How Kingland Runs Code Migrations 10x Faster with AI Agents on AWS

Kingland builds enterprise data management software for financial services. Their codebase is 15 years old. Dependency upgrades used to take engineers half a day each. Legacy code had no documentation. Compliance audits required a paper trail for every change. When Kingland deployed Ona's AI agents on AWS, a Jest v30 migration that would have taken five hours of engineering time was done in 30 minutes. The agents now handle dependency upgrades, code refactoring, and legacy documentation across the entire codebase while meeting the audit requirements of regulated financial services.

Key outcomes:

1. 10x faster migrations — A Jest v30 migration completed in 30 minutes instead of ~5 hours of engineering time.
2. 60% AI co-authorship — 60% of commits now include AI co-authorship, with 1,200 agent executions per month across 67 unique users.
3. Legacy codebase documented — 15 years of undocumented Java code receiving agent-generated JavaDocs, reviewed by engineers.
4. Full compliance maintained — Audit trails for every agent action, running entirely within Kingland's AWS VPC via Bedrock.

Company

Kingland is an enterprise data management provider serving financial services clients including banks, asset managers, and insurance companies. The company employs roughly 440 people, with about 100 engineers working across a Java and JavaScript codebase that has been in active development for over 15 years. Patrick Rice, Enterprise Architect, leads engineering and research with responsibility for developer productivity.

The challenge: a 15-year codebase that resists change

Kingland's software undergoes extensive third-party audits annually. The codebase is large, old, and carries the kind of accumulated weight that makes every change expensive.

Dependency upgrades were manual and slow. Each migration (a framework version bump, a test library upgrade, a build tool update) required an engineer to read the migration notes, understand the breaking changes, update the code, fix the tests, and verify everything worked. A single dependency upgrade could take four to five hours of focused work. Multiply that across dozens of dependencies and hundreds of files, and the backlog grew faster than the team could work through it.

Legacy code was undocumented. When Kingland implemented CheckStyle requiring JavaDoc comments on all public methods, the team faced the reality of 15 years of code with minimal documentation. Nobody was going to volunteer to write JavaDoc comments for thousands of methods they didn't author.

Environment setup ate days. Kingland's development setup had evolved from Windows machines distributed as VMware images, to HashiCorp Vagrant with centralized VM images. Performance issues persisted. Engineers with 32GB laptops still hit resource limits. Patrick Rice described the cost: developers were spending significant hours per year fixing development environments, often losing upwards of two days setting up their environment when something went wrong.

Compliance required audit trails for everything. Financial services clients expect to know who made every change and why. Any tool that touched the codebase needed to produce a clear record of what happened, who initiated it, and whether a human reviewed the result.

The solution: AI agents in secure, ephemeral environments on AWS

Kingland deployed Ona in two phases. First, they replaced Vagrant-based local development with standardized cloud environments, eliminating the setup problems entirely. Then they adopted Ona's AI agents to tackle the migration and modernization work that had been sitting in the backlog.

The architecture runs entirely on AWS:

• AWS Bedrock for LLM processing. All AI inference happens through Kingland's own AWS account. No code or data leaves their network. Patrick Rice noted the security advantage: AWS Bedrock models have a much stricter security stance than what you see from model providers directly.
• Self-hosted runner deployed in Kingland's VPC, with VPN and IP allowlisting. The runner operates within their existing security perimeter.
• Ephemeral environments for each agent execution. Every AI agent runs in an isolated, disposable environment that is destroyed after use. The philosophy: the best way to handle security is to not have environments that are alive for long enough to get compromised.
• Full audit trails tracking who initiates every action and what changes are made, critical for keeping track of the audit trail in a regulated environment.

Migration and modernization use cases

Kingland uses Ona's AI agents for three categories of work that previously sat in the backlog because the scope was too large for engineers to tackle manually.

Dependency migrations

When Kingland needed to migrate to Jest v30, an engineer provided the migration notes to an Ona agent and walked away. The agent read the notes, updated the code, fixed the tests, and produced a pull request. The Jest v30 migration was completed with about 15 to 30 minutes of actual engineering interaction, compared to what would have taken an engineer four to five hours.

The pattern is now repeatable. Engineers provide migration notes, the agent executes the upgrade, and the engineer reviews the result. Kingland uses Renovate to identify dependency updates, and agents handle the actual migration work.

Code refactoring

Kingland found that separating test code from production code gives agents a natural safety boundary. The approach: ask the agent to either refactor test files without changing production files, or refactor production files without changing test files. Both use cases work well because the existing test suite acts as a verification layer. If the agent refactors production code and the tests still pass, the change is likely correct. If it refactors tests and they still pass against unchanged production code, the tests are still valid.

Legacy code documentation

When CheckStyle enforcement required JavaDoc on all public methods, the team used agents to document the 15-year-old codebase package by package. Engineers provided examples of good and bad comments, pointed the agent at a package, and reviewed the output. Work that no engineer wanted to do, and that would have taken weeks of manual effort, moved through the backlog steadily.

Results

Beyond engineering, tech writers and data scientists at Kingland have started using Ona for documentation and SQL work.

Key takeaway

Kingland had a 15-year codebase, a backlog of migrations that grew faster than engineers could clear it, and compliance requirements that made every change expensive to track. Ona's AI agents, running on AWS infrastructure, turned migration and modernization from manual, engineer-by-engineer work into an automated, auditable process. The Jest v30 migration is the headline number (10x faster), but the broader shift is that dependency upgrades, code refactoring, and legacy documentation are no longer work that sits in the backlog. They're work that agents do while engineers review.

The innovative project is implemented using Ona's Enterprise solution, built on top of Amazon Web Services (AWS) using AWS-designed highly performant AWS services such as Bedrock. The composability on AWS is crucial to deliver the highest level of flexibility for Kingland. It allows Kingland to select the best-of-breed components they need, which will work seamlessly and reliably straight away, thanks to Ona and the high levels of service AWS has. Due to AWS, the onboarding of new developers takes as little as a few minutes, compared to months previously. With the Ona solution, built on AWS, Ona can scale computational and storage resources to accommodate Kingland's goals.