Imagine staring at a massive pile of ancient COBOL code from the 80s—millions of lines tangled like spaghetti, undocumented, and guarded by a handful of retiring experts who hold all the tribal knowledge in their heads. Modernizing that beast? It could take months or even years, costing a fortune in developer time and expertise. Now, picture doing it in days, with docs that are not just faster but 97% quicker to generate, 95% more comprehensive, and 60% more readable. That's not sci-fi; that's Fujitsu's Application Transform powered by Fujitsu Kozuchi, launched today, March 30, 2026, as a generative AI SaaS. [1]
If you're knee-deep in legacy system hell—think banks, insurance giants, or any enterprise running on COBOL dinosaurs—this tool is a lifeline. It auto-analyzes source code, spits out crystal-clear design documents, and paves the way for cloud migrations or rewrites without needing COBOL wizards. In this deep dive, we'll unpack the launch, how it works, the jaw-dropping stats from real verifications, and why it's a productivity rocket for devs everywhere. Buckle up; this could transform how we tackle enterprise modernization.[1]
What is Fujitsu Application Transform?
At its core, Fujitsu Application Transform is a cloud-based SaaS service that uses generative AI to reverse-engineer legacy source code into professional-grade design documents. Powered by Fujitsu's Kozuchi AI platform—their battle-tested suite for enterprise AI—it targets the nightmare of legacy systems, especially those written in COBOL and other outdated languages.[1]
Legacy code isn't just old; it's opaque. Enterprises spend billions annually maintaining these systems because no one fully understands them anymore. Developers waste weeks manually diagramming flows, data structures, and logic. Application Transform flips the script:
- Input: Upload your source code (COBOL, PL/I, or similar).
- Magic: AI analyzes it deeply, cross-referencing existing scraps of design docs or comments.
- Output: Structured design documents covering system architecture, modules, data flows, and behaviors—ready for modernization planning.
It's not some generic ChatGPT hack. Fujitsu built this on decades of system dev expertise, launching as SaaS in Japan starting today. Future updates in FY2026 will add code rewriting and ops/maintenance aids.[1]
Think of it as GitHub Copilot meets enterprise archaeology. But way smarter for big iron. See our guide on AI code tools for more on how this stacks up.
The Tech Under the Hood: Why It's Smarter Than Generic AI
What sets Application Transform apart? It's not throwing LLMs at code and hoping for the best. Fujitsu layers in proprietary sauce:
-
Fujitsu Knowledge Graph–Enhanced RAG for Software Engineering: This is the star. RAG (Retrieval-Augmented Generation) pulls relevant code chunks to ground the AI, but Fujitsu's graph version links massive codebases like a neural network. No hallucinations (AI making stuff up) or omissions—critical for COBOL's procedural mazes.[1]
-
Proprietary Code Analysis: Builds on a 2025 predecessor tool for visualization. It parses "remaining design information" (old specs) alongside live code, standardizing analysis that humans do inconsistently.
-
Kozuchi Backbone: Fujitsu's AI platform handles everything from GenAI to trust layers, ensuring enterprise-grade security and scalability.
Here's a simplified flow:
Upload Legacy Code (e.g., COBOL)
↓
Knowledge Graph Builds Relations (modules, calls, data)
↓
GenAI Generates Docs (diagrams, descriptions, flows)
↓
Human Review (minimal, thanks to 95% comprehensiveness)
Compared to vanilla GenAI:
- Generic AI: Misses interconnections in large repos → omissions galore.
- This: 95% more comprehensive.[1]
Result? Docs that are 60% more readable than manual methods—cleaner language, better visuals, logical structure.[1]
Killer Stats: 97% Faster, Proven in the Wild
Fujitsu didn't pull numbers from thin air. They verified with SMBC Nikko Securities Inc. (a major Japanese financial firm) starting FY2025. The test? Reverse-engineering COBOL-heavy legacy docs.
Key Results:
- Time Reduction: ~97% faster to generate design docs. What took experts weeks? Now days—or hours.[1]
- Comprehensiveness: 95% improvement over plain GenAI. Catches nuances like edge cases in financial logic.
- Readability: 60% better than conventional manual docs. Easier for juniors to grok, speeding team ramps.[1]
Toshihiro Horiuchi, Managing Executive Officer at SMBC Nikko:
"We have been conducting joint verification with Fujitsu on reverse-engineering design documentation for legacy languages, including COBOL, and through this collaboration we have come to recognize the significant potential of this technology."[1]
In enterprise terms: If a bank spends 6 months ($500K+) understanding a system before migrating to cloud, this slashes it to a week. Productivity boost? Massive.
| Metric | Manual/Generic AI | Fujitsu App Transform |
|---|---|---|
| Time | Baseline (weeks/months) | 97% reduction[1] |
| Comprehensiveness | Baseline | +95%[1] |
| Readability | Baseline | +60%[1] |
Real-World Impact: Game-Changer for Dev Productivity
For devs and IT leads, this is liberation. No more:
- Expert Bottlenecks: COBOL coders are retiring en masse. Now, any dev with upload access can kick it off.
- Manual Drudgery: Hours tracing
PERFORMstatements? AI does it. - Migration Risks: Bad docs lead to buggy rewrites. These are accurate baselines.
Use Cases:
- Finance: Like SMBC Nikko—audit-proof docs for compliance-heavy COBOL cores.
- Insurance/Gov: Modernize mainframes to microservices.
- Any Legacy Hauler: PL/I, Assembler—whatever's gathering dust.
Pair it with tools like AWS Mainframe Modernization or Azure Logic Apps for full pipelines. Check our legacy migration guide for strategies.
Productivity math: A 10-dev team saves thousands of hours/year. At $100/hr, that's millions unlocked for innovation.
Availability, Pricing, and Getting Started
Launched today (March 30, 2026) as SaaS in Japan via Fujitsu Kozuchi. Global rollout? Likely soon—watch Fujitsu's site.
- Pricing: Not public yet (typical for enterprise SaaS). Expect usage-based: per codebase size or docs generated. Contact Fujitsu for pilots.
- Onboarding: Upload code, AI processes in cloud. Support services coming for integration.
- FY2026 Roadmap: Auto-rewrite code, rebuild for new langs, O&M aids.[1]
Early access? Hit up Fujitsu sales—mention the SMBC verification for cred.
FAQ
What languages does Fujitsu Application Transform support?
Primarily COBOL and other legacy source codes in enterprise systems. It shines on procedural langs like PL/I. Future updates may expand.[1]
Is it secure for sensitive enterprise code?
Yes—runs on Fujitsu Kozuchi with enterprise controls. RAG keeps processing contained; no code leaves your vault without consent. Ideal for regulated industries.
How accurate are the generated docs? Any hallucinations?
95% more comprehensive than generic AI, 60% more readable. Knowledge Graph prevents hallucinations by grounding in full codebase links. Verified by SMBC Nikko.[1]
When can I try Fujitsu Application Transform outside Japan?
SaaS starts in Japan today. Global? Fujitsu's pattern—expect APAC/EU/US soon. Pilot programs via sales.
So, what's your legacy nightmare? COBOL mainframe or forgotten Fortran? Drop a comment—have you tried tools like this, or is Fujitsu's 97% speedup tempting you to pilot? Let's chat modernization wins! [1]