Copilot Studio is no longer "just" a low-code chatbot builder. At Build 2026 (June 2–3, 2026 in San Francisco) and in the simultaneously released May update, Microsoft has decisively rebuilt the platform into a governed enterprise agent platform — for building, orchestrating, testing and operating AI agents across Microsoft 365, Power Platform, desktops and web interfaces.

This article summarises the most important announcements and contextualises them for Swiss organisations.

Computer-Using Agents (CUA) — Now Generally Available

The biggest new feature: Computer Use in Copilot Studio has been generally available since 13 May 2026 — and is being rolled out to all commercial geographies.

What Computer-Using Agents can do: an agent interacts with a web interface or desktop application the way a human would — it sees the screen, identifies UI elements, clicks, types and navigates. No API, no pre-configured connectors required. The agent uses vision models to interpret the user interface in real time.

Concrete use cases:

• Automating legacy web applications without API access (e.g. older ERP interfaces, cantonal government portals)
• Data transfer between systems that have no connector
• Form filling in external portals based on internal SharePoint data
• Quality checks in web-based document management systems

An important governance note: CUA agents act in the context of the configuring user or service account. All actions are fully auditable in Microsoft Purview. For tenants with elevated compliance requirements (FINMA, Swiss FADP) we recommend evaluating CUA agents in a sandbox environment first and defining a tight scope of action.

Microsoft IQ — The New Intelligence Layer

Microsoft introduced the Microsoft IQ architecture at Build 2026 as a unified intelligence and context layer — and it is immediately available for GitHub Copilot, Microsoft Foundry and Copilot Studio.

Microsoft IQ consists of four layers:

The core difference from the previous approach: Microsoft Graph + connectors give you data; Microsoft IQ gives you intelligence. Instead of orchestrating raw data queries, IQ queries return contextualised, permission-checked answers — with organisational relationships, intent recognition and provenance.

For Copilot Studio agents this means: agents can answer questions about internal structures, responsibilities and relationships without developers writing complex Graph queries. Work IQ knows who is responsible for a project, what was discussed last week and which documents are relevant — in real time and permission-compliant.

Multi-Agent Orchestration — Several GA Milestones

Those who have been following developments in recent months know: multi-agent is the main focus of Copilot Studio 2026. Several components reached General Availability in April and May 2026:

GA since April 2026:

• Microsoft Fabric integration: Agents can directly access Fabric data (Lakehouse, Warehouse, Dataflows) and run AI actions on enterprise data.
• Microsoft 365 Agents SDK orchestration: Agents built with the M365 Agents SDK can be orchestrated seamlessly from Copilot Studio — regardless of whether they were built low-code or pro-code.
• Agent-to-Agent (A2A) communication: Agents can address other agents, delegate tasks and return results. This enables genuine multi-agent workflows where an orchestrating agent coordinates specialised sub-agents.

GA since May 2026:

• Extended Agent Builder in M365 Copilot: Agents configured in Microsoft 365 Copilot can now be customised more deeply in Copilot Studio — more knowledge sources, more actions, more control.
• High-value AI actions in automation: New pre-built AI actions for common automation scenarios (document analysis, email classification, content generation) directly in the workflow interface.

New Workflows Experience

Microsoft has fundamentally redesigned the workflow interface in Copilot Studio. The new experience aligns more closely with modern process design tools and replaces the classic Power Automate flow canvas inside Copilot Studio with an agent-centric view.

What has changed:

• Workflows are now modelled as agent actions, no longer as isolated flows
• Conditional logic can be described in natural language ("If the customer submits an urgent request, escalate to Level 2 support")
• Integration of loops and iterations over data sets directly in the agent context
• Better debugging: step-by-step trace of agent decisions in the test interface

Real-Time Voice Experiences

Copilot Studio now supports Real-Time Voice for agent experiences. This means businesses can deploy Copilot Studio agents as voice interfaces — for customer service hotlines, internal helpdesks or interactive voice portals.

The architecture is based on Azure Communication Services and Microsoft Teams Phone. Voice agents can transcribe conversations in real time, retrieve relevant data from enterprise systems and read out responses. Handoffs to human agents are supported seamlessly.

For Swiss organisations active in customer service or internal support, this opens a new domain: voice requests that previously could only be handled by human agents can now be partially or fully automated through Copilot Studio agents.

Agent Governance — Real-Time Security Assessment

Microsoft has significantly expanded the governance capabilities in Copilot Studio. New is Real-Time Security & Risk Assessment for agents:

• Continuous security assessment of agent actions based on DLP policies and sensitivity labels
• Automatic warning or blocking when an agent accesses classified content or contacts external services not explicitly allowed
• Agent Activity Dashboard in the Power Platform Admin Center: overview of all active agents in the tenant, their activity patterns and potential risks
• Improved support for Conditional Access policies specifically for agent identities

This is particularly relevant for Swiss organisations in regulated industries: FINMA-regulated financial institutions, organisations subject to the Swiss FADP and ISO 27001-certified businesses can now operate agents in a controlled framework that meets their own compliance requirements.

What This Means for Swiss Organisations

CUA for Legacy Systems — Worth Evaluating Now

Computer-Using Agents are the first feature that represents a genuine breakthrough for Swiss SMEs and mid-market companies: they enable automation where no API exists. Many Swiss organisations work with legacy ERPs, cantonal web portals or industry-specific systems that have no connectors. CUA addresses precisely that gap.

Microsoft IQ: Knowledge Queries Without Development Overhead

Work IQ in Copilot Studio significantly reduces the implementation effort for "intelligent" agents. Previously, developers had to write complex Graph API queries to retrieve contextual information. With Work IQ, these queries are abstracted away — the agent "understands" organisational context natively.

Multi-Agent GA: Production-Ready, Not Just Preview

A2A communication and M365 SDK orchestration are now GA — no longer preview. That means organisations planning multi-agent architectures can now build with support and SLAs, not just at their own risk. The preview limitations of recent months have been lifted.

Licensing: Copilot Studio Usage Remains Message-Based

The fundamental licensing logic has not changed: Copilot Studio is licensed via message packs (25,000 messages per month in the base package). CUA actions and voice interactions are billed through dedicated capacity units that can be purchased separately. For accurate current terms, we recommend checking directly with Microsoft or a partner — pricing adjustments following GA releases are possible.

Our Assessment

Copilot Studio demonstrated at Build 2026 that Microsoft is serious about the platform and developing it consistently. The combination of CUA (GA), Microsoft IQ, multi-agent orchestration (GA) and Real-Time Voice makes Copilot Studio one of the most capable low-code agent platforms available today.

Three recommendations for Swiss organisations:

1. 1Governance first: Before agents go into production, set up the Agent Activity Dashboard in the Admin Center and define DLP policies for agent actions.
2. 2CUA pilots for legacy systems: Identify two or three processes that were previously not automatable due to missing APIs — and evaluate CUA specifically for those.
3. 3Test Work IQ: Replace an existing, complex search-dialogue agent with a Work IQ-based approach and measure whether answer quality improves.

CNEXT accompanies Swiss organisations in evaluating, building and governing Copilot Studio agents — from use-case selection through to productive rollout.