Microsoft Copilot Studio vs a Custom AI Agent: What It Costs and When Each Wins (2026)

Microsoft Copilot Studio (also searched as MS Copilot Studio or Copilot Agent Studio) is Microsoft's low-code tool for building AI agents: assistants with their own instructions, knowledge and actions that run in Teams, Microsoft 365 Copilot, websites and apps. It is also one of the most searched AI tools in business software. Around 40,500 people a month search Google in the US for "copilot studio", 14,800 in the UK, and another 1,900 in the US ask "what is copilot studio" (DataForSEO, 18 October 2026). The searches that carry buying intent are about money: "microsoft copilot studio pricing" (1,000 a month in the US), "copilot studio pricing" (720 US, 260 UK), "copilot studio license" (720 US) and "copilot studio cost" (320 US).

When we asked AI assistants whether a company should use Copilot Studio or hire a firm for custom AI agents, the ChatGPT app searched for Microsoft's "official" pricing and licensing pages on every run, in all three countries. Several pages that rank for this question quote Microsoft's older per-message pricing, which Microsoft replaced with Copilot Credits on 1 September 2025. So this guide quotes Microsoft's own current figures, with dates and links, then sets out the limits, the Microsoft ladder of build tools, and when a custom agent is the better buy for a company of 10 to 50 staff, 50 to 500, or 500 and up.

What is Microsoft Copilot Studio?

Microsoft's Copilot Studio overview(updated 17 October 2026) calls it "a graphical, low-code studio for building and managing AI-powered agents and workflows". Low-code means you build by describing the agent, filling in forms and drawing flows, with little or no programming. Microsoft Copilot Studio AI agents follow your instructions, answer from the knowledge sources you connect, and take actions through tools and connectors. Microsoft says agents can work with employees and customers across Microsoft Teams, Microsoft Copilot, websites and mobile apps.

How to access Copilot Studio: open copilotstudio.microsoft.com in a modern browser and sign in with your work account. The Microsoft Copilot Studio free trial lets one person sign up and build agents, test them in the test chat panel, and stop there, because a trial agent cannot be published. Admins can block these individual sign-ups, which Microsoft's admin documentation calls viral sign-ups; the "Microsoft Copilot Studio viral trial" people search for is this same individual trial. For Copilot Studio training, Microsoft Learn's own documentation is free and is the source for most figures on this page.

Copilot Studio pricing at a glance (read 18 October 2026)

Microsoft prices Copilot Studio in two layers. People who build or use agents need a licence or a role. The agents themselves draw down Copilot Credits, a usage unit pooled across your whole Microsoft 365 tenant (the tenant is your company's Microsoft account space). Here is the Copilot Studio license cost and the Microsoft Copilot Studio price per credit as the Microsoft Copilot Studio pricing page and its sister pages listed them on 18 October 2026.

Two numbers are worth working out. A $200 pack of 25,000 credits is $0.008 a credit, so packs run about 20% cheaper than the $0.01 pay-as-you-go rate shown in Microsoft's estimator, if you use the whole pack. And Microsoft says the switch from "messages" to "Copilot Credits" on 1 September 2025 changed neither the quantity per pack nor the pay-as-you-go rate (Microsoft Learn, standard licensing page). Any guide still quoting a separate per-message price list is describing the same money under the old name.

Copilot vs Copilot Studio: which one you are paying for

"Copilot vs Copilot Studio" is a common search (260 a month in the US) because the names overlap. Microsoft 365 Copilot is the assistant each licensed person uses inside Word, Excel, Outlook and Teams. Copilot Studiois the workshop where agents get built. Microsoft's pricing page puts the line here: Copilot Studio included with Microsoft 365 Copilot lets licensed users "build and use internal agents within Microsoft 365" at no additional cost, while the standalone Copilot Studio licence is what lets you publish to external channels such as websites, apps and social platforms.

So the Microsoft 365 Copilot Studio license question has a short answer. If your agent only serves staff who already pay $30 a month for Microsoft 365 Copilot, you may need nothing more. If it serves customers, partners, or staff without that licence, the Microsoft Copilot Studio license you need is the standalone one: credit packs or pay-as-you-go. On 18 October 2026, Microsoft's pricing page mentioned Copilot Studio Lite only in a footnote, with no price of its own; the lightweight builder it describes in the same section is Agent Builder, covered in the ladder below.

What a Copilot Credit buys

Microsoft Learn's billing rates page (updated 15 October 2026) sets the credit price of each thing an agent does. The third column is the one most guides skip.

The "No charge" column has conditions. Microsoft says it covers employee-facing use when the user holds a Microsoft 365 Copilot licence and the agent runs under that user's own sign-in, within fair usage limits. Computer-using agents (agents that click through screens like a person) are not included. Reasoning models add a premium meter on top of the feature rate. Copilot Studio vs Power Automate is a billing question as well: Microsoft says ordinary Power Automate cloud flows use Power Automate licensing, not Copilot Credits, while agent flows built inside Copilot Studio draw credits.

Microsoft's own worked examples show how fast the numbers move. A website customer support agentgiving four classic answers and two generative answers per visitor, for 900 customers a day, uses [(4 x 1) + (2 x 2)] x 900 = 7,200 credits a day, in Microsoft's words. Our arithmetic on Microsoft's list prices: 7,200 x 30 days is 216,000 credits a month, which needs nine $200 packs ($1,800 a month), or about $2,160 on pay-as-you-go at $0.01 a credit. A sales performance agent grounded on Microsoft Graph, used by 50 licensed and 100 unlicensed staff, is billed only for the 100 unlicensed users: 4,800 credits a day. The licensed 50 cost nothing extra.

How to estimate your Copilot Studio cost in five steps

Searches for a "copilot studio pricing calculator" lead to Microsoft's free Copilot Credit Estimator. Microsoft's own disclaimer on it says it "should not be used as a pricing calculator", and the estimate is not a binding offer. Use it as a forecast, then check against real usage. These five steps use only Microsoft's rates and your own volumes.

1. Split users into two groups: staff with Microsoft 365 Copilot licences (internal use not charged) and everyone else, including customers.
2. For one typical conversation, count classic answers, generative answers, agent actions and graph-grounded answers, and price each at 1, 2, 5 and 10 credits.
3. Multiply credits per conversation by conversations a month for the charged group only.
4. Divide by 25,000 and round up to get packs at $200 each, or multiply by $0.01 for pay-as-you-go. Add 20% headroom, because unused credits expire monthly and the switch-off sits at 125%.
5. Pilot for a month, read actual use in the Power Platform admin center, and set a per-agent monthly cap there before going wide.

Five Copilot Studio limitations to know before you build

These come from Microsoft Learn's quotas and limits page (updated 15 October 2026) and the billing rates page. None of them is a reason to avoid Copilot Studio. Each is a reason to size the plan before launch.

1. The 125% switch-off.When a tenant on prepaid capacity reaches 125% of it, custom agents are disabled once the current conversation ends. Users then see "There is a billing issue" or "This agent is currently unavailable. It has reached its usage limit." Environments on pay-as-you-go are not cut off this way.
2. AI request limits. Generative requests are capped at 50 a minute and 1,000 an hour per environment on 1 to 10 prepaid packs, 80 a minute on 11 to 50 packs, and 100 a minute on pay-as-you-go. Trial and developer environments get 10 a minute. Microsoft warns customer-facing and autonomous agents can hit these even when monthly volume looks fine.
3. Knowledge and topic caps. 500 knowledge sources and 1,000 topics per agent, 25 SharePoint site addresses per agent with generative orchestration, and SharePoint list answers drawn from the first 2,048 rows.
4. The SharePoint file rule. For makers without a Microsoft 365 Copilot licence in the same tenant, generative answers can only use SharePoint files under 7 MB. With the licence and tenant graph grounding, Microsoft supports files up to 200 MB. Documents labelled confidential or highly confidential are not indexed.
5. Publishing outside the company needs the paid licence. The Microsoft 365 Copilot inclusion covers internal agents. Websites, apps and social channels need packs, pre-purchase or pay-as-you-go, and a trial licence cannot publish at all.

Copilot, Agent Builder, Copilot Studio, Foundry, Agents SDK: the Microsoft ladder

"Agent Builder vs Copilot Studio vs Foundry" is one of Google's related searches for this topic, and Microsoft has five rungs, simplest first. Microsoft's Agent Builder vs Copilot Studio page and its custom engine agents overview (both updated 30 September 2026) set out the differences.

The sentence most competing guides leave out is Microsoft's own: "If you need full control over AI models or orchestration, opt for a pro-code approach with Microsoft 365 Agents SDK or Foundry via Agents Toolkit." Pro-code here means a developer writes the agent in ordinary code. That is what a custom AI agent is, whether it is published into Microsoft 365 or runs on its own.

When Copilot Studio is the right call

Copilot Studio agents are strongest where Microsoft already holds the data and the identity. Pick it when most of these are true:

• The users are your own staff, and many already hold Microsoft 365 Copilot licences, so their use is not charged.
• The answers live in SharePoint, OneDrive, Teams or Dataverse, which Copilot Studio reads natively.
• The actions are standard: create a ticket, start a Power Automate flow, route an approval, update a Dataverse row.
• A prebuilt connector exists for each outside system the agent must touch.
• Your IT team already runs the Power Platform admin center and wants agents governed in the same place.
• You want a working internal pilot fast, built by a maker in your own team rather than a developer.

A 50-person accounting firm in Manchester or a 300-person distributor in Ohio whose staff live in Teams and SharePoint fits this profile well. So does an HR or IT help desk agent in a 2,000-person company. Microsoft's own Copilot Studio examples are an IT help desk triage agent, a customer support agent that raises tickets and hands over to a person, and a sales assistant that reads CRM data and starts an approval.

When a custom AI agent is the right call

A custom agent is code written for your job, on the model you choose, connected to your systems through their APIs. It is the right buy when the job does not sit inside Microsoft, or when credit pricing and platform limits would shape the product more than your customers do.

1. Customer-facing at volume. Website chat, WhatsApp or voice for thousands of customers a day, where every answer would otherwise draw credits and the 50-a-minute limit on small pack counts matters.
2. Systems outside Microsoft. Orders in NetSuite, Odoo, SAP Business One or ERPNext, stock in Shopify, deals in HubSpot or Salesforce, invoices in Xero, or EDI feeds from trading partners.
3. Model choice. You want to pick or switch between OpenAI, Anthropic Claude and Google Gemini, or run a model in your own cloud account.
4. Custom approval steps and audit logs. A person must approve before an order, refund or price change posts, and every step must be logged in a format your auditors already use.
5. Code in your own repository. You want the agent to keep running if you change software vendors, with no dependency on a monthly credit pool.
6. Work that is not a conversation. Reading emailed purchase orders and PDFs, matching invoices, or building quotes from a rule book, where the agent works in the background.

Our guides to buying or building enterprise AI agents and AI agent build vs buy cover the same decision across Microsoft, Salesforce and help desk platforms. For the Salesforce version of this page, see Agentforce pricing vs a custom AI agent.

Copilot Studio vs a custom AI agent, side by side

The middle route: Copilot Studio front end, custom back end

Many companies do not need to pick one side. Copilot Studio agents can call tools on a Model Context Protocol (MCP) server, a standard way of exposing a system's actions to an AI agent. Microsoft says each tool an MCP server publishes becomes available in Copilot Studio automatically, once generative orchestration is turned on. So a Copilot Studio agent in Teams can check stock in NetSuite or draft an order in Odoo, if someone builds the MCP server or API in front of that system.

The other middle route runs the opposite way. A custom engine agent, in Microsoft's words, provides "full control over orchestration, AI models, and data integrations" while still appearing in Microsoft 365 Copilot and Teams. Built with the Microsoft 365 Agents SDK, the logic lives in your code and staff still open it in Teams. Our guide to a custom MCP server for ERP and CRM shows what that back end involves.

Governance and security in each route

Governance means who can build agents, what data they can touch, and who can see what they did. On the Microsoft side, the Copilot Studio security and governance page (updated 15 October 2026) lists the controls an admin gets: data loss prevention (DLP) policies that decide which connectors, knowledge sources and channels an agent may use; maker audit logs in Microsoft Purview; alerts in Microsoft Sentinel; Entra ID sign-in; customer-managed encryption keys; sensitivity labels on SharePoint answers; and a switch to stop generative AI data moving outside the United States. Microsoft also notes that its content moderation covers predefined harmful-content categories, not subjects your admin defines.

On a custom agent, you build the same controls on purpose, and you decide their shape. The ones we would expect in any written plan:

1. An approval queue, so a person approves before the agent posts an order, refund, price or payment.
2. A log of every prompt, tool call and answer, kept in your own account for the period your auditors set.
3. Role-based access, so the agent can only read and change what the signed-in person could.
4. Data kept in your own cloud and AI accounts, with a written answer on where it is processed.
5. A test set of real past cases, run before each release, with the pass rate recorded.

Our AI agent governance policy guide and AI agent security risks and controls go deeper. For an outside reference, the US NIST AI Risk Management Framework is intended for voluntary use and gives a shared vocabulary for these risks.

Copilot Studio pricing in the UK and Australia

Microsoft publishes a local page for each country, and the local price is not a straight currency conversion of the US figure on the day you read it. On 18 October 2026:

For UK firms the bigger question is usually data, not price. The ICO's guidance on AI and data protection sets out what UK GDPR expects on accountability, transparency and accuracy when you deploy AI. Microsoft documents where Copilot Studio stores and processes data on its geographic data residency page. In Australia, the OAIC's guidance on commercially available AI products covers the privacy duties when staff use tools like Copilot Studio. Whichever route you take, write down where customer data goes before launch. See our UK AI agents page and Australian AI agents page.

What a custom AI agent costs to build

Market ranges vary by scope, so treat any single number with care. ProductCrafters' AI agent development cost guide(February 2026) puts building an AI agent at $5,000 to $180,000 or more. Layer3 Labs' AI consulting pricing guide (updated 9 September 2026) puts a production MVP workflow at $5,000 to $25,000 and a full enterprise AI system at $50,000 to $200,000 or more. Those are market ranges, not FactoryJet prices. Our enterprise AI agent cost guide breaks the build side down further.

Before comparing any build quote with a Copilot Studio plan, ask the builder five things in writing: what is in each phase, who owns the code, where the AI and cloud bills sit, what happens when a defect appears after launch, and whether you can see it working on your own data first. Our questions to ask an AI development company and why AI agent projects fail cover the rest.

What else ranks for "copilot studio pricing", and how this guide differs

These are business sites on Google's first page for "copilot studio pricing" (US and UK) and "copilot studio vs custom ai agent" (US) when we checked on 18 October 2026 (DataForSEO), with the title Google showed. Microsoft's own pages held the top spots in both countries. Forums and video sites are left out.

Most of these explain the price list or the Microsoft product family. This guide adds the "No charge" rule for Copilot-licensed staff, the 125% switch-off and the rate limits from Microsoft Learn, UK and Australian list prices read the same day, Microsoft's own pro-code advice, and the two middle routes.

How FactoryJet helps you choose and build

FactoryJet builds custom AI agents and the integrations behind them: APIs and MCP servers for Odoo, NetSuite, SAP Business One, ERPNext and custom ERPs, plus EDI, punchout, Xero, HubSpot and WhatsApp connections, Salesforce and Agentforce work, and AI voice agents. That means we can build the custom back end a Microsoft front end calls, or a custom agent when that fits better. How we work:

• You own the code and the repository from day one.
• AI and cloud bills can run in your own accounts, at cost, with no markup.
• Each phase is scoped, priced and signed on its own.
• Defects are fixed inside the build price for a cover period agreed before work starts.
• Support is optional: each change is quoted and approved, with no monthly minimum.
• You see working software on your own data before you sign.

See AI agent development, AI integration services, AI agents for ERP systems and AI customer support agents. For the time-saved side of the decision, try our AI agent ROI calculator.

Sources

Every page below was opened and read on 18 October 2026.

• Microsoft, Copilot Studio plans and pricing (US)
• Microsoft, Copilot Studio pricing (UK) and (Australia)
• Microsoft, Microsoft 365 Copilot Business
• Microsoft Learn, Billing rates and management (updated 15 October 2026)
• Microsoft Learn, Standard licensing for Copilot Studio (updated 15 October 2026)
• Microsoft Learn, Quotas and limits (updated 15 October 2026)
• Microsoft Learn, Copilot Studio overview (updated 17 October 2026)
• Microsoft Learn, Choose between Agent Builder and Copilot Studio (30 September 2026)
• Microsoft Learn, Custom engine agents for Microsoft 365 (30 September 2026)
• Microsoft Learn, Extend your agent with Model Context Protocol
• Microsoft Learn, Security and governance in Copilot Studio
• Microsoft Learn, Geographic data residency in Copilot Studio
• Microsoft, Copilot Credit Estimator
• ProductCrafters, AI agent development cost (February 2026)
• Layer3 Labs, AI consulting rates and pricing (updated 9 September 2026)
• NIST, AI Risk Management Framework
• ICO, Guidance on AI and data protection
• OAIC, Guidance on privacy and commercially available AI products
• Our own measurement: Google search volumes and search results, DataForSEO, 18 October 2026