Microsoft just made its most significant Copilot update in two years — and it does not just add features. It changes the fundamental architecture of how enterprise AI works.
On March 30, 2026, Microsoft launched three major additions to Microsoft 365 Copilot: Critique, a dual-model research system where one AI writes and another fact-checks; Council, a parallel system where multiple models answer the same question simultaneously; and Copilot Cowork, an autonomous agent that handles long-running, multi-step tasks across the entire Microsoft 365 environment. All three pull from both OpenAI and Anthropic's Claude — a direct signal that Microsoft is no longer betting exclusively on a single AI partner. The result is the most enterprise-ready version of Copilot yet, though the cost premium for the most powerful features is steep and general availability is still weeks away.
What You Need to Know
Critique and Council are live now inside Microsoft's Frontier early access program for M365 Copilot subscribers. Copilot Cowork — the headline feature — is also in Frontier preview and reaches full general availability on May 1, 2026, as part of the new $99/user/month Microsoft 365 E7 bundle.
- Switch to Critique if your team runs competitive analysis, regulatory research, or any document that needs verifiable sourcing. The accuracy gain is measurable.
- Use Council if you need to compare perspectives on a complex strategic question — it surfaces disagreement rather than hiding it.
- Evaluate E7 if your organization is already paying for E5 ($60/user/month) plus a Copilot add-on ($30/user/month). The combined $87/month cost is only $12 short of E7's full bundle.
- Stay on your current plan if your primary Copilot use is drafting emails, summarizing documents, and quick Q&A. The multi-model features add cost without adding value for those workflows.
What Launched: Critique, Council, and Cowork
Critique: One AI Writes, Another Reviews
Critique splits research tasks into two stages. In the first stage, one model handles planning, retrieval, and drafting. In the second stage, a different model reviews the output before it reaches the user. Specifically, Critique generates responses using OpenAI's model, then fact-checks them with Claude.
The review process focuses on three elements: source reliability, report completeness, and citation accuracy. Key claims must be supported by verifiable sources before the final report is produced.
The accuracy improvement is concrete. Microsoft says the approach enabled the research agent to score 13.8% higher on the DRACO benchmark — an industry standard for deep research quality — compared to the previous top-ranked system, Perplexity Deep Research. Microsoft ran these evaluations itself using the same DRACO protocol published in the original benchmark paper, with GPT-5.2 as the judge model.
Council: Multiple Models, One Comparison Layer
Council goes further: two models each produce a full report on the same query, and a third AI compares them to show where they agree, where they differ, and what unique insights each brings.
Council runs multiple models in parallel. A separate judge model compares the outputs, surfaces consensus and disagreement, and makes AI reasoning more transparent. For compliance, legal review, and strategy work — where users need to know what each model noticed and what each model missed — this matters more than a single polished answer.
Council compares ChatGPT and Claude responses side by side. Users select it from the model picker inside the Researcher agent.
Copilot Cowork: The Autonomous Agent
Copilot Cowork is designed for long-running, multi-step work in Microsoft 365. Users describe the desired outcome, and Cowork creates a plan, reasons across their tools and files, and carries work forward with visible progress and opportunities to steer.
When a user hands off a task to Cowork, it turns the request into a plan that continues in the background, with clear checkpoints to confirm progress, make changes, or pause execution at any time. Cowork checks in if it needs clarification.
Practical examples include: preparing for a client meeting by gathering relevant emails, files, and Teams messages, then producing a briefing document and deck; cleaning up an Outlook calendar by identifying conflicts, suggesting reschedules, and implementing changes after approval; or conducting research across internal and external sources and consolidating the output into a structured report.
Critically, Cowork integrates technology from Claude Cowork, Anthropic's digital desktop assistant — while Microsoft's version operates in the cloud, inside Microsoft 365's security and governance boundaries.
Feature and Cost Comparison Table
| Feature | What It Does | Models Used | Cost vs. Single Model | Availability |
|---|---|---|---|---|
| Standard Researcher | Single-model deep research | OpenAI (default) | Baseline | All M365 Copilot |
| Critique | Generate + independent review before delivery | OpenAI writes, Claude reviews | ~20% more | Frontier program now |
| Council | Two models answer in parallel, judge model compares | OpenAI + Claude + judge model | ~2.5× more | Frontier program now |
| Copilot Cowork | Autonomous multi-step task execution across M365 | Claude + Microsoft models | Included in Copilot / E7 | Frontier now; GA May 1 |
Cost premiums for Critique and Council per Axios reporting, April 2026. Pricing as of April 2026.
Pricing and Packaging
This update arrives alongside a significant restructuring of Microsoft's enterprise licensing.
| Plan | Price (per user/month) | Key Inclusions | Best For |
|---|---|---|---|
| Copilot Business | $18–$21 (promo through June 30, 2026) | Cowork, model choice (Claude + GPT), Wave 3 features | SMBs under 300 users |
| M365 Copilot add-on | $30 (on top of E3/E5) | Full Copilot integration in M365 apps | Enterprise add-on buyers |
| M365 E7 (Frontier Suite) | $99 | E5 + Copilot + Cowork + Agent 365 + enhanced security | Large enterprises |
| Agent 365 (standalone) | $15 | AI governance control plane | Agent management without full E7 |
Pricing as of April 2026. E7 and Agent 365 reach general availability May 1, 2026.
For organizations already paying for E5 ($60/user/month) plus the Copilot add-on ($30/user/month), the combined $87/month is only $12 short of E7's full bundle price of $99/user/month. For teams that need Cowork and Agent 365 governance, upgrading is a straightforward financial case.
What Changed Since Wave 2
Wave 3 of Microsoft 365 Copilot represents a genuine architectural shift, not a feature refresh. In its March 9, 2026 Microsoft 365 blog post, Microsoft said Wave 3 brings "multi-model intelligence" into everyday workflows and emphasized that Copilot is now model-diverse by design.
Before this update, Copilot Researcher used a single AI model to plan, search, write, and review. Every step — generation and evaluation — happened inside the same model, with no independent check.
The new architecture separates those functions. Critique assigns generation to one model and evaluation to another. Council runs both functions simultaneously across competing models. For research-heavy work, a single polished answer is often less useful than a structured comparison of several plausible ones.
The Cowork launch also marks the first time Microsoft has embedded a major Anthropic product — built on the same technology as Claude Cowork — directly into its enterprise platform.
The Adoption Problem Behind the Features
Here is the non-obvious story underneath this launch: as of January 2026, Microsoft reported 15 million paid Copilot seats — roughly 3.3% of its 450 million commercial Microsoft 365 users. That number came after eight consecutive quarters with no public disclosure of paid adoption. Microsoft's stock dropped 7% after the January earnings call.
Critique and Council appear to be a direct response. Features like Critique and Council appear designed to accelerate adoption by addressing the most common objection to AI-assisted research: trustworthiness. Enterprises are not avoiding Copilot because they do not understand it. They are avoiding it because they cannot verify it.
Notably, Microsoft executive VP Charles Lamanna told Axios: "It's becoming very clear to us that there will be many models. Come summertime there will be many more models than just these two inside of Copilot." That framing positions Microsoft not as an OpenAI proxy, but as a neutral model aggregator — a much stronger enterprise value proposition.
Who Should Care
Enterprise research and knowledge teams get the most immediate value. If your team produces competitive intelligence reports, regulatory summaries, or due diligence documents, Critique removes the most expensive part of that workflow: the manual accuracy review.
Switch to Critique if you run any research workflow where a wrong citation or unsupported claim creates downstream liability. The 13.8% DRACO benchmark improvement translates to fewer hallucinations reaching the final report.
Use Council if your work involves genuinely contested questions — strategy, policy, legal analysis — where you need to see where the evidence is ambiguous, not just get a confident answer. Council turns model disagreement into a user-facing tool rather than hiding it behind a single output.
Evaluate Cowork if you are already on E5 and spending $87/month per user on E5 plus Copilot separately. The $12/month gap to E7 includes Cowork and Agent 365 governance. For teams running repeatable workflows — monthly budget reviews, weekly briefings, regular calendar management — the math closes quickly.
Stay on your current plan if your Copilot usage is primarily chat-based — drafting, summarizing, and answering questions. Council costs 2.5 times more per query than a single model. For conversational use, that cost premium delivers no return.
This does not affect you immediately if you are not in the Frontier program. Critique and Council are not yet broadly available; general Copilot users will need to wait for broader rollout beyond Frontier.
What to Watch Next
Microsoft is also building more homegrown models, and Lamanna said those models might show up first working in conjunction with outside models in ensemble experiences rather than as standalone replacements. Expect Microsoft's own MAI-1 model family to appear inside Critique and Council workflows by late 2026. The May 1 GA launch of E7 and Agent 365 is the next concrete milestone. Watch whether Cowork's autonomous task execution holds up at enterprise scale — early adopter Capital Group has reported positive results, but broad adoption will surface the real edge cases.
Conclusion
Microsoft's Wave 3 Copilot update is the most structurally significant change to the product since its launch. Critique and Council solve the core enterprise objection — AI that cannot be verified — by making model disagreement visible and building independent review into every research output. Copilot Cowork shifts Copilot from a prompt-response tool into an autonomous workflow executor, built on Anthropic's Claude technology and running inside Microsoft 365's security boundaries. The upcoming E7 bundle on May 1 makes the financial case clearer for organizations already close to that price point. If your team does research-heavy or document-heavy work inside Microsoft 365, now is the time to join the Frontier program and test Critique before the broader rollout.
