Use case · AI-driven development

Bring your largest applications into AI-driven development.

Every other application gets AI-driven development. The largest and most critical ones are left out because AI coding tools are missing the inputs they need. vFunction gives them the specification, the tests, and the system model along with FDEs to drive the change.

What AI-driven development needs

Three inputs a large existing application doesn't have

AI-driven development (AI-DLC) needs three inputs. New applications produce them as they’re built. Large existing applications don’t have them, and code alone can’t supply them.

  • A specification of what each domain is meant to do
  • A model of how the system is structured and used
  • Tests that verify each change didn’t break behavior
The system model

How the application runs in production

vFunction observes the running application: actual structure, critical dependencies, which code paths execute and which never do, where the architectural debt sits.

  • Runtime analysis and proprietary data science, not code alone
  • Your architects set domain boundaries and the target architecture from it
Specifications

A specification for each domain, backed by evidence

Built from the runtime flows vFunction observed, with the architectural issues found during analysis factored in. Nothing inferred or guessed.

  • One specification per domain, in the order your team sets
  • Generated code doesn’t reintroduce the issues flagged for that domain
Integration tests

Proof a change didn't break behavior

Integration tests generated from production traffic, per domain. An application that never had tests gets substantial coverage before the first change.

  • Derived from what users do, not what the team assumed
  • Run after each rewrite to verify behavioral equivalence
Your AI coding tools

The tools write the code. vFunction gives them context.

Kiro, Claude Code, AWS Transform, and other AI coding tools get specifications, tests, and guardrails, and generate code one domain at a time under review.

MCP server and skills

Kiro, Claude Code, and other tools connect over MCP; a downloadable skill generates code from each specification.

How the MCP server and skills work →

AWS Transform

Used within the same workflow for .NET and Java framework upgrades.

How it works with AWS Transform →

Your team's review

Every change is proposed against a specification and checked by the tests before the next domain.

How an engagement works →
Let's talk

Tell us about the application.

Java or .NET, roughly how large, where it runs, which AI coding tools you use. We come back with what vFunction would generate for it.

What happens next

  1. A person reads it — not a scheduler.
  2. You hear back with pricing and the funding path.
  3. If it fits, a conversation with an engineer.

A person replies — not an automated sequence. We use what you send only to reply about your application.

How it runs

vFunction's engineers run it with your team

In your environment, on the platform and your AI coding tools, one domain at a time, until your team can run it alone.

  • Interpret the analysis; set domain boundaries and target architecture with your architects
  • Write and refine the prompts; review every change with your team
  • Transfer the process as your team builds proficiency

How an engagement works →

Questions about AI-driven development on existing applications

Does vFunction write the code?

No. Kiro, Claude Code, AWS Transform, or your tool of choice does. vFunction gives it the specification, the tests, and the context.

Which AI coding tools does it work with?

Kiro and Claude Code (CLI or VS Code), AWS Transform for Java and .NET, and anything that connects over MCP.

Can't the AI coding tool build the specification from the code?

Code shows structure, not which paths run or which dependencies matter. Those are facts about production, so vFunction observes the running application.

We already have tests. Why generate more?

Tests written from assumptions cover what the team assumed. Tests generated from production traffic cover what users do. Most large applications need both.

Who decides the domain boundaries and target architecture?

Your architects do. vFunction proposes a structure from the evidence; your team sets the boundaries and decides what to preserve, change, or retire.

Which applications is this for?

Java or .NET applications past roughly 500,000 lines of code or 1,000 classes. Below that, an AI coding tool alone is usually enough.

Does vFunction run AI-driven development for us?

No. vFunction's engineers run the process with your team, in your environment, until your team runs it alone. See How an engagement works.

Is this only for applications on AWS?

No. vFunction installs in your environment and works with the AI coding tools you use. For applications on AWS, vFunction is often funded by AWS.

How is vFunction priced?

Per application, no per-user fees, often funded by AWS for applications on AWS. See Pricing.

Where to go next

Keep going

vFunction 5.0 brings large existing Java and .NET applications into AI-driven development: specifications and integration tests generated from the running application, and the database migrated with it. See what’s new in vFunction 5.0 →

The platform →

Full-Stack Windows Modernization →

How an engagement works →

Pricing →

Customer stories →

About vFunction

vFunction is a modernization platform for large existing Java and .NET applications. It gives AI coding tools such as Kiro and Claude Code the context and guardrails they need to refactor, rewrite, and modernize existing systems with speed, accuracy, and control: runtime architectural context, a target architecture your architects define, and specifications and integration tests generated from how the application actually runs in production. vFunction is built for applications past roughly 500,000 lines of code or 1,000 classes.

Tell us about the application

Java or .NET, roughly how large, where it runs, which AI coding tools you use. We come back with what vFunction would generate.