Research Overview

ModernizeSpec is grounded in a 9-document research series produced during PearlThoughts’ ERPNext modernization effort in February 2026. The research combined deep web search, direct codebase analysis, experiment review, intern team output evaluation, GitHub organization analysis, Microsoft Teams channel analysis, and industry benchmarking.

Executive Summary

A full ERPNext migration (all 521 doctypes, 768 API endpoints, 177 reports, full Frappe framework replacement) in 3 months is not achievable — even with unlimited AI credits and experienced engineers. The codebase represents ~216 person-years of effort (COCOMO estimate).

However, a strategically scoped migration of the core accounting engine (the highest-value 15-20% of the codebase) is achievable in 3 months with 3-5 skilled interns using AI tooling.

Scenario	Scope	Team	Timeline	Feasibility
Full ERPNext replacement	521 doctypes, all modules	3-5 interns	3 months	Not feasible
Core Accounting engine	~60 doctypes, GL/AP/AR/Tax	3-5 interns	3 months	Achievable with AI
Core + Stock + Selling/Buying	~150 doctypes	3-5 interns	6-9 months	Achievable with AI
Production-ready ERP (80% parity)	~400 doctypes + framework	8-12 engineers	12-18 months	Achievable with AI

Research Methodology

Deep web search — ERPNext history, market position, AI migration landscape
Codebase analysis — Direct measurement of ERPNext Python LOC, doctypes, API endpoints, test coverage
Experiment review — 3 completed Go extraction iterations, 68 tests
Intern team output review — All 7 teams: GitHub commits, PR activity, parity test results
GitHub organization analysis — 20 repos, commit activity over 3 weeks
Teams channel analysis — 1,168 messages across 11 channels, synced February 8
HR internship records — Team structure, evaluations, pain point analysis
Industry benchmarking — Airbnb (3,500 files in 6 weeks), Amazon (30K apps), Microsoft CAMF

Research Documents

Codebase Complexity

316,679 Python LOC, 521 doctypes, controller inheritance chains, complexity tiers, and what makes ERPNext hard to migrate.

Read the assessment

AI Migration Landscape

Enterprise platforms (Amazon Q Transform, IBM watsonx), success stories (Airbnb 12x compression), and the 8 key patterns from industry research.

Read the landscape

Feasibility Forecast

4 scenarios analyzed, Go vs Kotlin comparison, intern velocity data, cost projections ($16.8K-$38.3K for 3 months).

Read the forecast

Standard Feasibility

Market gap analysis: no existing standard combines legacy measurement + DDD mapping + extraction sequencing + parity verification. 100% generalizable.

Read the analysis

Key Findings Across All Research

What AI Can Compress

Phase	Compression	AI Impact
Understanding legacy code	5-10x with graph-RAG + code intelligence	High
Mechanical translation	10-50x for typed code	High
Test generation	3-5x for table-driven tests	Medium-High
Architecture decisions	1x (no compression)	None — needs human judgment
Framework design	1x (no compression)	None — needs senior engineering

Eight Key Patterns from Industry

Multi-agent pipelines (analyze, translate, validate, iterate) are the dominant architecture
Retry with error feedback outperforms sophisticated prompt engineering
Rich context is the primary success driver — choosing the right related files matters more than prompt engineering
Specialized models outperform general-purpose LLMs for targeted language pairs
Strangler Fig + AI is the lowest-risk combination
Shadow testing and parity validation are non-negotiable before production cutover
Team sizes shrink but skill requirements shift — 5-10x fewer people, but they need architectural judgment
Vendor claims require verification on your own codebase

The Market Gap

No standard combines:

Legacy codebase measurement in machine-readable format
DDD bounded context mapping consumable by AI agents
Extraction sequencing with risk scoring and priority
Parity verification standards
Multi-agent coordination for modernization

This gap is what ModernizeSpec fills.