Process

From Idea to
Production Agent

Agent Accelerator takes you from a business need to a production-deployed, compliance-validated, governance-documented AI agent. Here's the complete journey — from your first sentence to a live agent in Teams.

6

Clear Phases

15 min

Design Time

73

Compliance Checks

1

Command Deploy

01
Phase 1 15-30 minutes

Design

Start a Build Studio session. Tell the AI Architect what your agent should do.

"We need an agent in Microsoft Teams that helps employees submit IT support tickets. It should check our ServiceNow knowledge base first, and if there's no answer, create a ticket. Users shouldn't have to type their details — the agent should know who they are."

Example user requirement

Build Studio walks you through 7 structured steps

1

Initialize

Name the agent, set up the project, capture context.

2

Define Identity

The architect recommends a persona, expertise areas, and communication style. You refine. It configures Entra ID for automatic user identification.

3

Recommend Tools

Build Studio researches official Microsoft documentation, platform capabilities, and enterprise patterns. It proposes MCP Gateway tools, knowledge base search with confidence thresholds, and auto-populated user data via Entra ID. You review, adjust, approve.

4

Design Workflows

The architect recommends Generative Orchestration with a Cascading Resolution Pipeline: Intent Check, Knowledge Base Search, Service Catalog Search, Ticket Creation, and Graceful Fallback.

5

Identify Dependencies

Build Studio classifies each integration as platform-native (Entra ID, SharePoint KB, Adaptive Cards) or engineering dependency (ServiceNow MCP plugin).

6

Confirm Platforms

Primary: Copilot Studio (Teams). Secondary: none for now — can regenerate for Bedrock or Vertex later.

7

Review & Approve

You see the complete agent design. Everything looks right. Approve.

Output

A platform-agnostic it-support.dsl.yaml file plus requirements.md, design-decisions.md, and research-notes.md.

02
Phase 2 Instant

Generate

You select "Generate for Copilot Studio." Build Studio produces a complete Power Platform solution.

Generated file structure

ITSupportAgent_1_0_0_0/
  solution.xml
  customizations.xml
  bots/ITSupportBot/
    bot.xml
    configuration.json
  botcomponents/
    Topics/         (Greeting, Fallback, custom topics)
    GPT/            (System instructions with tool orchestration)
    Actions/        (AI Actions for each ServiceNow tool)
  Connector/
    MCPGateway/     (Custom connector with OpenAPI spec)
  Workflows/        (Power Automate flows if needed)

A real, importable solution

This is not a scaffold or template. Every component is configured and wired together — topics, GPT instructions, AI actions, connectors, and workflows. Import it directly into your Copilot Studio environment.

03
Phase 3 Automatic

Validate

Build Studio automatically scans the generated package against 73 compliance checkpoints.

Compliance Report
Compliance Report — IT Support Agent
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Agent-Level Security     [11/11 PASS]
   Authentication: Entra ID configured
   Content moderation: Medium
   No secrets in GPT instructions
   No PII collection instructions

Environment-Level        [12/12 PASS]
Flow-Level               [14/15 PASS]
   WARN: Consider adding error handling to catalog flow

Policy-Level             [10/10 PASS]
Topic/Code Security      [6/6 PASS]
Risk Classification      [3/3 PASS]

Overall: 72/73 PASS | 1 WARN | 0 FAIL
Frameworks: GDPR  | SOC 2  | HIPAA  | ISO 27001 

Auto-Generated Governance Plan

Business owner and review cadence

Data classification (Internal)

DLP policies and connector restrictions

Sharing model (IT dept, then all employees)

Usage analytics and feedback collection

Incident response procedures

User and admin documentation requirements

04
Phase 4 Auto-Generated

Engineering Handoff

For dependencies that need development, Build Studio produces engineering specifications — not vague requirements documents. Your team knows exactly what to build, in what order, and how to test it.

P1 Critical Path

MCP Gateway Plugin

  • 8 tools: get_user_tickets, create_incident, search_knowledge, search_catalog...
  • OAuth 2.0 Client Credentials to ServiceNow
  • Each tool fully specified: params, outputs, error handling, retry logic
  • Estimated effort: 2-3 sprints
P2 Infrastructure

Infrastructure Setup

  • ServiceNow OAuth application setup (scopes, roles)
  • MCP Gateway deployment and agent registration
  • Secrets management (Key Vault)
P3 Integration

Integration & Testing

  • Update Copilot Studio connector with real gateway URL
  • End-to-end testing plan
  • Deployment checklist
05
Phase 5 1 Command

Deploy

One command deploys the agent. It appears in your Copilot Studio environment, ready for connection configuration and channel publishing.

Copilot Studio deployment

$ pac solution import --path ./ITSupportAgent_1_0_0_0.zip --publish-changes

MCP Gateway agent registration

{
  "AgentId": "it-support",
  "DisplayName": "IT Support Agent",
  "AllowedToolNamespaces": ["servicenow", "datetime"],
  "RateLimit": { "RequestsPerMinute": 200 }
}

Copilot Studio Import

The agent appears in your environment ready for configuration of connections (ServiceNow credentials, MCP Gateway endpoint) and channel publishing (Teams, web chat).

MCP Gateway Registration

Your engineering team deploys the MCP Gateway (if not already running) and registers the new agent with its tool namespaces and rate limits.

06
Phase 6 Ongoing

Iterate

When requirements change, open the existing DSL in Build Studio. The original requirements and design decisions are right there. Add the new capability, regenerate, re-validate, and deploy.

Modification history in DSL

modification_history:
  - date: "2026-04-15"
    change: "Added password reset capability via MCP Gateway"
  - date: "2025-12-04"
    change: "Initial design and generation"

When Requirements Change

Six months later, the business wants password reset capability. Instead of starting over:

  1. 1. Open existing DSL in Build Studio
  2. 2. Add the new capability
  3. 3. Regenerate the deployment package
  4. 4. Re-run compliance validation
  5. 5. Deploy the update

When Platforms Change

The business decides to deploy on AWS for a subsidiary. No redesign needed:

  1. 1. Open the existing DSL
  2. 2. Select "Generate for AWS Bedrock"
  3. 3. Get Bedrock agent definition, Lambda scaffolds, and CloudFormation templates
  4. 4. Deploy to AWS

One design, multiple platforms, zero rework.

Timeline

From business idea to production agent in days, not months.

For agents with only platform-native tools, the entire process from idea to Teams deployment can happen in under an hour.

01

Design

15-30 min

Stakeholder + Build Studio

02

Generate

Instant

Automated

03

Validate

Instant

Automated compliance

04

Handoff

Instant

Auto-generated specs

05

Engineering

2-6 weeks

Custom integrations

06

Deploy

1 command

DevOps / Platform admin

Output

What You Get at the End

Every artifact is version-controlled, traceable, and updatable. Nothing is lost between design and deployment.

Complete output file tree

src/agents/it-support/
  design/
    it-support.dsl.yaml         — Platform-agnostic agent definition
    requirements.md             — Original requirements (preserved)
    design-decisions.md         — Why decisions were made (ADRs)
    research-notes.md           — AI research with sources
  generated/
    copilot-studio/
      ITSupportAgent_1_0_0_0/   — Complete importable solution
      compliance-report.md      — 73-point validation results
      governance-plan.md        — 8-domain governance document
    engineering/
      engineering-handoff.md    — Development roadmap
      mcp-plugins/
        servicenow-spec.md      — Plugin specification

Ready to build your first agent?

See how Agent Accelerator takes you from a business need to a production-deployed, compliance-validated AI agent. Start with a proof of concept.

Australian entity · ISO 27001 certified · Your data stays yours