GenAI for software modernization: upgrading legacy systems

The potential for GenAI 

Previously, machine learning (ML) was a specialized branch of AI that allowed data scientists to create predictive models for business use. GenAI is an emerging branch of AI that democratized the development and deployment of AI capabilities over a common set of APIs that any software engineer can understand and leverage. These new GenAI models have lowered the barrier to adopt this technology and broadened potential use cases and GenAI’s potential impact.

Early GenAI models were large language models (LLMs) trained on extensive text data sets. LLMs can comprehend language, predict likely sequences of words and generate new text. However, the applications for GenAI have exploded beyond these early use cases. One significant area of recent interest has been the adoption of GenAI in software development.

GenAI can be applied as easily in the development of standardized user stories as in code development, code review and test generation. At the moment, humans still need to verify and validate the outputs of GenAI, but there is potential for GenAI to significantly reduce time and costs associated with software development.

Current challenges with mainframe modernization and cloud migration initiatives

Mainframe systems have been the backbone of many enterprises for decades, providing reliable, secure and scalable performance for mission-critical applications. However, as the world shifts to a more digital, cloud-based and agile environment, legacy mainframes face several challenges that hinder their ability to meet the evolving needs and expectations of customers, partners and regulators.

To address the dynamic IT landscape, mainframe modernization and cloud migration have been frequent initiatives and investments among enterprises year after year. Unfortunately, these projects are frequently complex and expensive. While fully operating in the cloud is the goal for many enterprises, the path may be long and may require significant and specialized labor to achieve.

The cost of inaction: risks of maintaining the status quo

Mainframe modernization and cloud migrations have been in-progress for at least the last 10 to 15 years across various industries. At this point, some of the most technically complex challenges remain, including technical debt, limited interoperability, resource constraints, security vulnerabilities and regulatory compliance issues, with a sizable footprint for both modernization and migrations. Additionally, the talent capable of addressing them is increasingly aging out. 

While the technical debt landscape hasn’t changed significantly with GenAI, the rate of change in addressing technical debt has accelerated. Originally, the ability to tackle challenging technical debt was a function of quantity and capability of technical talent. With the emergence of GenAI, junior technical staff can quickly become proficient and contribute faster, and senior technical staff can gain context on and address larger and more ambiguous problems quicker as well.

For enterprises that have lagged in addressing technical debt in the past, slow adoption of GenAI-enabled tools and approaches will further increase the gap with competitors doing both. For many use cases, but especially for developers, GenAI will be a force multiplier for enabling professionals to get more done with less time and effort.

Real-world use cases and emerging trends 

Leveraging GenAI and collective intelligence, organizations have the opportunity to support and automate the developer experience now. While LLMs were originally designed for natural language text generation and summarization, they can also understand software languages and explain and extrapolate software functionality.

Potential GenAI use cases across the software development lifecycle (SDLC) are emerging to improve software quality and velocity for developers in:

Design — While planning and architecting software, GenAI tools can aid in recommending a target state architecture, producing technical designs and documenting existing systems.

Development — During development, coding copilots and similar tools improve efficiency when writing application and infrastructure code. They can also guide developers during code conversion between languages and versions.

Emerging uses of GenAI models by software developers fall into three broad categories within the development section of the SDLC: 

1. Developer assistance — A developer uses GenAI tools as a part of their workflow. For example, GenAI copilots are currently leveraged to provide developer assistance by suggesting code snippets as developers write, helping streamline the coding process and embed necessary controls aligned to risk and compliance policies. 
2. Developer automation — A GenAI-enabled system generates code that developers review. Code generation platforms leverage GenAI to automate routine coding tasks, such as bug detection and fixing, and generate new code in response to developer queries. This assists developers in improving code quality and efficiency throughout the SDLC.
3. Developer documentation — GenAI is used in developing technical documentation, exemplified by its application for mainframe migration projects. By autonomously generating human-readable inline comments for previously undocumented legacy code (e.g., COBOL), GenAI facilitates re-engineering efforts into modern languages like Java. This significantly accelerates delivery timelines and supports ongoing modernization endeavors. Future iterations aim to integrate seamlessly with tools like GitHub Copilot, further enhancing development efficiency and effectiveness.

Testing — After software is written, GenAI solutions can not only automate test generation, but they can also automatically review pull requests for bugs and security issues.

Monitoring — GenAI solutions can also be used to assess the security of existing code, monitor logs to detect current issues or inefficiencies, and monitor for security incidents.

There are many opportunities for GenAI assistance and automation to increase developer efficiency. Both major tech providers and startups are building GenAI-enabled solutions across the SDLC. While the space is exciting, the rapid pace of innovation has resulted in a lack of consistent standards and cross-platform compatibility for products. And choosing a tool may imply choosing an entire vendor ecosystem. While this was partially true when considering cloud providers, this is even more pronounced in the current state of GenAI tools.

Code conversion and upgrades pose new and exciting opportunities when harnessing the power of GenAI. EY teams have conducted various successful pilot programs to help enterprises move toward a target architecture:

• SAS to PySpark conversion pilot — This initiative saw the seamless conversion of 4,000 lines of SAS (Statistical Analysis System) code into the more dynamic PySpark environment, marking a 50% leap in efficiency. With a conversion accuracy of 85%, the pilot demonstrated intelligent processing and optimization and was completed within three weeks.
• PostgreSQL to Google BigQuery migration — Over 600 PostgreSQL queries were migrated to Google BigQuery through GPT-4 assisted modeling with 90% accuracy in conversion. This four-week project not only optimized data warehousing processes but also enhanced real-time data analytics capabilities, yielding a significant 60% increase in efficiency.
  
• PL/SQL to Spark SQL transformation — The conversion of PL/SQL to Spark SQL, facilitated by GenAI, culminated in an 85% conversion accuracy and a 35% efficiency gain. This two-week sprint showcased the EY organization’s agile methodologies and AI engineering capabilities, setting a new standard for database transformation and management.
• Accelerating legacy code documentation — GenAI actively develops technical documentation in EY programs, where it generates human-readable inline comments for code written in a legacy programming language (e.g., COBOL), expediting mainframe migration to more modern languages, like Python and Java. This initiative, known as EY Code Assist, has accelerated modernization efforts and uses a private OpenAI instance to ensure EY controls data security.

These case studies show GenAI is not the future — it’s the present. GenAI is reshaping the landscape of enterprise IT infrastructure. 

What this means for your business

To take advantage of GenAI for modernizing and transforming your organization’s mainframe, consider the following three steps to get started:

Evaluate — The current landscape of GenAI solutions presents a diverse array of options, but no one solution fits all developer needs. Moreover, interoperability between GenAI solutions from different providers remains a challenge. Enterprises seeking to leverage GenAI effectively must first assess their existing capabilities and infrastructure. Teams equipped for safe and agile experimentation will excel in evaluating different tooling ecosystems to find the most suitable solutions.

Experiment — Innovation teams within enterprises can establish isolated innovation sandboxes within cloud environments. These environments facilitate semi-independent operation, enabling teams to experiment freely without the constraints of procurement, vendor evaluation and security reviews. Collaboration between innovation teams and various business, tech and security groups is crucial for refining chosen tools to meet enterprise-wide needs.

Scale — For enterprises facing stringent constraints, such as security and procurement policies that hinder rapid iteration, outsourcing innovation and experimentation becomes a viable option. By leveraging external expertise, enterprises can navigate complexities and adopt GenAI solutions at a scale that aligns with their operational requirements.