Cloud platformsMicrosoft Azure

Build on Azure and Microsoft 365

Agents in the Microsoft world,
where your people already work.

Microsoft Foundry, Copilot Studio, Microsoft 365 and Azure, engineered into agents that know your business, respect your permissions and hand off cleanly to people.

Claude, OpenAI and other Foundry Models feed Microsoft Foundry, surrounded by Agent Service, Foundry IQ, model router, Entra Agent ID, guardrails and the Control Plane, above a knowledge and data layer, with a human approval stepService desk agentMicrosoft Teams · built on Foundry Agent ServiceMy laptop can’t reach the VPNsince this morning.This matches a known certificate issue.The fix is step 4 of the VPN runbook.Foundry IQ · VPN runbookYour open ticketPROPOSED ACTIONReissue the VPN certificate for this deviceActing as: Entra Agent IDApproveEditEvery step traced and evaluated in Foundry Control Plane
Talk about your projectExplore the Azure stack

Copilot, Copilot Studio or Foundry?

Pick the Microsoft surface that fits the work.

Microsoft offers several ways to put AI in front of people. Choose the situation closest to yours.

Which situation is closest to yours?

Our recommendation

Microsoft 365 Copilot

The work happens in content people already have. Copilot reaches it with their own permissions, so the job is adoption and measurement rather than a build.

First step. Pick two or three teams, agree what good looks like and measure it for a month.

  • Microsoft 365 Copilot
  • Work IQ
  • Agent 365

Our recommendation

Copilot Studio

A low-code agent with Power Platform connectors and Power Automate flows is quicker to change, and the team can own it after launch.

First step. Map the process, the systems it touches and who approves each action.

  • Copilot Studio
  • Power Automate
  • Dataverse

Our recommendation

Microsoft Foundry

Foundry Agent Service and Microsoft Agent Framework give full control over tools, memory and evaluation, and agents can still be published into Teams and Microsoft 365 Copilot.

First step. Build the hardest step as a Challenge24 prototype and evaluate it on real cases.

  • Foundry Agent Service
  • Agent Framework
  • Foundry IQ

Our recommendation

Foundry Models in your application

The user stays in the product they know. The application calls Foundry Models directly, behind its existing identity and logging.

First step. Choose the one screen where AI saves the most time and prototype it.

  • Foundry Models
  • Model router
  • App Service

How we see it

Your agents should inherit your identity model, not invent a new one.

Microsoft now gives agents their own Entra identities, a registry in Agent 365, a control plane in Foundry and a home in Teams and Microsoft 365 Copilot.

We connect those pieces to your data and processes, so an agent sees only what its user may see, acts only where it is allowed and leaves a trail your security team already knows how to read. Where Microsoft 365 Copilot already does the job, we say so.

An illustrative design

A service-desk agent in Teams.

An employee asks for help in Teams. The agent finds the relevant policy, reads the open ticket and drafts the fix. A colleague approves it before it is applied.

  1. Employee asksTeams · Microsoft 365 Copilot

  2. Agent plans the stepsFoundry Agent Service
  3. Finds the policyFoundry IQ · Azure AI Search
  4. Reads the ticketMCP tool → service desk
  5. Acts as itselfEntra Agent ID · Guardrails
  6. Colleague approvesApproval card in Teams
  7. Traced and scoredFoundry Control Plane

The Azure AI stack

Six layers. We build across all of them.

Open a layer to see the Azure services involved and the part we take on.

ExperienceWhere people ask, review and approve.
  • Microsoft 365 Copilot
  • Teams
  • Copilot Studio
  • Your web and mobile apps

What we build. We decide with you whether a workflow belongs in Microsoft 365 Copilot, a Copilot Studio agent or a custom application, then design the handoff to people.

AgentsWhere the plan is made and the steps are run.
  • Foundry Agent Service
  • Microsoft Agent Framework
  • Copilot Studio agents
  • A2A

What we build. We build agents with Microsoft Agent Framework or Copilot Studio, run them on Foundry Agent Service and keep multi-step work recoverable.

Tools and actionsWhere the agent touches your systems.
  • MCP tools
  • Logic Apps
  • Power Automate
  • Microsoft Graph
  • API Management

What we build. We connect Dynamics 365, line-of-business APIs and Microsoft Graph as tools, each with a contract, a permission and a failure path.

Knowledge and dataWhere answers come from.
  • Foundry IQ
  • Azure AI Search
  • Fabric and OneLake
  • Work IQ
  • Azure SQL
  • Cosmos DB

What we build. We design retrieval that respects SharePoint and document permissions, connect Fabric data and show people the source behind every answer.

ModelsWhere reasoning happens.
  • Foundry Models
  • Model router
  • Claude
  • OpenAI models

What we build. We evaluate the models available in your tenant on real tasks and use the model router where a cheaper model does the job.

Trust and operationsWhere you keep control.
  • Entra Agent ID
  • Agent 365
  • Foundry Control Plane
  • Guardrails and controls
  • Defender
  • Azure Monitor

What we build. We give each agent an Entra identity under Conditional Access, register it in Agent 365, apply guardrails and trace every run in Foundry.

What changed on Azure

A year of Microsoft agent releases.

The releases that change how we design and build on Azure, with a link to each announcement.

  1. Azure AI Foundry became Microsoft Foundry

    Foundry IQ for knowledge, Fabric IQ for business data and a Control Plane for agent fleets arrived together at Ignite.

    Read the announcement (opens in a new tab)

  2. Azure SRE Agent reached GA

    An operations agent that investigates incidents across Azure resources and proposes mitigations for an engineer to approve.

    Read the announcement (opens in a new tab)

  3. Evaluations and tracing became standard

    Evaluations, monitoring and tracing in Foundry reached general availability, so agent quality can be measured in production.

    Read the announcement (opens in a new tab)

  4. Agent 365 for governing agents

    A single place to register, secure and govern agents, joined to Microsoft Entra Agent ID.

    Read the announcement (opens in a new tab)

  5. Claude GA in Microsoft Foundry

    Claude models became generally available in Foundry, including use inside Foundry Agent Service.

    Read the announcement (opens in a new tab)

Our services on this platform

From first idea to production on Azure.

How each stage of our work maps to Azure. Start at whichever stage fits where you are.

01Start

Decide what is worth doing.

  • AI Innovation Workshop

    Sort ideas into what Microsoft 365 Copilot already covers and what needs a custom build.

    Microsoft 365 Copilot · Copilot Studio · Foundry

  • Challenge24

    A Foundry prototype your people can try in Teams, in three business days.

    Foundry · Claude · Foundry IQ

02Build

Build the system.

  • Applied AI and agents

    Agents on Foundry Agent Service, published where your people work.

    Agent Framework · Agent Service · MCP

  • AI-infused software

    .NET and web applications with AI built into the workflow.

    App Service · Functions · AKS · Azure SQL

03Strengthen

Make it dependable.

  • Enterprise AI Hardening

    Agent identities, Conditional Access, guardrails and evaluations before wider rollout.

    Entra Agent ID · Guardrails · Defender

  • Legacy Modernization

    .NET Framework and Java upgrades, with migrations planned and tested.

    GitHub Copilot modernization · Azure Migrate

04Run

Keep it healthy.

  • SRE as a Service

    Azure Monitor and Azure SRE Agent prepare the evidence. Our engineers decide.

    Azure Monitor · Azure SRE Agent

  • AI FinOps

    Model and Azure spend attributed by team and product, with budgets that hold.

    Cost Management · FinOps hubs

  • AI Governance and Evaluation

    Agents registered, risk-assessed and evaluated against the Govern and Secure guidance.

    Agent 365 · Foundry Control Plane

05Extend

Add capacity.

  • Engineering teams

    Engineers fluent in .NET, Power Platform and Azure, inside your team.

    Bicep · Azure DevOps · GitHub

Claude on Azure

Claude in Microsoft Foundry and Microsoft 365

Claude models are generally available in Microsoft Foundry and can power agents in Foundry Agent Service. Claude is also available inside Microsoft 365 Copilot and Copilot Studio, so the same model family can serve both your custom agents and your everyday tools.

How we build with Claude

Under the agents

The cloud foundation we build on.

Compute

App Service · Functions · Container Apps · AKS

Hosting that fits the workload, the release process and the team that will run it.

Data and analytics

Fabric · OneLake · Azure SQL · Cosmos DB · Power BI

Application data, analytics and reporting with deliberate access boundaries.

Integration

Service Bus · Event Grid · Logic Apps · API Management

Messages, events and APIs with retries and clear failure handling.

Networking

Virtual Network · Private Link · Front Door

Private paths to models and data, and defined boundaries around each service.

Security and identity

Entra ID · Key Vault · Defender · Sentinel

Identity for people and agents, secret handling and security monitoring in one model.

Delivery and hybrid

Bicep · Azure DevOps · GitHub Actions · Azure Arc · Azure Local

Versioned infrastructure, repeatable releases and consistent management across cloud and on-premises.

Questions

Common questions.

Microsoft 365 Copilot, Copilot Studio or a custom Foundry agent?
Use Microsoft 365 Copilot when the job is general productivity on content people already have. Use Copilot Studio when a business team needs a focused agent with connectors. Build on Foundry when the agent needs custom logic, deep integration or strict evaluation. We help you place each workflow.
What is Microsoft Agent Framework?
It is Microsoft’s open-source framework for building agents in .NET and Python, succeeding Semantic Kernel and AutoGen. It supports MCP for tools and A2A for agent-to-agent work. We use it for custom agents that run on Foundry Agent Service.
How do agents get permissions in our tenant?
Each agent gets its own Microsoft Entra Agent ID, so Conditional Access, access reviews and audit logs apply to it the way they apply to people. We scope its tools narrowly and keep consequential actions behind an approval.
Which model should we use?
We evaluate the models available in your environment, including Claude and OpenAI models, on representative tasks. Quality, latency and cost are made explicit, and the model router can send simpler requests to cheaper models.
Can retrieval respect SharePoint permissions?
Yes. We design retrieval with Foundry IQ and Azure AI Search so results are filtered by the user’s access, and we test that boundary with real accounts before launch.
Can you help with sovereign or disconnected requirements?
We assess what your obligations require, from EU data boundaries to disconnected operation with Azure Local and Foundry Local, and design the architecture around the options that fit.
How do you manage ongoing AI costs on Azure?
We make spend visible by team and workload with Cost Management and FinOps hubs, then review model choice, routing, caching and capacity against quality.
Can you modernize our .NET applications?
Yes. We use GitHub Copilot modernization to accelerate upgrades, keep the business rules intact and test behaviour against the existing system before cutover.

Explore another platform

Amazon Web ServicesGoogle CloudIBM Cloud & watsonx

Let’s start with the work you want to improve.

Tell us about your application, your Azure environment, or the idea you want to explore.

Talk about your project