01
Build & orchestrate
Choose the authoring and runtime layer based on the team, channel, and level of control.
Microsoft FoundryFoundry Agent ServiceCopilot StudioGitHub + Visual Studio Code
Azure AIField GuideStart a customer scenario →Customer conversation toolkitUpdated Jul 2026
Choose the right Azure AI building blocks.
A practical, interactive guide to position the Azure AI portfolio, design agentic solutions, and explain where Foundry, Copilot Studio, models, knowledge, Foundry Tools, safety, and Azure databases fit.
12 service & solution areas3 deep dives6 architecture patterns
How the pieces connect
Business
outcome
outcome
＋ Entra ID · Private Link · API Management · Azure Monitor
01 · Quick comparison
One portfolio. Different jobs.
Select a service to get a customer-ready explanation, then use the filters to focus the discussion.
02 · Azure AI landscape
Navigate the full solution stack.
Start with the customer outcome, then identify the layers needed to build, ground, secure, operate, and scale it. Most production solutions span several groups.
02
Models & intelligence
Select, compare, adapt, and deploy the right model for each workload and operating environment.
Foundry ModelsAzure OpenAI in Foundry ModelsAzure Machine LearningFoundry Local
03
Knowledge & Foundry Tools
Ground agents in governed enterprise knowledge and add prebuilt content, language, vision, speech, and safety capabilities.
Foundry IQ (Azure AI Search)Content UnderstandingDocument IntelligenceSpeech · Vision · Language
04
Control & operations
Evaluate quality and safety, trace behavior, control access and networking, and monitor production AI systems.
Foundry Control PlaneAzure AI Content SafetyAzure Monitor + Application InsightsEntra ID + Azure Policy
05
Data & application runtime
Store operational data, vectors, and application state, and run the surrounding application, API, and workflow services.
Azure Cosmos DBAzure SQL DatabaseAzure Database for PostgreSQLFunctions + Container AppsMicrosoft Fabric + Azure Databricks
Customer rule of thumbExperience + intelligence + knowledge + actions + controls + runtime = a production AI solution.
03 · Foundry deep dive
Explore the AI application factory.
Use these seven views to explain the current Microsoft Foundry structure: platform, Agent Service, Models, IQ, Tools, Control Plane, and Local.
Unified Azure PaaS
Build and operate AI apps and agents in one governed platform.
Microsoft Foundry brings agents, models, tools, evaluation, tracing, monitoring, identity, networking, and policy controls into one platform experience.
Use whenUse it when engineering teams need a repeatable, governed path from experimentation to production.
Commercial lensThe platform can be explored without a platform fee; deployed models, agents, tools, data, and Azure resources carry their own meters.
Core capabilities
01Portal, SDK, CLI, and REST development
02Shared project resources for teams
03Agent and application lifecycle
04RBAC, networking, policy, and cost controls
01DesignOutcome · data · risk→
02BuildModels · agents · tools→
03EvaluateQuality · safety · cost→
04PublishIdentity · channels · APIs→
05OperateTrace · monitor · optimize
04 · Agentic AI
Match autonomy to business risk.
A Foundry agent has three core components: a model, instructions, and tools. Production designs add knowledge, state, identity, evaluation, and operational controls as required.
Selected pattern
Task agent
Uses approved tools to complete a defined process with validation, approval, and clear transaction boundaries.
- Example
- Create a ticket, prepare a quote, update a CRM record
- Required controls
- Tool permissions · Human approval · Idempotency · Transaction audit
Reference blueprint
Core agent plus production services
01ModelReason and generate
02InstructionsDefine role and boundaries
03ToolsRead or change approved systems
04KnowledgeGround in trusted context
05State & memoryPreserve useful context
06ControlsIdentity, safety, evaluation
Multi-agent pattern
Coordinate specialists—do not create a crowd.
Start with a single agent. Add narrowly scoped specialists only when domain, permission, model, or workflow boundaries justify the extra latency and operational complexity.
Lead / planner→Knowledge specialistAction specialistRisk reviewer→Approved outcome
05 · Copilot Studio capabilities
From low-code idea to governed agent.
Copilot Studio coordinates instructions, knowledge, topics, tools, flows, triggers, and channels through a graphical experience for business and IT teams.
Author
Design agents and agent flows visually
Define instructions, topics, inputs, variables, prompts, triggers, and flows through a graphical low-code canvas or natural language.
Ask the customer
Who owns the agent design, and how often will makers and IT teams change it?
1Instructions and generative orchestration
2Topics and conversation paths
3Inputs, triggers, and agent flows
4Test canvas and iterative authoring
Maker-led, low-code delivery
Best when speed, graphical authoring, Microsoft channels, connectors, agent flows, and IT-governed lifecycle management are priorities.
or combine
Developer-led, pro-code control
Best for prompt agents or custom Hosted agents when application UX, tools, frameworks, code, or runtime behavior needs engineering control.
06 · Guided decision
Convert requirements into an architecture position.
Complete five design decisions aligned to the Foundry stack and Azure Well-Architected guidance. The output is a preliminary architecture position—not a final design.
01
WorkloadKnowledge agent
02
OrchestrationFoundry Agent Service
03
Data patternEnterprise knowledge
04
AutonomyRead and recommend
05
TopologyPrivate and regulated
1
Select the workload archetypeDefine the dominant workload before selecting products.
2
Choose the orchestration and delivery modelBase this on ownership, extensibility, runtime control, and target channels.
3
Select the grounding and data patternChoose according to freshness, authorization, query shape, and transactional requirements.
4
Define the permitted autonomyUse the lowest autonomy that can achieve the outcome and preserve approval boundaries.
5
Set the production topologyApply nonfunctional requirements before estimating the final service footprint.
07 · Architecture patterns
Show how the pieces connect.
Six conceptual patterns aligned to Microsoft reference architecture guidance. Use them to explain flow, controls, ownership, and cost drivers before detailed design.
01Enterprise sourcesSharePoint · Fabric · files
→
02Foundry IQAzure AI Search · retrieve · cite
→
03Foundry Agent ServiceReason · cite · use tools
→
04Secure experienceWeb · Teams · M365 · API
Shared foundation
Foundry Control PlaneMicrosoft Entra IDVNet + Private LinkKey VaultAPI Management / FirewallContent SafetyEvaluation & tracingMonitor + App Insights
08 · Pricing concepts
Estimate the shape of cost.
Use the six solution cost stacks to explain what is metered and where hidden consumption appears. Editable figures are teaching assumptions, not a quotation.
Usage-based model
Input tokens + output tokens
Different models and deployment types have different rates. Output usually costs more because generation consumes more compute.
Monthly cost(input M × rate) + (output M × rate)
PTU alternativeReserve model processing capacity for predictable, sustained production throughput. Billed on provisioned capacity, not consumed tokens.
Illustrative monthly model cost$6012M × $2.5 + 3M × $10
Important Microsoft-published list figures are identified where shown. All editable rates and calculated totals are planning assumptions—not quotations. Pricing varies by region, model, deployment, currency, agreement, tax, and date.
09 · Supporting services
Production architecture extends beyond the AI layer.
The model is one line in a production architecture. Security, integration, data, monitoring, and operations complete the customer solution.
10 · Customer scenarios
Turn the technology into a customer story.
Choose an industry to frame the business need, recommended Azure stack, discovery questions, cost drivers, and a practical next step.
Banking & financial servicesCustomer-ready
Secure customer-service knowledge assistant
Service teams need fast, consistent answers from approved policies, product documents, and internal procedures without exposing regulated data.
Business outcomeShorter handling time, more consistent responses, and traceable answers with citations.
Recommended direction
Foundry Agent ServiceAzure OpenAI in Foundry ModelsFoundry IQ (Azure AI Search)API ManagementMicrosoft Entra ID
Ask the customer
- Which documents are approved as answer sources?
- Must the assistant identify the customer or access account data?
- What audit, residency, and human-review controls are required?
Commercial lens
Model tokens · Search units · API traffic · Monitoring and retention
Run a controlled proof of concept on one service journey and an approved document set.
11 · Customer talk track
A clean way to explain the landscape.
Use this progression in the first five minutes of a discovery call.
“Microsoft Foundry is Azure’s unified platform for building and operating AI apps and agents. Foundry Models provides model choice, including Azure OpenAI. Foundry Agent Service provides managed prompt and Hosted agent options. Foundry IQ, built on Azure AI Search, grounds agents in enterprise knowledge. Foundry Tools adds content, document, speech, language, vision, and safety capabilities. Copilot Studio is the graphical low-code route for agents, flows, connectors, and business channels.”
- Who will use the experience, and through which channel?
- Which data is authoritative, classified, regional, or permission-sensitive?
- Must the agent only answer, or also change business state?
- Is the delivery team primarily makers, developers, or both?
- What availability, latency, recovery, and cost boundaries apply?
- “Foundry is another model.”
It is the Azure platform around agents, models, tools, knowledge, and operations. - “The model already contains our enterprise knowledge.”
Foundry IQ and Azure AI Search retrieve governed customer context separately from generation. - “AI-ready databases replace enterprise search.”
Databases serve live operational patterns; Foundry IQ is optimized for governed enterprise knowledge and retrieval. Choose by access pattern. - “Copilot Studio is one flat price.”
Internal Microsoft 365 use and standalone Copilot Studio have different licensing contexts; standalone usage is measured in Copilot Credits.