Home › AI Agents › AI Agent Development
AI agent development and Copilot Studio consulting
AI Agents Built with Microsoft Copilot Studio
Agents that answer questions from your own content, take real action in your systems, and hand off to your team when needed, all inside Microsoft 365.
The basics
What is an AI agent?
An AI agent understands a request in plain language, finds the answer in your approved business content, and can complete a task for the person asking, like logging a ticket or submitting a request.
We build them with Microsoft Copilot Studio, inside your Microsoft 365 environment, so they use your content, apps and security settings. When an agent isn’t sure, it passes the conversation to a person.
Traditional chatbot
- Fixed, scripted replies
- Matches keywords
- No real actions
- Dead end when stuck
- Separate login and permissions
AI agent
- Answers from your approved content, with sources
- Understands everyday language
- Creates tickets, submits requests, updates records
- Hands off to the right person
- Your Microsoft 365 sign-in and permissions
Agents we build
AI agents for the work your team repeats every day.
Most companies start with one agent that removes a clear daily burden, then add more once it proves its value.
HR policy and leave agent
Answers questions about benefits, PTO and company policies, checks leave balances, and submits leave requests for approval.
IT help desk agent
Walks people through common fixes, resets and access requests, and creates a ticket with the right details when it can’t solve the issue.
Employee knowledge agent
One place to ask about processes, templates, contacts and “how do I…” questions, with answers drawn from your intranet and documents.
Onboarding agent
Guides new employees through their first weeks, answers common questions, and reminds them and their managers about pending tasks.
Customer support agent
Answers product and order questions on your website or WhatsApp, collects details, and hands complex cases to your support team.
Document Q&A; agent
Finds and summarizes information across contracts, policies and manuals, and points people to the exact source document.
Connections
What your agent can connect to.
An agent is only as useful as the information and systems it can reach. We connect it to the right sources and tools, and nothing more than it needs.
Knowledge
- SharePoint and OneDrive
- Dataverse tables
- Azure SQL and SQL Server
- Uploaded files and attachments
- Public websites and Azure AI Search
- Fabric IQ and Work IQ preview
Tools
- Agent flows and workflows
- Prebuilt and custom connectors
- REST APIs and MCP servers
- Computer use for apps without an API
- Other agents, over A2A
- Hand-off to a person
Channels
- Microsoft Teams and Microsoft Copilot
- SharePoint pages
- Your website and mobile app
- WhatsApp and Facebook
- Voice and telephony preview
- Event triggers (no one has to ask)
AI models
- GPT-5.5 Chat (default)
- Claude Sonnet 5 and Claude Opus
- Model per agent or per prompt
- Your admin allows external models
- Memory and reusable skills
- Content safety tests on every release
Microsoft Learn, “What’s new in Copilot Studio”, October 2026. Preview features are confirmed as production-ready during design.
Types of agents
The right kind of agent for each job.
Microsoft offers several ways to build agents. We recommend the simplest one that does the job well, so you don’t pay for complexity you don’t need.
Copilot Studio agents
Custom agents with your own knowledge, tools and channels. Best for business processes that need actions, integrations and governance.
Agents for Microsoft Copilot
Extend Microsoft Copilot with your instructions, knowledge and actions, using Copilot’s own model and orchestration. Best when your team already uses Copilot.
Agents in SharePoint
Answer questions from one SharePoint site or library. A fast way to make a policy or project site easier to use.
Agents that act on their own
Start work when something happens, like a new file, an email or a record change, instead of waiting for someone to ask.
Multi-agent solutions
A front-door agent that routes HR questions to an HR agent and IT questions to an IT agent, with child and connected agents.
Security and governance
Safe by design, from day one.
Your security team should be comfortable with every agent we launch. We use the controls Microsoft builds into Copilot Studio and Microsoft 365, and we set them up properly.
Permissions respected
When an agent uses SharePoint, it only shows content the person asking is already allowed to see.
An identity per agent
Every agent gets a Microsoft Entra Agent ID, so Conditional Access and connector permissions apply to it like a user.
Secure sign-in
People sign in with their Microsoft Entra ID account, so the agent knows who it’s talking to.
Data policies
Power Platform data policies control which connectors, knowledge sources and channels an agent may use.
Sensitivity labels
Answers drawn from labeled SharePoint content show the label, so people know how to handle the information.
Hand-off to people
Agents pass conversations to Dynamics 365 Customer Service, or escalate through Teams or a ticket, with context included.
Audit and oversight
Changes are recorded in Microsoft Purview audit logs, and Agent 365 gives admins one inventory of every agent.
The honest part
AI agents can still get things wrong. That’s why we design them to show their sources, stay within approved content, and hand off to a person when they’re unsure.
Problems we fix
Common Copilot Studio problems, and how we solve them.
Most agent projects run into the same handful of issues. Here’s how we deal with each one.
We clean up knowledge sources, write clear agent instructions, and test against real questions with agent evaluations until answers are reliable.
We use Microsoft sign-in so SharePoint answers follow each user’s permissions, and review sharing and sensitivity labels.
We estimate usage up front, then set per-agent monthly limits with alerts in the Power Platform admin center.
We set up environments, sharing limits and an agent inventory, so you know what exists and who owns it.
We define success criteria, fix blockers, and move the agent through test into production with a proper release.
We move agents between environments as solutions with pipelines, so nothing is edited directly in production.
Our process
How we build your agent.
You see it working early, and nothing goes live until it passes testing with your own questions.
Assess
Best first use case; content, systems and license check
Design
Conversations, knowledge, actions and hand-off rules
Build
In your tenant, AI-assisted, with senior review
Evaluate
Test sets from your real questions, multi-turn
Launch
Publish to your channels, train users and admins
Improve
Monitor answers, usage and Copilot Credits
Tested before launch
Tested with your real questions.
We build a test set from questions your team actually asks, run Copilot Studio agent evaluations, including multi-turn conversations and content safety checks, and fix weak answers before anyone uses the agent. You see the results and sign off.
Copilot Studio consulting, done properly
How changes reach production.
We treat agents like any other business system. They’re built in development, tested in a separate environment, and released to production only after you approve.
- Power Platform solution with environment variables
- Source control through Dataverse Git
- AI-assisted build, senior review
- Agent evaluations with real questions
- User acceptance testing
- Credit usage check
- Authentication and channels reconfigured
- Monitoring and analytics on
- Nothing edited directly in prod
For larger programs, we also use Microsoft’s Copilot Agent Kit for batch testing and agent inventory, so quality checks run the same way every release. Classic agents are upgraded to generative orchestration and the new build experience before they go through this process.
Licensing, explained
What it costs to run an agent.
Copilot Studio agents use Copilot Credits, and each response uses more or fewer depending on what the agent does. There are four ways to cover them.
Prepaid packs
A set monthly amount of Copilot Credits for your organization.
Pay-as-you-go
Billed monthly through Azure for the credits your agents actually use.
Prepurchase plan
A one-year commitment of credits you can draw on.
Microsoft Copilot users
Many of their agent answers in Teams, SharePoint and Copilot Chat use no credits.
Choosing the right tool
Agent Builder, Copilot Studio or Microsoft Foundry?
Microsoft gives you three ways to build an agent. Here’s how we help clients choose.
Agent Builder
in Microsoft Copilot- Best for
- Simple personal or team agents that answer from a few documents
- Who builds it
- Your employees, with no build work
- Our role
- Guidance and training
Copilot Studio
business agents- Best for
- Agents that use your data, take actions, run in several channels and need governance
- Who builds it
- Our team, with your input
- Our role
- Full design, build, testing and support
Microsoft Foundry
pro-code- Best for
- Agents that need custom models or complex orchestration
- Who builds it
- Developers
- Our role
- Full design and build on Foundry, connected to Copilot Studio
Clear expectations
What you get, and what we need from you.
What we deliver
- A working agent in your tenant
- Test set and evaluation results
- Development, test and production environments
- Admin guide and user guide
- Training for users and admins
- A full handover
What we need from you
- A business owner for the agent
- Access to the content and systems it will use
- The right Microsoft licenses
- Real example questions from your team
- Feedback during demos and final sign-off
Who owns it
- The agent lives in your Microsoft 365 tenant
- You own the agent and everything we build for it
- Your team can manage it after handover
- Optional support plan if you want us to stay involved
Enablement
Train your makers to build with confidence.
Many teams want to build some agents themselves. We train your makers and admins on Copilot Studio, set up simple standards, and stay available when they need help.
Hands-on Copilot Studio sessions using your own scenarios.
Environments, data policies, monitoring and credits.
Naming, instructions, testing and release checklists.
Office hours and reviews as your team builds.
Free offer
Free AI Agent Readiness Assessment
Not sure where an agent would help most? In 30 minutes, we’ll review your goals and your Microsoft 365 setup, and give you a clear, practical starting point.
Book Your Free Assessment- Your top three agent opportunities
- A quick check of your content and data
- Licensing and running cost guidance
- A recommended first step
Why TSinfo
Why teams choose us for AI agents.
FAQ
AI agent development questions
What’s the difference between Microsoft Copilot and Copilot Studio?
Microsoft Copilot is the AI assistant built into apps like Teams, Outlook and Word. Copilot Studio is the platform for building your own agents, with your knowledge, actions and channels. You can also use Copilot Studio to extend Microsoft Copilot with agents for your business.
Do we need Microsoft Copilot licenses to use an agent?
No. Copilot Studio agents can run in Teams, on your website and in other channels without Microsoft Copilot licenses, using Copilot Credits instead. If your people already have Microsoft Copilot, many of their agent answers in Teams, SharePoint and Copilot Chat don’t use credits, which can lower running costs.
What does it cost to run an agent?
Running costs are measured in Copilot Credits, bought as prepaid packs, a prepurchase plan or pay-as-you-go through Azure. Usage depends on how many people use the agent and what it does. We estimate your credits during the assessment using the Copilot Studio agent usage estimator.
Can the agent see confidential data?
Only what you allow. We choose the agent’s knowledge sources with you, and when it uses SharePoint, it only shows content the person asking already has permission to see. Power Platform data policies, sensitivity labels and the agent’s own Microsoft Entra Agent ID add further control.
Can the agent work with Salesforce, ServiceNow or our own systems?
Yes. Agents can use prebuilt and custom connectors, REST APIs and MCP servers to read and update other systems, and computer use for applications without an API. Real-time knowledge connectors for systems such as Salesforce and ServiceNow are still in preview, so we confirm what’s ready for production during design.
What happens after the agent goes live?
We monitor usage, answer quality and Copilot Credit consumption, rerun evaluations as content changes, and train your team to manage the agent. You can keep us on a support plan or run it yourself. Learn more in Microsoft’s Copilot Studio overview.
Related services
Keep exploring
Ready to put an AI agent to work?
Tell us which questions or tasks take up your team’s time. We’ll show you what an agent could handle.