✓ Survived Rank 1 of 17 proposedVenture DevTools Oct 21, 2026

Verified AI Porting Platform for Legacy Codebases

Frontier-model code ports (COBOL, Java, TS, Python) shipped with machine-checked behavioral equivalence proofs instead of vibes.

Open the discussion ↗

A ready-to-paste brief for Claude Code, Cursor or any coding agent.

I want to build a venture-scale product. Help me plan and start the MVP.

Product: Verified AI Porting Platform for Legacy Codebases
What it is: Frontier-model code ports (COBOL, Java, TS, Python) shipped with machine-checked behavioral equivalence proofs instead of vibes.
Who pays: VP Engineering / Head of Platform Modernization at enterprises with large legacy codebases (banks, insurers, SaaS with 1M+ LOC); $150k–$1M migration budgets that today go to consultancies.
Closest alternatives: Mechanical Orchard (COBOL modernization, services-heavy), Moderne / OpenRewrite (rule-based automated refactoring, not model-driven ports), Sourcegraph Batch Changes (large-scale edits, no equivalence verification), In-house agent + senior-reviewer time
The gap: Existing players either do deterministic rule-based refactors or sell human consulting hours; none productize automated differential/conformance evidence that a model-generated port is behaviorally equivalent to the original, which is exactly the trust gap egonschiele describes.
Why now: In the last 12 months frontier models crossed the threshold where full compiler-scale ports complete in days (this article is the proof case), so the bottleneck in every migration has shifted from writing code to trusting it.
Moat to build toward: Compounding corpus of language-pair conformance test suites, semantic-diff techniques, and equivalence-failure data from every engagement — each port makes the next one faster and safer; switching means losing the accumulated certification history.
Biggest risk: Behavioral equivalence testing turns out to be as hard as the port itself for stateful, side-effecting enterprise systems, and buyers default to established consultancies regardless of cost.

First three steps:
1. Pick one language pair (e.g., Java 8 monoliths to modern Java/Kotlin or Python 2-era services to Python 3) and build a differential test harness that auto-generates behavioral equivalence tests from the source system's traces and types
2. Run 2–3 paid pilot migrations with a human-in-the-loop agent pipeline, publishing an 'equivalence report' as the deliverable
3. Productize the report + regression harness as a standalone verification offering that plugs into any team's own agent workflow

Start by asking me up to five questions about constraints (stack, budget, timeline), then propose the smallest version a paying customer would use, and scaffold it.

Overall6.5/10

Pain7

Willingness to pay7

Timing7

Gap6

Moat6

Feasibility5

For

Who pays

VP Engineering / Head of Platform Modernization at enterprises with large legacy codebases (banks, insurers, SaaS with 1M+ LOC); $150k–$1M migration budgets that today go to consultancies.

What people said · 5

Frontier models can now do large-scale language ports in days, and the ecosystem already has 'slopforks', faithful ports, and re-architected variants of TypeScript — fidelity is the differentiator. From the article ↗
Heavy agent user still manually reviews AI output because coding agents make unsound architectural decisions, and notes large codebases outgrow existing tooling. Read the comment ↗
Three independent AI-assisted TypeScript ports already exist (tsrs, tsgo-rs, bun check), showing porting demand and fragmentation. Read the comment ↗
Commenter wonders whether the same AI-port approach has been tried for CPython, indicating appetite beyond TypeScript. Read the comment ↗
A solo developer ported TypeScript to C# with AI for about $75, showing porting cost has collapsed but with no correctness guarantee attached. Read the comment ↗

Paraphrased from the discussion. Follow the links for the original comments.

Competitors and gap

Mechanical Orchard (COBOL modernization, services-heavy)Moderne / OpenRewrite (rule-based automated refactoring, not model-driven ports)Sourcegraph Batch Changes (large-scale edits, no equivalence verification)In-house agent + senior-reviewer time

Existing players either do deterministic rule-based refactors or sell human consulting hours; none productize automated differential/conformance evidence that a model-generated port is behaviorally equivalent to the original, which is exactly the trust gap egonschiele describes.

Why now

In the last 12 months frontier models crossed the threshold where full compiler-scale ports complete in days (this article is the proof case), so the bottleneck in every migration has shifted from writing code to trusting it.

Moat · 6/10

Compounding corpus of language-pair conformance test suites, semantic-diff techniques, and equivalence-failure data from every engagement — each port makes the next one faster and safer; switching means losing the accumulated certification history.

Against

Red team: Frontier models collapsing porting cost this year genuinely shifts the bottleneck to trust, enterprises already spend $150k–$1M per migration, and no consultancy productizes machine-checked behavioral equivalence for model-generated ports outside Mechanical Orchard's COBOL niche.
What would kill it: Behavioral equivalence testing turns out to be as hard as the port itself for stateful, side-effecting enterprise systems, and buyers default to established consultancies regardless of cost.

Start

First three steps: Pick one language pair (e.g., Java 8 monoliths to modern Java/Kotlin or Python 2-era services to Python 3) and build a differential test harness that auto-generates behavioral equivalence tests from the source system's traces and types
Run 2–3 paid pilot migrations with a human-in-the-loop agent pipeline, publishing an 'equivalence report' as the deliverable
Productize the report + regression harness as a standalone verification offering that plugs into any team's own agent workflow
Source discussion: Tsrs: A Port of TypeScript to Rust ↗ 426 points209 commentsFrom the storyLinked article

For

Against

Start

More from this issue