ChoyTein Shen

I made Microsoft Dynamics 365 Finance and Operations application development AI-native. Work that was locked inside Visual Studio now ships in hours, not days.

8+ YRS ENTERPRISE SYSTEMSX++ / .NET

ZERO METADATA CORRUPTION

HOURS, NOT DAYS

THROUGHPUT PER DEV
4 → 17 TASKS / WEEK

READ THE STORY ↓

The wall

Everyone assumed AI coding agents couldn't do D365 F&O development. They had three good reasons.

LOCK 01

Bound to Visual Studio

Build, deploy, DB sync, debugging, element design — every operation lives inside the IDE. No IDE, no development.

LOCK 02

Fragile XML metadata

Tables, forms and elements aren't plain code — they're strict XML. A naive AI edit corrupts project, model and element files.

LOCK 03

No intelligence outside the IDE

No go-to-definition, no find-references, no symbol knowledge for X++ exists anywhere outside Visual Studio.

So AI stayed locked out of X++ —

The breakthrough

I designed the enablement layer that lets an AI agent run the entire Microsoft Dynamics 365 Finance and Operations application development lifecycle on its own — with zero metadata corruption. The agent doesn't stop at writing code: it builds, deploys, syncs the database, tests, and iterates until the requirement is met.

DEVELOP→BUILD→DEPLOY→TEST→SHIP

Every change passes the same review and testing gates as hand-written work — the loop raises throughput, not risk.

AI AGENT · LIVE TASK FEED

Nothing left behind in the IDE

Not a demo on one task type — systematic coverage of the work a D365 developer actually does.

01Code & logic

·New feature development & code-logic changes
·X++ patterns — batch jobs, service operations — correct by construction
·Event-handler discovery + generation
·Debugging & defect fixing
·Formatting to standard

02Data & UI design

·Table design — fields, groups, indexes, relations, delete actions
◆Form patterns applied with the control skeleton auto-populated
·Any element property changed, valid values known up front
·Extensions of standard tables, forms, menus, queries
·Data entity work

03Ship & operate

·Build — models, projects, labels
·Deploy to cloud environments
·Database synchronization
·Model management — creation, references, versioning
◆Feature exposure — menu items, security privileges, entry points

04Navigate & investigate

·Find references, implementations, definitions
·Metadata search across the entire application
·Impact analysis before a change
·Signatures, members, extension lookup

If a developer does it in Visual Studio, the AI does it here — end to end.

What it did for the team

SPEED

Change requests estimated at 3 days of development, delivered in 3 hours.

SCALE

+0%

Throughput per developer — 4 to 17 tasks a week — rolled out across the whole team, not one desk.

QUALITY

Corruption errors across project, model and element files — the failure mode that kept AI out of X++.

The point was never my own output. I built the enablement, rolled it out to the developer team, and mentored them into the new workflow — the productivity gain belongs to everyone.