AIToday
Large Language ModelsAI Coding AssistantsAI Business & IndustryZenn AI/MLPublished: Nov 15, 2026, 16:00 JST

ProAg shifts legacy .NET Framework 4.8 process in 5 weeks with AI

Send on LINE
ProAg shifts legacy .NET Framework 4.8 process in 5 weeks with AI

3 Key Points

  1. What happened

    ProAg moved one of its most complex legacy processes from .NET Framework 4.8 to .NET 8 on AWS EKS using AI agents in about five weeks, versus over six months per process internally, with 100% data match.

  2. Why it matters

    The result suggests AI agents can be trusted on a high-stakes migration only because a validation harness proved output parity at every stage boundary, not just at the end, so the proof of equivalence—not the code generation—is the reusable asset.

WHO IT HITSLegacy modernization teams at regulated insurers and financial firms, who must prove line-by-line equivalence before retiring old systems, are the clearest audience for this approach.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

ProAg's legacy estate is typical of specialty insurance: more than a decade of business logic accumulated in Windows-only .NET Framework libraries, with hundreds of processes to move. The internal team's pace of over six months per process implied a multi-year program, which is why the company brought in an outside partner and targeted one of the hardest processes rather than an easy one.

The choice of the hardest process appears deliberate: a success on a trivial task would not answer whether the method works on the work that actually has to be done. The architecture also had to clear a regulatory bar first—US crop insurance is heavily regulated, so the team ran Claude via Amazon Bedrock inside ProAg's own AWS account, keeping every model call under existing controls.

The technical sequence matters as much as the speed. Before code was touched, agents mapped dependencies into a graph that subject-matter experts reviewed, and a separate agent drafted a migration plan that flagged Windows-only libraries for replacement or rewrite. Migration then proceeded stage by stage, with a validation harness built alongside the migration rather than after it, so mismatches could be localized early. A final field-level comparison on production-like data became the formal evidence ProAg keeps. The public case is explicit that this does not mean production readiness without further hardening, so the five-week figure should be read as a pilot result—the reusable parts being the harness, documentation steps, and agent templates.

FAQ
How did ProAg handle regulatory restrictions on sending code to external AI services?
ProAg called Claude through Amazon Bedrock from within its own AWS account, so all model calls stayed under its existing controls and regulated code and data did not leave its boundary.
What did ProAg build before touching the code?
In week one, AI agents produced an ontology graph of dependencies across classes, stored procedures, data stores, and feature flags, which business subject-matter experts then reviewed for correctness.
How was output equivalence verified?
The process was split into four stages—data retrieval, filtering, calculation, and report generation—and old and new versions were run on identical inputs at each boundary until outputs matched. A final field-by-field diff on production-like data was kept as an official record.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleLlama 3.2 1B on QCS6490: 1,024 tok/s prefill, 16.4 tok/s decode