Rebuilding a VB6 loan platform with AI in eight months

A lending team cut the average time from opening a pull request to merge from 4.1 calendar days in April–May to 2.5 in July–October 2026. Working with Sigao, the team rebuilt a VB6 platform that handles hundreds of thousands of loan applications a month. The rebuild took eight months. AI usage cost about $5,000.

The platform led its market, but its VB6 core had become slow and costly to change.

  • Everything ran on VB6

    The full feature set lived in several million lines of VB6, behind a dated interface.

  • A hand rewrite was too slow

    The business wanted a faster, less costly path than rewriting all of it by hand.

  • Customers wanted more, sooner

    Customers across the United States process hundreds of thousands of loan applications through it each month, and they wanted new features faster.

  • Lending allows no shortcuts

    Lenders answer to regulators, so moving faster couldn't make the software any less secure or reliable.

The goal was to make the same team faster and show that quality held as it sped up.

We learned what the old system really did, then rewrote it for what users need now.

  • We mapped what the system actually did

    Our reverse engineering framework analyzed the VB6 code, the user documentation and recorded interviews with the client. It showed what the system was supposed to do and what it actually did.

  • We found where the skeletons were hidden

    Code built up over decades holds rules and workarounds nobody wrote down. We surfaced them before the rewrite started, so the team decided what to keep with the facts in front of it.

  • We rewrote it for today's users

    Converting the VB6 line by line would have carried its old limits into the new platform. We rewrote it around what users need now, and their feedback shaped every release.

We planned the rewrite one user problem at a time.

  1. Product brief

    Each brief stated one problem the platform's users had. Digital twins of those users and the reverse engineering findings shaped it.

  2. Feature plan

    A planner agent broke each brief into features and grounded them in Azure DevOps and the shared context.

  3. UX design

    Design agents produced high-fidelity mockups, and feedback from the client's users shaped each revision.

  4. Specification

    A spec planning agent turned each feature into a specification of the outcome the user needed. It lived in the repository, where engineers and agents read the same documents.

The team built each feature with agents and checked every change against its specification.

  1. Technical plan

    For each feature, engineers and planning agents wrote a technical plan. Architectural decisions went into the shared context, so later features built on them.

  2. Implement

    Engineers built in small steps with agents, in an environment defined with Microsoft Aspire so every engineer and agent ran the same setup.

  3. Verify

    Test planning agents derived a test plan from each specification. Branch protection and lint rules held every change to it, whether a person or an agent wrote it.

  4. Document

    A documentation agent reviewed each pull request and kept the user and technical documentation current as the new platform grew.

The VB6 application became a flexible workflow builder in eight months.

The rebuilt platform works at scale and gives the client's customers a modern interface for loan applications, workflows and report templates.

The report template editor on the rebuilt platform.

Pull requests reached merge sooner, and more passed testing and review with no findings.

The same team completed 373% more story points in its tenth sprint than in its first, using the same point scale. Over the comparison months, average monthly merges rose from 33 to 78, and a larger share of reviewed pull requests cleared testing and code review with no P0–P2 findings.

  • Story points completed per sprint

    +373%Sprint ten compared with sprint one, April to early November 2026. The dashed line marks sprint one. The slowest sprint after the second still finished 105% above the first.

  • Time to release a feature

    1.5 weeksDown from about a month.

  • Time from opening a pull request to merging it

    2.5 daysAverage calendar time, down from 4.1 days in April–May to 2.5 in July–October 2026.

  • Pull requests merged per month

    78Up from 33.

  • Pull requests that passed testing and code review with no issues

    1 in 4Up from 1 in 10. The team fixed every issue found before the code merged.

Velocity is story points per sprint across ten consecutive sprints, from April to early November 2026. Merge times, monthly merges and review results compare April–May with July–October 2026. Review results count P0 to P2 issues found by automated tests, regression tests and code review. Each reviewed pull request counts once, and pull requests that were not reviewed are left out. Chris Sims, Sigao’s Managing Partner, confirmed the merge-time and review figures on November 10, 2026.

Tell us where delivery is getting stuck.

If a legacy platform is slowing your roadmap, start with a conversation about how work moves through your teams today.