Legacy code to modern systems — with verified equivalence
Scriba AI converts COBOL, RPG, PL/I, Natural and 60+ other languages into modern, idiomatic applications — without LLMs. Migration as a service: you send us the code, we return it converted, tested and documented.
Scriba.AI on stage at Codemotion 2026
We are pleased to announce that on October 29, as part of Codemotion 2026 in Milan, two Scriba.AI founders will present the platform in detail, its advantages over generic LLMs, and how to use it.
A chance to discover the only automatic code-to-code conversion solution: fully Italian, proprietary technology.
- The Scriba.AI platform in detail
- Advantages over generic LLMs
- How to use it, with real migration cases
SCRIBA.AI: an Italian AI solution to modernize legacy code
Marco Landi
Senior Advisor
Michele Laurelli
Head of AI
Migrating a legacy system is hard for one specific reason
The system's behaviour is documented only in the code, or in the minds of the people who built it — and fewer and fewer people know it.
Rules layered over decades
Business logic built up over time, rarely with up-to-date documentation.
Hidden edge cases
Exceptions handled by a single conditional branch, easy to lose in a rewrite.
Data semantics
Rounding and truncation that depend on the physical representation of the data.
Logic in tables
Parameter tables drive behaviour: part of the rules doesn't live in the code at all.
A successful migration has to preserve all of this — and be able to prove it.
What makes
Scriba AI different
Not a line-by-line translator: Scriba reconstructs the application logic and business rules, then rewrites them in the target stack, following the conventions you set.
No LLMs
Proprietary specialized models and deterministic components, coordinated by fully proprietary orchestration. No language model — neither commercial nor open source.
Verified equivalence
Differential tests compare the migrated system's outputs with the original's on the same inputs. Anything that doesn't pass the deterministic gates isn't delivered.
Idiomatic code
No JOBOL: classes, layers, types, error handling and naming follow your rules — written the way a team that works with that stack every day would write it.
Confidentiality
NDA and DPA, code never sent to model providers, air-gapped environment when needed. Delivery via encrypted storage or a private Git repository.
A six-phase pipeline
A system of specialized models and deterministic components, each dedicated to a well-defined task, coordinated by a proprietary orchestration layer.
Parsing
Syntax tree built with parsers specific to each language and dialect.
Semantic analysis
Types, dependencies, control flow and side effects.
Logic recovery
Business rules made explicit, traced and separated from the legacy runtime.
Multi-model
Specialized models in an ensemble: disagreement triggers an escalation.
Validation
Compilation, types, client rules, differential tests on behaviour.
Generation
Idiomatic target code, project structure, build and evidence.
Deterministic verification gates: no phase moves forward unless the previous one has been verified.
Why Scriba AI and not a commercial AI service?
Sooner or later, every approach produces code that compiles. The difference is what you can prove about the result.
Understanding of business logic
- Generic LLMs
- Limited to the fragment
- Agents on LLMs
- Limited to the agent's context
Scriba AIReconstructed across the whole scope
Quality of the code produced
- Generic LLMs
- Variable
- Agents on LLMs
- Variable
Scriba AIIdiomatic, following your rules
Equivalence verification
- Generic LLMs
- None
- Agents on LLMs
- To be built
Scriba AIDifferential tests against the original
Code confidentiality
- Generic LLMs
- Code sent to third parties
- Agents on LLMs
- Code sent to third parties, or infrastructure to run
Scriba AIGoverned by NDA and DPA, nothing sent to model providers
Cost predictability
- Generic LLMs
- Low on large scopes
- Agents on LLMs
- Low: token consumption or infrastructure
Scriba AIFixed, guaranteed price
Accountability for the result
- Generic LLMs
- Yours
- Agents on LLMs
- Yours
Scriba AIScriba's, until every gate is passed
| Criterion | Generic LLMs | Agents on LLMs | Scriba AI |
|---|---|---|---|
| Understanding of business logic | Limited to the fragment | Limited to the agent's context | Reconstructed across the whole scope |
| Quality of the code produced | Variable | Variable | Idiomatic, following your rules |
| Equivalence verification | None | To be built | Differential tests against the original |
| Code confidentiality | Code sent to third parties | Code sent to third parties, or infrastructure to run | Governed by NDA and DPA, nothing sent to model providers |
| Cost predictability | Low on large scopes | Low: token consumption or infrastructure | Fixed, guaranteed price |
| Accountability for the result | Yours | Yours | Scriba's, until every gate is passed |
Ratings describe the typical behaviour of each approach; individual tools or projects may differ.
How do you prove that the migrated system behaves like the original?
For Scriba, a migration is complete when the code passes every test and quality gate set by the client, including differential tests that compare its outputs with the original system's on the same inputs.
COBOL ILE batch on IBM i for a leasing company
Extraction of financial transactions, migrated according to the client's backend and frontend instruction files.
COBOL ILE batch on IBM i
source members mapped to Java classes
documents delivered, from the executive summary to the operations runbook
validation methodology for equivalence with the original system
rounds of review by independent reviewers, with every resolution traced
Evaluate Scriba on your own code
The PoC applies the same working model as a full engagement to a smaller scope: you send us a closed scope of code, we return it migrated, tested and documented, with the evidence that every criterion has been met.
Up to 30,000 lines
A closed scope: a complete process with its dependencies, not a sample of files.
Criteria decided in advance
Tests and quality gates declared before kick-off, including differential tests against the original.
A single pass
Complete configuration up front, frozen at kick-off: no drafts, no mid-course corrections.
Seven phases, from qualification to delivery
- 01
Qualification
NDA and candidate scope
- 02
Package
Code, instructions, tests, data
- 03
Eligibility
≤ 30,000 lines, closed scope
- 04
Kick-off
Frozen criteria, PoC agreement
- 05
Conversion
Single pass, verification gates
- 06
Validation
Every test and quality gate
- 07
Delivery
Deliverables and a verifiable outcome
Estimate your migration scenario
Configure your scenario and get an indicative estimate. The actual price is fixed and is set after we analyse your package.
60+ languages, from mainframe to cloud
From COBOL, RPG, PL/I and Natural to modern stacks, with the target you define: language, architecture and conventions.
Ready to evaluate Scriba on your code?
Tell us about the scope you have in mind: together we'll check eligibility, target stack and success criteria.