Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchAI, startups & growth insights. No spam.
TL;DR: Microsoft announced Wave 3 of Microsoft 365 Copilot at NVIDIA GTC 2026, introducing multi-model support that includes Anthropic's Claude alongside next-generation OpenAI models. Simultaneously, Microsoft is launching Agent 365, a new AI agent automation framework priced at $15 per user per month, and a new M365 E7 Frontier Suite at $99 per user per month that bundles all AI agents and premium features. Both SKUs go generally available on May 1, 2026. This is the most significant pivot in Microsoft's AI strategy since Copilot launched — from a single-model product to a multi-model orchestration platform.
Wave 3 of Microsoft 365 Copilot is not an incremental update. It represents a fundamental architectural shift in how Microsoft thinks about AI inside its productivity suite. Where Wave 1 introduced Copilot as a GPT-4-powered assistant embedded in Word, Excel, and Teams, and Wave 2 expanded those capabilities with autonomous agent workflows, Wave 3 opens the platform to models it does not own.
The headline feature is model diversity. Microsoft is introducing what it calls "expanded model diversity" — the ability for Copilot to route tasks to the most appropriate model based on task type, cost, latency, and capability requirements. Claude from Anthropic is the first third-party model confirmed for this framework. Next-generation OpenAI models will also be part of the expanded stack.
This is not a cosmetic change. It reflects a recognition that no single model is best at everything. Long-form document analysis, code generation, reasoning tasks, and real-time chat assistance have different performance profiles across different models. Giving Copilot the ability to dynamically select the right model for the right task is a meaningful capability upgrade for enterprise users who push the product beyond simple Q&A.
Microsoft announced these changes at NVIDIA GTC 2026, which is an interesting venue choice. GTC is primarily a hardware and infrastructure conference. Announcing Wave 3 there signals that the changes are as much about compute infrastructure as they are about product features — multi-model routing requires significant backend investment in model serving, load balancing, and inference optimization.
Adding Anthropic's Claude to the Microsoft 365 Copilot stack is a significant business development move for both companies, and the implications go beyond any individual product feature.
For Microsoft, Claude fills a specific gap. Anthropic has built a reputation for models that perform well on long-context tasks, nuanced instruction following, and applications where safety and reliability matter. Enterprise workflows involving legal document review, compliance analysis, financial modeling, and technical documentation are areas where Claude's architecture has shown consistent strengths. Adding Claude to Copilot's model routing layer gives Microsoft a credible answer to enterprise buyers who ask whether Copilot can handle complex, high-stakes tasks where errors are expensive.
For Anthropic, Microsoft distribution is transformative. Microsoft 365 has approximately 400 million licensed users globally. Even if only a fraction of enterprise Copilot users interact with Claude-routed tasks, the exposure and inference volume dwarfs what Anthropic can reach through its own direct channels. The partnership also validates Anthropic's positioning as an enterprise AI provider, not just a safety-focused research lab.
The relationship is also not as paradoxical as it might appear on the surface. Microsoft has a major investment in OpenAI, but that investment does not prevent Microsoft from sourcing models from other providers. The Azure AI Foundry already hosts models from dozens of providers. Adding Claude to Copilot's routing layer is consistent with Microsoft's cloud platform strategy: offer the best available tools from the market, regardless of source, and capture value through the platform layer.
The practical question for end users is how model selection will surface in the product. Early indications suggest that most users will not be manually choosing between models. The routing will be largely automatic, optimized by Microsoft's infrastructure based on task characteristics and user tier. Power users and administrators may have more direct control through admin console settings.
Agent 365 is the most commercially significant announcement in the Wave 3 package. It is a new AI agent automation framework built natively into Microsoft 365, priced at $15 per user per month, and going generally available on May 1, 2026.
The framework allows organizations to build and deploy AI agents that automate workflows across M365 applications. These are not simple rule-based automations. Agent 365 agents can understand context, make decisions, trigger actions across multiple applications, and handle exceptions that would trip up traditional automation tools. The integration spans Teams, Outlook, SharePoint, OneDrive, Dynamics 365, and the broader Power Platform ecosystem.
Several agent types are shipping at launch. Scheduling agents can manage complex calendar coordination across teams, handling the back-and-forth that currently eats hours of administrative time. Document processing agents can review, summarize, route, and archive documents based on content analysis rather than metadata alone. Meeting agents can attend calls, generate structured summaries, extract action items, and assign follow-up tasks to the right people. Customer communication agents can draft, review, and in some configurations send responses to standard customer inquiries, with human review steps configurable based on organization policy.
The $15 price point is deliberately positioned below the existing M365 Copilot SKU, which runs at $30 per user per month. That creates a tiered adoption path. Organizations can start with Agent 365 for process automation use cases, demonstrate ROI, and then upgrade users who need the broader Copilot assistant experience. From Microsoft's revenue perspective, this is a land-and-expand model designed to increase AI attach rate across the M365 installed base.
Agent 365 also integrates with Copilot Studio, Microsoft's low-code agent builder. Organizations that have already built custom agents in Copilot Studio can connect them into the Agent 365 framework, enabling more complex multi-agent workflows where custom-built agents collaborate with the standard Agent 365 agents.
The M365 E7 Frontier Suite is a new enterprise SKU launching May 1, 2026, at $99 per user per month. It is the most expensive Microsoft 365 offering ever released.
E7 bundles everything. The base M365 E3 or E5 productivity licenses, the full Copilot experience with multi-model access including Claude, the Agent 365 framework, advanced security features from Microsoft Defender and Purview, and a set of AI-specific capabilities not available at lower tiers. The exact capability differentiation between E7 and lower tiers has not been fully disclosed, but Microsoft has indicated that E7 will include priority model access, higher usage quotas, access to frontier models before general availability, and enhanced governance and auditing tools specifically designed for regulated industries.
The $99 price point requires some context. For organizations currently paying for M365 E5 (approximately $57 per user per month) plus M365 Copilot ($30 per user per month), the effective blended rate for a fully licensed user is already close to $87. E7 at $99 represents a modest premium for a cleaner, all-inclusive package that eliminates the complexity of managing multiple add-on licenses.
For organizations currently on E3 ($36 per user per month) without Copilot, E7 is a much larger jump. Microsoft's bet is that the ROI case for the full AI-enabled suite is strong enough to justify the increase. The company has been building the ROI narrative for months, publishing case studies showing time savings, productivity improvements, and cost reduction from Copilot adoption.
The E7 tier also serves a signaling function. It creates a visible "AI-complete" benchmark for enterprise buyers, a defined product tier that represents full commitment to the AI-augmented workplace Microsoft is selling. For procurement teams evaluating AI strategy, having a single SKU that covers the complete stack simplifies the purchasing decision and Microsoft's enterprise sales process.
Copilot Cowork is a feature concept within the Wave 3 roadmap that is currently in limited research preview, with broader access planned for the Frontier program in late March 2026. It represents Microsoft's most ambitious vision for what Copilot can eventually become.
The core idea behind Copilot Cowork is persistent agent context. Rather than Copilot operating as a reactive assistant that answers questions when asked, Cowork envisions an agent that maintains ongoing awareness of a user's work context, proactively surfaces relevant information, manages long-running tasks across multiple work sessions, and coordinates with other agents working on related problems elsewhere in the organization.
The "Cowork" framing is deliberate. It positions the AI not as a tool but as a collaborator. Microsoft has been using language like "digital coworker" and "AI teammate" with increasing frequency across its product communications. This language mirrors what OpenAI is saying about Frontier and what Anthropic used when introducing Claude Cowork earlier in the year. The industry is converging on a framing that positions AI agents as organizational participants, not just software features.
The limited research preview means Microsoft is being cautious about releasing Cowork broadly. Persistent agent context creates new data governance questions. If an agent has ongoing awareness of your work, what data is it retaining? How is that stored? Who can access it? These are not resolved questions, and Microsoft's measured rollout suggests the team is working through the privacy and compliance implications before pushing it to the full M365 customer base.
The Frontier program access in late March will give enterprise customers an early look at Cowork in a more structured setting. Frontier program members typically include Microsoft's largest and most sophisticated enterprise customers, who can provide the complex, real-world feedback needed to harden the feature before general availability.
Microsoft's decision to move from Copilot-only to multi-model orchestration reflects three converging pressures that have been building since Copilot launched in 2023.
The first is performance differentiation. No single model leads on all benchmarks. As enterprises push AI into more specialized, high-value workflows, the limitations of any single model become apparent. Legal teams working with Copilot on contract analysis have different model requirements than developers using Copilot for code generation. Routing to the optimal model per task type is a direct response to enterprise feedback that Copilot delivers inconsistent results across different use cases.
The second is competitive defense. Google Workspace AI is tightly integrated with Gemini, giving Google customers a seamless multi-modal experience. Anthropic's Claude has significant enterprise adoption, particularly in legal, financial services, and life sciences. OpenAI's ChatGPT Enterprise is actively sold into organizations that also use Microsoft 365. By making Copilot model-agnostic, Microsoft removes a reason for enterprise customers to evaluate competing platforms. Whatever model a customer wants, they can get it through Copilot.
The third is the Azure AI Foundry business. Microsoft already hosts over 1,800 models in Azure AI Foundry. Every model call routed through Copilot's multi-model layer is an Azure inference workload. Multi-model Copilot is also a mechanism for driving Azure consumption. The more diverse the model portfolio in Copilot, the more inference revenue flows through Azure.
The timing — announced at NVIDIA GTC — also suggests Microsoft is signaling a deeper relationship with the AI infrastructure layer. NVIDIA's GPU platforms power a large portion of the world's model inference. Microsoft's Azure infrastructure is one of NVIDIA's largest cloud customers. Co-announcing at GTC reinforces the partnership and the narrative that Microsoft is building the infrastructure layer for the multi-model enterprise AI era.
Microsoft's Wave 3 announcements shift the competitive dynamics of the enterprise productivity AI market in ways that Google and Salesforce will need to respond to.
Google Workspace AI has been the most credible competitor to M365 Copilot for the past eighteen months. Google's advantage has been its native multi-modal capabilities — Gemini can process images, documents, audio, and video across Google Workspace applications in ways that Copilot historically could not match. Google has also been aggressive on pricing, bundling Gemini capabilities into existing Workspace plans at no additional cost for certain tiers.
Wave 3's multi-model architecture partially answers Google's multi-modal advantage. If Copilot can route to the most capable model for a given task type, the gap between the two platforms narrows. The critical question is execution. Multi-model routing sounds elegant in announcements but is difficult to implement reliably at scale. Users who experience inconsistent behavior — different results from the same request on different days because different models handled it — will develop trust problems with the product.
Slack AI, Salesforce's AI offering embedded in the workplace communication platform, is a more targeted competitive concern. Slack's strength is in asynchronous communication workflows, where its AI features summarize channel activity, surface relevant threads, and draft messages. Agent 365's communication and scheduling agents are a direct competitive challenge to these features. Microsoft Teams already competes with Slack for enterprise communication market share. Agent 365 gives Microsoft another weapon in that fight.
The deeper competitive question is whether enterprise buyers will standardize on one AI platform or run multiple. Most large organizations are currently running AI experiments across multiple vendors simultaneously. Microsoft is betting that Wave 3, with its multi-model support and competitive pricing, gives them a reason to consolidate. Google and Slack are betting on the opposite: that enterprise AI spending grows large enough for multiple platforms to coexist.
| Platform | Company | AI pricing | Model strategy | Key enterprise strength |
|---|---|---|---|---|
| M365 Copilot Wave 3 | Microsoft | $30/user + Agent 365 at $15 | Multi-model (Claude, OpenAI) | Installed base, Teams integration |
| Workspace AI | Bundled in select tiers | Gemini-native | Seamless Google Cloud integration | |
| Slack AI | Salesforce | Add-on to Slack plans | OpenAI-powered | Async communication workflows |
| Claude Cowork | Anthropic | Not yet public | Claude-native | Safety reputation, long context |
The pricing structure Microsoft has built around Wave 3 is strategically layered in ways that deserve careful analysis before procurement decisions are made.
At $15 per user per month, Agent 365 is priced to be a relatively easy approval in enterprise budgets. For a 1,000-person organization, Agent 365 runs $180,000 annually — a meaningful number, but within the range that department heads and IT leaders can often approve without board-level sign-off. Microsoft has designed the product and the price to minimize friction in the initial adoption decision, with the expectation that demonstrable ROI drives expansion and eventual upgrade to higher tiers.
The $30 per user per month M365 Copilot license remains the core AI productivity offering. With Wave 3's multi-model support, the value proposition for that $30 improves materially. Organizations that previously saw Copilot as "GPT-4 in Office" now see it as "the best available model for any task in Office." That is a meaningfully better product at the same price.
At $99 per user per month, E7 targets the segment of enterprise customers for whom AI is a strategic priority, not just a productivity experiment. This group is willing to pay for comprehensive access, premium support, and the assurance that they will not be left behind on model access as the AI landscape evolves. The all-inclusive nature of E7 also matters for procurement teams managing budget complexity — one line item instead of multiple add-on SKUs.
The structural question for CFOs is total cost of ownership. E7 at $99 looks expensive until you add up E5 ($57) plus Copilot ($30) plus any additional security or compliance add-ons. The bundled math often favors E7 for organizations that would have purchased those components anyway. Organizations that do not need the security or compliance features of E5 and E7 have less incentive to upgrade, and the $99 price will simply appear large in budget discussions.
For organizations currently on E3 evaluating an AI upgrade, the decision tree involves three realistic options: add Agent 365 at $15 for automation-focused use cases, move to Copilot at $30 for the full assistant experience, or commit to E7 at $99 for the complete platform. Microsoft's sales team will be very active in steering customers toward E7, where the margins and contract values are highest.
Wave 3 creates several categories of work for IT administrators and enterprise technology teams.
The most immediate impact is governance complexity. Multi-model Copilot means that data processed by Copilot queries may flow through different model providers depending on routing decisions. Organizations in regulated industries — financial services, healthcare, government — need to understand what models their data is touching and where inference is happening. Microsoft will need to provide clear documentation on data residency and processing for each model in the routing stack, including Claude. IT teams should request this documentation before enabling multi-model features, not after.
Agent 365 deployment at scale requires process design work that many organizations are not yet equipped for. Deploying an agent that handles customer communications or routes financial documents is not just a software installation. It requires workflow mapping, exception handling design, audit trail configuration, and change management for the employees whose work is being automated. Organizations that treat Agent 365 as a self-service tool and deploy it without process design rigor will encounter the same problems that plagued early RPA deployments: high initial enthusiasm, followed by maintenance burden and quality issues.
For enterprise buyers evaluating the E7 Frontier Suite, the May 1 launch date means procurement timelines are running now. Organizations that want E7 on day one need to have budget approvals, legal review of updated licensing terms, and deployment planning underway immediately. Microsoft's enterprise licensing agreements typically involve 90-day notice periods for major SKU changes. The companies best positioned to take advantage of Wave 3 features on day one are the ones treating this as an active procurement project today, not in April.
The multi-model shift also changes vendor relationship dynamics. IT leaders who have built their AI strategy around the assumption that Microsoft means OpenAI now need to account for Anthropic as an indirect technology vendor in their stack. That has implications for vendor risk assessments, data processing agreements, and security reviews. Claude running through Copilot is still Anthropic's model processing your organizational data, even if Microsoft is the contractual counterparty.
"Microsoft's Wave 3 Copilot with multi-model support including Claude represents the clearest signal yet that the enterprise AI platform wars are no longer about which model is best — they are about which platform can orchestrate all models best." -- Microsoft at NVIDIA GTC 2026
Wave 3 is the third major product generation of Microsoft 365 Copilot, announced at NVIDIA GTC 2026. The headline change is expanded model diversity, including Anthropic's Claude and next-generation OpenAI models alongside the existing Copilot architecture. Wave 3 also introduces the Agent 365 framework and the M365 E7 Frontier Suite.
Agent 365 and the M365 E7 Frontier Suite both go generally available on May 1, 2026. Some Wave 3 features, including Copilot Cowork, are in earlier preview stages with broader access expected in late March 2026 for Frontier program participants.
Agent 365 is a new AI agent automation framework built into Microsoft 365. It allows organizations to build and deploy AI agents that automate workflows across Teams, Outlook, SharePoint, Dynamics 365, and the Power Platform. It is priced at $15 per user per month.
Agent 365 is priced at $15 per user per month, launching May 1, 2026. This makes it the lowest-priced dedicated AI add-on Microsoft has offered for M365, positioned below the existing Copilot license at $30 per user per month.
M365 E7 is a new enterprise SKU at $99 per user per month that bundles the full M365 productivity suite, Copilot with multi-model access, Agent 365, advanced security features, and priority access to frontier models. It is designed for organizations that want an all-inclusive AI platform without managing multiple add-on licenses.
Claude from Anthropic performs well on long-context tasks, document analysis, and compliance-sensitive workflows where accuracy and instruction-following matter. Adding Claude to Copilot's model routing layer gives enterprise customers access to a proven alternative model for tasks where it outperforms other options in the stack.
Routing is primarily automatic, optimized by Microsoft's infrastructure based on task characteristics and user tier. Power users and administrators may have more control through admin console settings, but the default experience will be transparent to most end users — Copilot selects the best model without requiring user intervention.
When Copilot routes a task to Claude, Anthropic's model infrastructure processes that request. Microsoft remains the contractual counterparty for enterprise customers, but the data processing implications need to be reviewed against your organization's data governance policies. IT teams should request updated data processing documentation from Microsoft before enabling Claude routing in regulated environments.
Copilot Cowork is a feature concept within Wave 3 focused on persistent agent context. Rather than Copilot operating as a reactive assistant, Cowork envisions an agent with ongoing awareness of a user's work that proactively surfaces information and manages long-running tasks across work sessions. It is currently in limited research preview, with broader Frontier program access expected in late March 2026.
Agent 365 is a dedicated agent automation framework, whereas existing Copilot features are primarily assistant-oriented. Agent 365 agents can run autonomously on defined workflows without continuous user prompting. They can make routing decisions, trigger cross-application actions, and handle exceptions, which goes beyond the interactive assistant model of standard Copilot.
Organizations on M365 E5 ($57/user/month) plus Copilot ($30/user/month) are already spending $87 per user. E7 at $99 represents a $12 premium for a bundled package that also includes Agent 365 and enhanced AI governance features. For organizations with full E5 and Copilot deployments, the incremental cost to E7 is relatively modest.
Google Workspace AI uses Gemini natively with strong multi-modal capabilities across Google's productivity applications. Microsoft's Wave 3 approach is to route tasks to the best available model from a curated pool, including Claude and OpenAI models. Google's advantage is deep native integration; Microsoft's advantage is model choice and the breadth of the M365 installed base.
Wave 3 significantly improves the Copilot value proposition compared to previous generations. The multi-model routing means better task-specific performance, and Agent 365 adds automation capabilities beyond what earlier versions offered. Organizations that deferred Copilot adoption because of capability gaps have stronger reasons to re-evaluate by May 2026.
Agent 365 complements rather than replaces Power Automate. Power Automate handles rule-based workflow automation with structured triggers and actions. Agent 365 adds AI-native agents capable of understanding unstructured context and making judgment-based decisions. Both can be used in the same organization for different automation scenarios, and Agent 365 integrates with the Power Platform ecosystem.
IT admins should take four steps now: review Wave 3 licensing terms and assess which SKU tier fits organizational needs, map existing Copilot usage to identify which workloads would benefit from Agent 365 or multi-model routing, assess data governance requirements for Claude integration in regulated workflows, and engage Microsoft account teams about E7 pricing and migration paths from current licensing agreements.
Microsoft has not announced a free trial program for Agent 365 or E7. Frontier program members may have early access before the May 1 general availability date. Organizations interested in early access should contact Microsoft's enterprise sales team or reach out through their existing M365 licensing agreement.
Microsoft's Frontier program is a pre-release access program for enterprise customers that gives participants early access to upcoming product features. Wave 3 features including the Copilot Cowork preview are expected to reach Frontier program members in late March 2026, several weeks before general availability on May 1.
Microsoft unveils its premium E7 tier with Claude integration and announces Agent 365 GA for May 1, signaling a multi-model enterprise AI strategy.
Microsoft unveiled the Frontier Suite (Microsoft 365 E7) on March 9, 2026 — a $99/user enterprise AI bundle bundling Copilot Wave 3, Agent 365, and Copilot Cowork. Here's what it means for the future of enterprise software.
Tesla unveils Digital Optimus — a software-based AI agent built with xAI that observes and replicates human computer interactions to automate complex office workflows, targeting September 2026 rollout.