Legacy
1959
The year COBOL was born. 220 billion lines of it still run in production.
Modern

Sixty yearsof code.Finally translated.

2028
The year we ship the AI that finally rewrites it.
01
The Premise

The world runs on yesterday.

Half a century ago, engineers wrote the systems that quietly run the modern world — payrolls, settlements, air-traffic, utility grids. Their authors have retired. Their languages have been declared dead a dozen times. The code persists.

Modernization has remained slow, brittle, and outrageously expensive. We are building the AI that finally makes it tractable.

0B
Lines of COBOL still running in production
More than every line of Python, Java, and JavaScript ever shipped — combined.
fig.01
$0T
In commerce processed every day through COBOL systems
Behind every wire, swipe, and settlement — code older than the internet.
fig.02
0 / 100
Of the world's largest banks still run on mainframes
The systems quietly underpinning the modern economy.
fig.03
0%
Of IT leaders cite legacy as the #1 barrier to modernization
It isn't optional anymore. It's overdue.
fig.04

“The hardest software problem of the next decade isn't building new systems. It's translating the ones we can no longer afford to rewrite by hand.”

— From the founding memo
02
The Product

An AI that reads, reasons, and rebuilds.

Built from the ground up for the dialects, idioms, and architectural assumptions of 1970–2005. Not a generic copilot retrofit.

IRead

Understand intent, not just syntax.

A semantic model of the codebase: control flow, data shape, hidden invariants, and the business intent encoded over forty years of patches.

  • Cross-file dependency and call graphs
  • Pattern, idiom and anti-pattern detection
  • Dead-code, dark-data and ghost-branch surfacing
  • Cyclomatic complexity, coupling, drift
IIReason

Map decades of intent into modern architecture.

Choose your target stack. Our planner reconciles the source intent with idiomatic patterns in the destination — typed, modular, observable.

  • Target-aware refactoring strategies
  • Domain decomposition and module boundaries
  • Schema and storage migration plans
  • Risk-ranked execution roadmap
IIIRebuild

Generate code with provable equivalence.

Production-grade output verified against the source through symbolic execution, generated property tests, and side-by-side replay on real workloads.

  • Idiomatic Java, Kotlin, Scala, C#, F#, TypeScript, Python, Go, Rust, C++
  • Auto-generated test parity suites
  • Differential replay against legacy traffic
  • Auditable transformation map per commit
03
In Practice

From this. Into this.

Real output. A 1987 COBOL deposit routine, modernized to idiomatic Java with the original business logic preserved line-for-line in the transformation map.

ACCT-DEPOSIT.CBL
COBOL · 1987
01 * ACCT-DEPOSIT — applies a deposit to a customer balance
02 IDENTIFICATION DIVISION.
03 PROGRAM-ID. ACCT-DEPOSIT.
04 
05 DATA DIVISION.
06 WORKING-STORAGE SECTION.
07 01 WS-AMOUNT PIC S9(9)V99.
08 01 WS-BALANCE PIC S9(11)V99.
09 01 WS-FEE PIC S9(5)V99 VALUE 2.50.
10 
11 PROCEDURE DIVISION.
12 PROCESS-DEPOSIT.
13 IF WS-AMOUNT > 0
14 COMPUTE WS-BALANCE = WS-BALANCE + WS-AMOUNT - WS-FEE
15 PERFORM LOG-TRANSACTION
16 ELSE
17 DISPLAY 'INVALID AMOUNT'
18 END-IF.
○ source · indexed18 ln · ↑ ingested
AccountDeposit.java
Java 21 · 2026
01// Generated by Legacy Modernization AI · semantic equivalence verified
02public final class AccountDeposit {
03 private static final BigDecimal FEE = new BigDecimal("2.50");
04 
05 public Result deposit(Account account, BigDecimal amount) {
06 if (amount.signum() <= 0) {
07 return Result.invalid("amount must be positive");
08 }
09 
10 var balance = account.balance()
11 .add(amount)
12 .subtract(FEE);
13 
14 ledger.record(account, amount, FEE);
15 return Result.ok(balance);
16 }
17}
● equivalence verified17 ln · ↻ regenerated
Transformation Map · excerpt
PIC S9(11)V99
Replaced with BigDecimal — preserves exact decimal arithmetic, eliminates COMP-3 packing risk.
PERFORM LOG-TRANSACTION
Inlined as a method call to ledger.record(), traced to existing audit hooks.
IF WS-AMOUNT > 0 / ELSE DISPLAY
Promoted to early-return guard — same branch, modern control flow, no GO TO.
04
The Process

Three movements, one outcome.

  1. 01/ Sweep

    Feed it the codebase. Get the full map.

    A complete semantic atlas of your system in hours, not months. Modules, dataflows, dependencies, dead branches, security debt — all surfaced and ranked.

    artifact · 01
    deep-sweep.report.json
    Cyclomatic82
    Coupling61
    Dead code34
    Dark data49
  2. 02/ Translate

    Choose your stack. Watch the transformation.

    Pick a target — Java, Python, Go, Rust, TypeScript. The model rebuilds your system idiomatically, with module boundaries that match your domain, not the legacy file structure.

    artifact · 02
    modernize.plan.diff
    -PERFORM CALC-INTEREST UNTIL EOF.
    -MOVE WS-RATE TO WS-OUT-RATE.
    +interest = computeInterest(account);
    +audit.record(account, interest);
    ledger.commit(transactionId);
  3. 03/ Verify

    Every transformation comes with proof.

    Generated property tests, symbolic equivalence checks, and shadow-replay against production traffic. You get a transformation map, not a black box.

    artifact · 03
    equivalence.proof.log
    Symbolic equivalence on 1,247 paths
    Shadow replay · 14 days production traffic
    Generated property tests · 312 / 312 passing
    Numeric drift · ±0.000 across all decimals
05
The Coverage

Sixty years of dialects. One model.

Trained on every major mainframe and enterprise dialect from the punch-card era through the early web. Targeting the modern stacks teams actually want to ship on.

est. 195901
COBOL
Java, Kotlin, Go
Banking · Insurance · Government
est. 195702
FORTRAN
Python, Rust, C++
Scientific · Defense · HPC
est. 199803
VB6
C#, TypeScript
Enterprise apps · Internal tools
est. 197004
Pascal
Go, TypeScript
Embedded · Education · Tooling
est. 196405
PL/I
Java, Scala
Mainframe transaction processing
est. 198006
Ada
Rust, C++
Aviation · Defense · Real-time
est. 195907
RPG
Java, Kotlin
IBM i · ERP · Manufacturing
est. 199508
Delphi
C#, F#
Desktop · POS · Industrial
est. 199109
PowerBuilder
TypeScript, C#
Client-server · Forms-over-data
est. 196610
MUMPS
Python, Java
Healthcare · Epic systems · Banking ledgers
+ ALGOL · JCL · CLIST · Forth · Modula-2 · Smalltalk on the roadmap
06
The Waitlist

Bring us your hardest system.

We're in the early phase of building what may be the most demanding software product of the decade. Public launch is targeted for Q1 2028. Until then, we're talking to enterprises with mission-critical legacy stacks — to learn what to build first.

For organizations with active modernization mandates
Conversations under NDA, on your terms
Help shape what we ship in v1
First in line when access opens
Form · 01● The Waitlist is currently closed

We'll only reach out as we get closer to launch — likely with questions, not pitches. No drip sequences, no growth hacks.