From Copilot to Autonomous Agent

Most AI coding tools wait for you to type. Stage 2 doesn't wait. It picks up requirements, assigns them to a coding agent, and manages the entire development lifecycle — analysis, implementation, build, test, fix — while you focus on higher-value work.

When it needs human input, it asks. When it doesn't, it ships.

Stage 2 is an agent harness platform: the infrastructure layer that turns AI skills into a reliable, unattended production system.

Orchestration

How It Works

A deterministic pipeline from requirement to done

Requirement

→

Claim

→

Analyze

→

Code

→

Build

→

Test

→

Fix Loop (HITL)

→

Upload

→

Deliver

→

Done

Human feedback when needed

Automatic task pickup. Stage 2 polls for pending requirements and claims them — no manual assignment needed.

Full lifecycle orchestration. Each requirement flows through analysis, coding, compilation, testing, and fix iterations — all managed by the platform.

Built-in feedback loop. When a task needs human review or hits a blocker, Stage 2 routes it back with full context. Human feedback triggers a new iteration automatically.

Lightweight task monitoring. Platform-level liveness and progress tracking that watches the agent, detects phase transitions, and surfaces status without interrupting the coding workflow.

Self-healing. If the coding agent crashes or stalls, Stage 2 detects it, saves partial work, and reports the failure — no silent hangs, no lost progress.

Continuous delivery integration. Completed artifacts are pushed to the customer's source repository and merged into their existing Azure DevOps pipeline — no manual handoff between the agent and the release process.

Complete audit trail. Every phase is logged. Every iteration is archived with full context — requirement, code, build output, test results, agent transcript.

Two Layers

AutoXPP + Stage 2

Two layers, one system

	AutoXPP	Stage 2
Role	The AI developer	The factory floor
What it does	Analyzes requirements, writes X++ code, triggers builds, runs tests, fixes errors	Claims work, provisions environments, monitors health, manages iterations, uploads results
Who runs it	Any developer, interactively in Claude Code	Runs unattended on dedicated agent machines
Feedback	Takes direction from the developer in real time	Routes structured feedback between iterations automatically

Difference

Not Just Code Generation

A complete development system

Requirement to Deployment

Stage 2 doesn't generate code fragments for you to assemble. It produces compiled, tested, deployable D365 customizations from a plain-English requirement.

Feedback Loops, Not One-Shot

Every failed build, every test failure, every human review triggers a new iteration with full context. The system converges on working code through structured retry.

Observable, Not Opaque

Every phase is logged. Every iteration is archived with full context — requirement, code, build output, test results, agent transcript. You can audit exactly what the system did and why.

Harness-Level Reliability

Liveness monitoring, crash recovery, and stall detection are platform concerns — handled by the infrastructure, not by the AI model. The agent focuses on coding; the harness keeps it alive.

Architecture

Architecture at a Glance

Three layers working together

XGen Backend

API

Requirement intake, status queries

Task Queue

Work items pending assignment

Feedback Store

Human reviews, iteration history

↓

Stage 2 Harness (per machine)

Orchestrator

Poll, claim, spawn, kill

Task Monitor

Liveness, progress tracking

Watchdog

Crash recovery, stall detection

↓

Coding Agent

AutoXPP Skills

X++ generation, pattern library

Build

VS compile, DB sync

Test

Validation, evidence capture

Get Started

Ready to Scale Your D365 Development?

Start with AutoXPP today. Stage 2 scales it to a fleet.

Get AutoXPP Contact Us About Stage 2

Questions? contact@xplusplus.ai

Autonomous D365 Development, End to End