Turning Undocumented Systems Into Usable Knowledge With AWS
How enterprises can use AI and cloud services to generate usable documentation from legacy systems, supporting onboarding, maintenance and long-term modernization
Whether an enterprise decides to migrate its on-premises workloads to an entirely cloud-based system or build pipelines to IBM and Power mainframe systems to pull data, the process requires specialized technical knowledge that may be lacking from its existing documentation.
Incomplete or missing documentation for older COBOL programs and business flows is often raised as a challenge for enterprises attempting to modernize, making it difficult to bridge the gap with newer, API-driven workflows. A Stack Overflow survey of software developers found that “disparate and uncoordinated documentation that few have ample time afforded to complete” is a source of frustration.
But AI is changing how organizations extract and reconstruct system knowledge. AWS can help bring documentation into the process using large language models (LLMs) and AI agents.
The Documentation Gap Is a Structural Problem
The technical details of how long-established systems run critical workloads—commonly with tweaks or quick fixes that account for small oddities or technical snags in day-to-day operations—often sit with long-tenured, or even former, employees.
There are many reasons why documentation of valuable knowledge might be missing. Systems evolved over decades, often under pressure to deliver projects, with developers layering changes on top of existing code rather than formally documenting them. That snippet of code a developer wrote in the 1990s in a core business application but was never fully documented has the potential to cause disruption when it comes time to refactor or migrate the system.
“That’s institutional knowledge, and most organizations haven’t codified it,” notes Michael Bevilacqua, VP of AI Product Management at data automation firm Adeptia. “Many of these systems have incomplete or non-existent documentation.”
Reliance on current employees simply understanding how specific setups work can prevent businesses from keeping on top of maintaining formal records.
While these systems work without major problems day to day, onboarding of new employees is slowed down, troubleshooting depends on a few subject matter experts, and small changes carry the risk of breaking workflows.
Even access to official documentation has not always been reliable. A redesign of IBM’s website back in 2018 temporarily broke links to z/OS resources, highlighting how easily embedded knowledge can become fragmented. And when documentation does exist, it may contain outdated Information that causes confusion when referenced alongside newer documentation.
Building an AI Documentation Pipeline on AWS
Better documentation is central to modernization projects. Many transformation efforts stall because organizations don’t fully understand the systems they are trying to change. As enterprises look to adapt their systems to the data demands of AI-driven applications, AI also has a role to play in tackling the documentation problem by making systems more understandable.
“Now with AI, we can generate a comprehensive set of test skills, covering edge cases and production data. We can create AI reports around translation, and most importantly, create a set of documentation that is absolutely useful,” Srikara Rao, CTO at R Systems, an AWS Advanced Consulting Partner, tells TechChannel.
“Those are the kind of changes coming in where you are building an architecture based on AWS services.”
AWS Transform, its agentic AI service, can help uncover application structure and logic, while services like Amazon Bedrock generate explanations that make complex logic more accessible to engineers. Bedrock can use LLMs to parse code including COBOL, generate human-readable explanations of how applications work and identify hidden core dependencies and business rules.
Enterprises can build Bedrock agents to provide automated updates and dynamic support, interacting with users to understand their queries, analyze the documentation and take action to provide the information needed, such as looking up policies or API details.
This can bring documentation into data pipelines, in which code, metadata and system data is aggregated into platforms like Amazon Simple Storage Service (Amazon S3), integrated and prepared using the AWS Glue data management service, and then analyzed by AI via Amazon Bedrock.
Developers can extend this by using tools such as AWS Kiro to generate specifications, tests and supporting materials directly from code and prompts. They can embed documentation into DevOps workflows and observability systems, ensuring it evolves alongside code changes.
The result is that teams can move faster by working from searchable knowledge bases, opening up access to that valuable context stored in employees’ heads. Refactoring becomes less risky because assumptions can be tested against documented behavior, and modernization efforts are less likely to fail because of overlooked dependencies or misunderstood logic.
Why Human Expertise Still Matters
But AI is not a cure-all here. It is fast, but not always right. Using AI can reduce the time it takes to get documentation in order from weeks or even months, but that does not eliminate the need for human validation.
As Bevilacqua tells TechChannel, “AI-powered inference can compress that to hours, but you need deep domain expertise to validate what the AI produces.”
AI can produce confident but incorrect interpretations, creating misleading documentation and potentially leading to incorrect assumptions in transformation projects.
Subject matter experts remain essential to the process to validate the outputs from the models, confirming that descriptions, mappings and inferred logic reflect how systems actually behave. Without that validation, there is a risk of undocumented systems becoming incorrectly documented ones.
AI reduces the effort, but it does not entirely replace the institutional knowledge.
From Static Documentation to Maintaining Understanding
The value of AI-generated documentation extends beyond the initial project implementation. Systems need to remain understandable as they evolve, making documentation part of ongoing operations, not a one-off task.
By integrating document generation into cloud workflows, enterprises can keep system knowledge up to date as code, data pipelines and architectures change. On AWS, this means combining AI services such as Bedrock with data and integration layers like Glue and S3, so that documentation reflects current system behavior rather than a point-in-time snapshot.
The result is a shift in how systems are maintained. New engineers can onboard faster because knowledge is accessible. Teams spend less time reverse-engineering code and more time improving it. And as organizations continue to modernize, introducing micro-services, real-time pipelines and AI-driven features, documentation evolves alongside the system rather than falling behind it.