A woman reviews an AI-powered workflow dashboard with automation, analytics, and security features.
At Microsoft Ignite 2026 in San Francisco, Microsoft's case for Copilot Studio has grown. It no longer stops at "build a chatbot that answers HR questions." Microsoft now pitches it as a place to design and run whole business processes, combining agents, deterministic workflows and apps in one environment. Microsoft also wants IT to see the same governance and cost controls everywhere.

Jason Moore, Microsoft's vice president of product for Copilot Studio, announced the changes in a November 18 post. Some features are generally available now, some are in public preview, and one of the most important, dedicated agent accounts, reaches public preview in December. That mix of release states matters more than the marketing. Below we sort what you can use today from what is still a trial, and flag a shared-memory caveat that IT should read before any agent gets a mailbox.

A woman reviews an AI-powered workflow dashboard with automation, analytics, and security features. What's new, sorted by release status​

CapabilityStatus per Microsoft
Customer conversation harness (voice and chat, live-agent handoff with context)Generally available
Apps in Copilot Studio and Copilot CoworkGenerally available, rolling out in November 2026
Agent sharing among multiple makersGenerally available
Classify node using a decision-scoring model by defaultShipped (no preview label given)
Agent and workflow creation from natural language in Copilot Cowork or Copilot CodePublic preview
Copilot Studio command-line tools and Copilot Studio MCPPublic preview
Dedicated agent accountsPublic preview starting December 2026

Summary: Most of the team-building and app features can be used now. The persistent "agent as teammate" story still depends on preview features.

The Microsoft Copilot harness and harder processes​

Microsoft built the announcement around the Microsoft Copilot harness, the orchestration layer that gives agents reasoning, planning and multi-step execution. Moore's post says the harness works alongside hooks, computer use and Model Context Protocol (MCP) integration to give agents more reach into business systems and more control over how they act. The post doesn't define what hooks do in practice, so builders will need the documentation before they plan around them.

Moore's post says only that internal evaluations show "substantial gains." Jeff Teper's companion Dynamics 365 Ignite post gives numbers for agents on the new harness compared with the standard harness:

  • Knowledge quality improved by more than 20 points
  • Code-analysis quality more than doubled
  • The file-analysis pass rate went above 90%
  • Multitool performance rose from 87.0% to 98.9%
  • Input tokens per task fell 45%

Microsoft's footnote says these figures count only successful evaluation runs and may vary by day, build, model and configuration. Treat them as directional rather than as a guarantee for your workload. Still, a 45% drop in input tokens matters if you pay per credit.

The customer stories, read carefully​

Microsoft gives two examples:

  • Asurion: Its finance team had trouble with exception-heavy invoices, the ones that break traditional automation. After adding a Copilot Studio agent to its finance automation process, Asurion raised touchless invoice processing from under 20% to over 70%.
  • Microsoft's own support team: A "dynamic resolution agent" that uses reasoning, knowledge and tools resolved more than 35% of cases with no human involvement in August 2026.

Both results are reported by Microsoft and not independently verified. The Asurion figure comes with no timeframe or invoice volume. The support figure doesn't say which case types were counted or whether the 35% covers all cases or a subset. A 35% autonomous rate also means people still handled most of those cases. That fits Microsoft's own framing: the agent handles part of the process, not all of it.

Apps, workflows and building in plain language​

The process-design argument rests on three parts. Agents make judgment calls, workflows handle predictable automation, and apps give people a structured place to enter information or approve things.

  • Apps in Copilot Studio and Copilot Cowork are now generally available. Teper's post adds that app creation runs on the Copilot Managed Runtime, which handles identity and lifecycle management inside your tenant.
  • Workflows: The classify node now uses a decision-scoring model by default. Microsoft says this makes classification faster and more accurate for jobs like email triage and routing. Makers also get clearer visibility into credit usage, though Microsoft hasn't described the display.
  • Building in plain language: Employees can describe an agent or workflow in Copilot Cowork or Copilot Code (public preview). Makers can then develop it further in Copilot Studio. Microsoft's governance point is that these artifacts live in Copilot Studio, so IT can oversee them like any other organizational solution and doesn't end up with shadow automations scattered around the tenant.

Summary: The trend is toward more people building, with one place to govern what they build.

Agent accounts: from tool to teammate​

This is the most consequential part of the announcement. It's also the one that most needs caveats.

Microsoft says dedicated agent accounts give Copilot Studio agents managed credentials and access to approved resources. Memory and signal-based triggers then let an agent carry work forward over time. In the intended experience, people @mention the agent, hand it a task, see what actions it took, and rely on it around the clock.

Moore's example is a finance team's weekly budget review. Every Tuesday, the agent gathers planning records, compares forecasts with explanations, and asks budget owners through Teams or Outlook to explain unexpected changes. At the next review, it remembers which questions are still open. This is an illustrative scenario, not a documented deployment.

This isn't brand new​

Agent accounts have been in Microsoft's early-access Frontier program since October. Microsoft announced a set of Copilot Studio releases on October 15 that turn agents into persistent participants in shared business processes. An agent can now get its own organizational account with approved credentials and scoped permissions, so people can invoke it in Teams and Outlook, @mention it in Word, Excel, and PowerPoint comments, and hand off work by email. The December date marks a wider public preview, not first availability.

Microsoft Learn's page "Give your agent an account (Frontier)" says an account gives an agent a mailbox, appears in Microsoft Teams and your directory, and acts with its own identity. In addition, Copilot Studio also creates a Microsoft Entra Agent ID that admins manage separately, for each agent that has an account.

Setup as currently documented (Frontier)​

The Learn documentation is prerelease and subject to change. As of now, it describes this sequence:

  1. Admin, Microsoft 365 admin center: activate the Copilot Studio agent template in the Microsoft 365 admin center, choosing which users or groups may create account-based agents.
  2. Admin, Power Platform admin center: Under Copilot Studio, select Agent accounts and turn on the setting for your environment.
  3. Maker: Open Copilot Studio Preview, create the agent and select Agent account. A summary of the guidance by consultancy 2toLead notes that the account must be added before the agent is published for the first time.
  4. Publish: Your agent's account isn't active until you publish the agent.

If the Agent account option is missing: first verify that you're using Copilot Studio Preview and that your environment uses the early release cycle. Ask your admin to turn on agent accounts, include you in the users or groups selected during template activation, and ensure a valid license exists for your account. Teams presence also has a licensing dependency. The agent still gets its account, mailbox, and directory presence, but it appears in Microsoft Teams only if your tenant separately licenses Teams.

On licensing during the Frontier phase, Beri.net reports that each published agent uses a Microsoft 365 Frontier for AI Teammates license, and usage consumes Copilot Credits. The FutureWork blog adds that makers need a maker license with Agent 365 (Microsoft 365 E5 + Agent 365 add-on, or Microsoft 365 E7). Microsoft hasn't said what licensing will look like for the December public preview or at general availability, so budget cautiously.

The shared-memory problem​

Microsoft Learn's agent-accounts FAQ deserves a careful read. Account-based agents use shared memory across everyone they work with, even in one-to-one chats. The agent's People card warns users that files or chats they share can be summarized into answers for any user, including people without access to the originals, regardless of sensitivity label. The FAQ also says:

  • There is no privacy boundary between users. The agent is only "guided" not to repeat one person's details to another.
  • Anyone who interacts with the agent can view or delete its saved memories.
  • In this preview, end users can't opt out, and admins can't turn off shared memory without turning off agent accounts entirely.
  • Shared memories are deleted automatically after 28 days with no interaction.
  • Some data is excluded: credentials, ID and card numbers, health information, protected characteristics, and judgments about anyone's performance.

In practice, an employee could paste a draft reorganization memo into a chat with the team's budget agent, and a colleague without access to that memo might later get an answer shaped by it. That isn't a bug. It's how Microsoft documents the feature. Developer Holger Imbery suggests treating each agent account like a service account, with a job description, scoped permissions, a named human owner, clear memory limits, a budget, and a plan for retiring it. That's sound advice, and it costs nothing.

Summary: Agent accounts are the most promising part of this announcement and the riskiest to govern. Pilot them with low-sensitivity processes first.

Team development and governance​

Copilot Studio is also moving toward team development:

  • Agent sharing (GA): Several makers can work on the same agent directly in Copilot Studio.
  • Code-first developers can build and edit agents locally, with artifacts stored in GitHub repositories.
  • Copilot Studio CLI and Copilot Studio MCP (public preview) give developers, and the AI assistants they work with, access to the harness, tools and models. However an agent is built, whether in Cowork with natural language or with the CLI in Visual Studio Code, the process owner monitors and refines it in Copilot Studio.

On the governance side, Microsoft lists publishing and reuse with access controls, built-in evaluations, monitoring and telemetry, and organization-level cost policies. Management is centralized across the Microsoft 365 admin center and the Power Platform admin center. Teper's post gives specifics for Agent 365 Cost Management in the Microsoft 365 admin center: spending policies, limits, alerts and departmental chargeback across Cowork, Copilot Studio, Copilot Managed Runtime and Dynamics 365 agents. Spend forecasting arrives in Frontier this month, and value insights are due in December.

What we think​

Copilot Studio is no longer pitched as a low-code chatbot tool. It's a process platform where agents, workflows and apps share one place to govern them. That makes sense. The main reason agent projects stall is that one person's clever agent can't be audited, budgeted or handed over when that person leaves.

The counterpoint is that the "teammate" features, persistent identity, memory and triggers, are still previews. They carry documented data-exposure trade-offs and licensing that is still changing. The impressive numbers come from Microsoft's own customer stories and internal evaluations.

For IT teams, the practical plan:

  1. Use the GA features now: agent sharing, apps and the customer conversation harness. Review credit visibility and cost policies before usage grows.
  2. Pilot the previews in a separate early-release environment. Don't use your production default environment.
  3. Before enabling agent accounts, keep the Frontier template inactive until you have a list of approved makers. Scope each agent's tool permissions, and tell users plainly that anything they share with the agent may be shared with the whole team.
  4. Measure your own results. Asurion's jump from 20% to 70% is encouraging, but your exception-heavy process deserves its own baseline.

Agents are becoming coworkers with accounts, mailboxes and memories. Like any new hire, they need a manager, an access review, and a clear rule about which secrets they keep.

 

References

  1. Build and manage business processes with Microsoft Copilot Studio - Microsoft Microsoft 2026-11-18T16:00:00+00:00
  2. Copilot Studio Agent Accounts vs Service Accounts: What IT Leaders Need to Know 2tolead.com
  3. Copilot Studio Agent Accounts Share Memory Past File Permissions beri.net