CBA has built a multi-agent workflow that covers all the steps needed to prepare applications for migration to the cloud, allowing banks to move “20-30” applications per quarter.
CBA's Ash Moulin.
The workflow is branded as “Lumos,” and is essentially a rethink of the application modernization and migration process “AI-First.”
“Twelve to 18 months ago, I was standing in front of my team and said, 'We're trying to move X-number applications to the cloud.' And everyone on my team, including myself, found the task to be challenging.”
“So, how did we think about it [could] Apply engineering-driven approaches and ideas to how applications are migrated to the cloud. What does it look like to migrate this AI-first, agent-first, smart way?
“Essentially, all of that step [migration] Journey, we put the AI first and the agent first in the way that put it together. ”
This work, carried out in collaboration with AWS' Professional Services Arm Proserve, was born as a way to make the standard AWS migration framework “CBA-related” and increase the amount of possible application migrations in a given time frame.
“Last year, my team migrated around 10 applications in a year. Our ambition is to do more,” Mulan said.
“We ended at the end of this quarter and we are moving to ballparks with 20-30 applications per quarter.”
Mulan said over the past year, his team scoped “370 Legacy Applications” as a potential modernization and immigration candidate.
“We did a lot of application evaluations, but that wasn't easy,” he said. He likened the process to “documented archaeology” and understood it enough to understand enough applications to advance modernization and the cloud migration.
Aside from fragmented documents, the slides showed other issues that delay CBA application modernization and cloud migration efforts. This involves cloud readiness assessments that take more than six weeks per application, as well as the time and effort required to improve technical debt.
Lumos – built into the bank's cloud migration portal – is designed to help with this – using AI agents to handle specific steps, humans carry them in the loop, provide monitoring, and request additional changes if necessary.
AI agents are used for workflows throughout the assessment. Requirements gathering, networking, code analysis, solution architecture and cyber documentation generation, code transformation, UI testing, continuous feedback steps.
To get started, the agent summarises and analyzes conversations with the application owner, drawing out key points and converting them into requirements.
Mulan said this ensured there were no misunderstandings or mistakes in collecting candidates for transition.
Another agent is used to analyze and understand the application's codebase and “what this application actually does, various features, features, and how [these are]and “network connectivity discovered as part of its code analysis”, including API calls, databases, storage connections, etc.
This determines how your application is cloud prepared and highlights potential issues in code that require repairs or dependencies that require upgrades.
Moollan said that AI-generated component diagrams and sequence diagrams have also been created to show how the application works.
Once an application is understood and determined to be a candidate for modernization and cloud migration, “many agents” work together to create design and cyber documents.
Writer agents ensure that they are relevant to the CBA by performing checks against the bank's technical knowledge base, and “review content written by the first agent, organize it among the workflow agents, coordinate it between the workflow agents to improve and upgrade the quality of what is being produced,” Moollan said.
Mulan pointed out that there are also people in the loop who are challenged by what was written and who are even more refined.
The results are fed to another agent “trained in cyber knowledge trained” to prepare a “diagram of a security zone deploying resources to a security zone within an organization” and generate the required documentation.
In addition to understanding what an application does with an agent and how to migrate it to the cloud, agents are also involved in making changes to prepare their applications.
Agents “prepare the codebase” for work performed by other agents.
“There is an agent that uses open source Openrewrite [auto-refactoring engine] It does code conversion and framework upgrades and has an AWS Q Developer CLI [which] These two work hand in hand to upgrade that application, as we look at the code that OpenRewrite generates and make changes if there is a problem,” says Moollan.
Mulan said that many iterations are usually required, and changes were accepted and integrated into human surveillance and codebase.
Once Lumos is doing the work, use the DevOps hosting platform or DHP to deploy the application to cloud and “day 2” operations.
DHP has landed the CBA previously itnews Benchmark Award for the 2024 Finance Award.
Mulan said through Lumos that engineers can access “DHP's visual designers who create cloud environments in a CBA-compliant drag-and-drop manner.”
Banks continue to optimize their Lumos workflow.
Moolan said one of the optimizations is to change the multi-agent orchestration that enhances the underlying technology.
“We previously experimented with Kuruwai and Pidantic as agent orchestration frameworks,” he said.
“AWS just announced [Amazon Bedrock] AgentCore, so we are now working very heavily in the background and moving from Pydantic to Agentcore. ”
In addition to allowing more agent orchestration, Moollan has flagged Lumos' ability to handle “large legacy codebases” and the addition of more unit and system tests.
Ry Crozier has joined AWS Financial Services Symposium as an AWS guest.
