve accessibility and accuracy. Companies use an average of 230 applications, and for very large organizations, this figure rises to thousands. With AI, companies can keep pace with the growing number of applications in use.