1. Netsmart leveraged agentic AI to modernize over 2 million lines of application code
Framework upgrades tend to rank low on the list of exciting engineering tasks.
They may also be precisely the type of work where AI excels.
Netsmart, a healthcare technology firm, built an agent-assisted modernization workflow for several large Angular applications using AWS Transform, Kiro, and Amazon Q Developer.
This was not a minor trial.
Across the three systems covered in the case study, the codebase surpassed 2 million lines.
Five engineering teams applied the process to different products, spanning straightforward Angular upgrades as well as complex multi-version migrations.
One migration involved moving more than 100,000 lines of AngularJS code to Angular 21.
The results reported are considerable:
- roughly 90% less effort for routine Angular upgrades
- around 85% less effort for more involved migrations
- work previously scoped at several months finished by individual engineers in about two weeks
- one migration originally estimated at 18 months reportedly wrapped up in five
What makes this case notable
Here, AI clearly goes past autocomplete.
Responsibility was divided across multiple tools in the workflow.
AWS Transform took on large-scale transformation and breaking changes.
Kiro assisted with validating functionality and UI consistency.
Amazon Q Developer was embedded in the development and code-review workflow.
The framing shifts from:
AI writes code.
to:
AI executes a repeatable engineering process.
That difference is significant.
Read the original Netsmart case study on AWS
2. DXC moved Angular 9 and Angular 15 applications up to Angular 20 using Amazon Q Developer
DXC Technology published another notable modernization case in 2026.
Its customer ran an enterprise platform with more than 80 backend microservices and multiple Angular microfrontends.
A number of frontend applications were still on Angular 9 or Angular 15.
Rather than treating the upgrade as a strictly manual migration, the team turned to Amazon Q Developer to help:
- surface dependency conflicts
- swap out deprecated APIs
- debug build failures
- propose compatible package versions
- fix security issues
- support the transition to Angular 20
Crucially, AI-produced changes still had to pass standard engineering checks.
Those included unit tests, integration tests, regression testing, SAST, and DAST.
The reported productivity gain is substantial.
DXC estimated that upgrading a single service by hand took about 16-24 developer hours.
With Amazon Q Developer, that figure reportedly fell to 4-6 hours.
For the wider modernization program, DXC calculated a 60-75% reduction in upgrade effort, which works out to roughly 1,200-1,400 developer hours saved.
What makes this case notable
DXC did not start by turning an entire platform over to an agent.
The team began with a small proof of concept, measured the results, and then scaled the workflow up.
That approach is likely far closer to how AI should be adopted in large Angular codebases.
AI took care of repetitive transformation and troubleshooting.
The existing engineering system stayed responsible for proving the outcome was correct.
Read the DXC case study on AWS
3. A controlled enterprise study showed AI cutting delivery time by as much as 69%
Vendor case studies have their value.
Controlled experiments are arguably more compelling.
A 2026 research paper examined design-system-aware AI development at a large Brazilian enterprise.
Three approaches were compared in the experiment:
- manual implementation
- development with a design system
- development with a design-system-aware AI workflow
The study spanned Angular, iOS, and Android development.
Over two experimental cycles, AI-assisted development cut time-to-delivery by 46.7% to 69.4%.
Researchers also noted higher task completeness, less variation between developers, and better design consistency.
What makes this case notable
This may point toward one of the more significant directions for AI-assisted frontend work.
A generic model understands Angular.
A model wired into your:
- design system
- components
- constraints
- documentation
- architectural patterns
understands how your organization builds Angular.
That is far more valuable.
The interesting future may not be AI churning out arbitrary components faster.
It may be AI making it harder to produce a component that violates the architecture in the first place.
One important caveat applies here.
The 46.7-69.4% improvement is reported across Angular, iOS, and Android combined, so it should not be read as an Angular-only figure.
Read the research paper on arXiv
4. Issuer Direct used generative AI to guide a legacy Angular 9 application toward Angular 20
Another modernization story comes from Issuer Direct, which builds corporate communications and compliance software.
Its real-time reporting application was on Angular 9 and had built up significant technical debt, legacy dependencies, and more than 180 active security vulnerabilities.
Instead of attempting a single massive migration, DBServices describes starting from a fresh Angular 20 application and shifting functionality over incrementally.
Generative AI supported the refactoring effort throughout, speeding up repetitive work and standardizing code while engineers validated the output continuously.
The modernization also brought in modern Angular patterns, including:
- Standalone Components
- Signals
- the new control flow syntax
Per the case study, the project cleared the identified security vulnerabilities while lowering architectural coupling and improving real-time rendering performance.
What makes this case notable
A pattern is emerging across several of these stories.
AI appears especially effective when the destination is well defined.
Going from:
legacy Angular
to:
a known modern Angular architecture
gives an agent a concrete target to optimize for.
The engineer still determines what the target architecture should be.
AI makes much of the mechanical work in between less expensive.
Read the Issuer Direct case study
5. Quantatec embedded an LLM in an Angular product and shipped to production in about six weeks
Not every Angular + AI story involves AI writing Angular.
Sometimes Angular is simply the application layer users go through to reach AI.
Quantatec offers fleet-management software that processes data from more than 10,000 devices.
Customers kept asking for more customized reports.
That translated into more database queries, more report variations, and mounting strain on the existing analytics architecture.
Rather than building yet another dashboard configuration tool, Quantatec created Movias AI.
Users could pose questions about fleet data in natural language, and an LLM converted those questions into requests against the company's analytics layer.
The stack consisted of:
- Angular
- PostgreSQL
- Azure Kubernetes Service
- Azure OpenAI
- LangChain
According to the case study, the system went from proof of concept to production in roughly six weeks.
English, Spanish, and Brazilian Portuguese were among the languages evaluated.
What makes this case notable
This is arguably the most conventional AI application on this list.
And that is exactly what makes it one of the most useful examples.
The Angular frontend is not the AI.
It supplies the deterministic product surrounding a probabilistic system.
Authentication, navigation, state, data visualization, permissions, and interaction continue to behave like normal software.
The LLM turns into another capability within the application rather than a replacement for its architecture.
That distinction is likely to stay important even as models grow dramatically more capable.
A common thread runs through all five
These examples differ widely.
Yet they all lean in a similar direction.
The strongest results do not come from merely asking:
Can AI write this Angular component?
They come from supplying AI with context, constraints, and a verifiable process.
Netsmart handed agents a repeatable modernization workflow.
DXC paired AI with testing and security pipelines.
The enterprise design-system study tied AI directly to established UI constraints.
Issuer Direct mixed generation with ongoing human validation.
Quantatec put an LLM behind an application architecture built around real business data.
AI turns out to be far more useful when the engineering system around it is stronger.
That is the more interesting shift taking place in software development.
We spent decades building abstractions that reduced how much low-level work humans had to do.
Frameworks were one such abstraction.
AI is simply the next one.
Bonus: a useful counterexample - AI did not do everything
There is another 2026 Angular migration worth reading precisely because it is less dramatic.
Brainence documented moving a 200,000-line application with roughly 570 components from Angular 12 to Angular 21.
AI was used throughout the migration, but the team describes it as most helpful for narrow, verifiable tasks:
- generating codemods
- finding obsolete CSS selectors
- translating deprecated APIs
- explaining breaking changes
Architectural decisions, dependency strategy, and validation stayed with the engineers.
Delegate the mechanical work, not the decision-making.
Read the Brainence Angular 12 to Angular 21 migration story
What lies ahead?
The interesting question is not whether AI can produce Angular code.
Of course it can.
The more significant questions are:
- Can it grasp the architecture of a large existing system?
- Can it follow the same conventions as the rest of the organization?
- Can it carry out migrations without slipping in subtle regressions?
- Can it operate against a company's actual design system?
- Can we reliably verify the changes it makes?
The case studies above suggest the answer is increasingly yes - but only when the AI works inside a strong engineering process.
The tools keep getting more capable.
That does not make architecture, testing, code review, or engineering judgment any less important.
It makes them the constraints that let AI operate at a much larger scale.
