Microsoft’s recent Excel updates bring four practical changes this December that aim to speed data work, tighten collaboration, and — unsurprisingly — fold AI more deeply into everyday spreadsheet tasks. The headline items are expanded
Agent Mode with live web search and optional Anthropic Claude routing, a modernized
Get Data dialog for Windows Power Query,
tagged comment previews for protected workbooks that appear directly in email notifications, and a refreshed
Liquid Glass-style interface for Excel on iPhone, iPad, and Apple Vision Pro. These changes are rolling out in stages and carry licensing, tenant, and device prerequisites you need to understand before flipping switches.
Background / Overview
Excel has been moving from a feature-first spreadsheet into an
agentic workspace where AI assistants plan, execute, and iterate multi-step tasks for you. That strategy now shows up in production features rather than just labs: Agent Mode chains multiple steps inside a workbook, Power Query’s entry point is being modernized for discoverability, collaboration flows are being streamlined for protected files, and mobile apps are being aligned with the latest Apple UI conventions. These updates are being released across preview channels (Insiders, Frontier) first and are gating access via Microsoft 365 licensing and tenant admin controls.
As of November 26, 2025, public reporting and Microsoft guidance indicate that availability is staged: some features are Insider-only or Frontier preview, some are rolling out to Windows Current/Preview channels, and some mobile UI changes remain Insider-limited. Administrators and power users should plan for staggered adoption and confirm tenant-level settings before wider deployment.
Agent Mode: now with web search and a Claude option
What changed
Agent Mode — Excel’s Copilot-driven workflow engine that plans and executes multi-step spreadsheet tasks — now supports two practical additions:
- Web search inside Agent Mode. Agents can pull live facts (market numbers, statistics, published figures) directly from the web and cite those sources while building tables, charts, or models. This removes many manual copy/paste steps and improves the speed of fact‑gathering tasks.
- Model choice: Anthropic’s Claude routing. By default Agent Mode uses OpenAI-based models, but enterprises and Insiders now have the option to route Agent Mode requests to Anthropic’s Claude (where the tenant and admin policy permit). That provides a different reasoning and explanation style and can be valuable for finance-heavy use cases that prefer Claude’s transparency and trace traces-of-thought presentation. Anthropic’s recent Opus 4.5 release further emphasizes spreadsheet/agent strengths, which explains why vendors are enabling model choice inside productivity apps.
Who can use it (prerequisites)
- A Copilot-eligible license on the Microsoft account is required to access Agent Mode features. Microsoft’s Copilot guidance covers which Microsoft 365 plans and account types get Copilot features; some Copilot experiences are included with Microsoft 365 Personal/Family/Premium for personal accounts, while enterprise-grade Copilot features often require paid Copilot licenses or tenant enablement. Check your subscription and tenant settings before assuming access.
- Insider or Frontier program enrollment is required for early access. Agent Mode web-search and Claude routing are being distributed through preview channels (Frontier program and Insiders) before wider GA. Desktop parity and Mac support are explicitly being phased in after initial Windows/web availability.
- Admin controls. For organizations, an admin must enable model-choice and other Copilot-related provider options in the Microsoft 365 Admin Center (for example, the “AI Providers For Other Large Language Models” control) before Claude routing is available for tenant users.
Why it matters — strengths
- Faster, grounded research: Agents that can cite web sources reduce the friction of assembling background data for financial models, market snapshots, or research tables. The live search capability is most useful when you need current facts (prices, published statistics) without switching contexts.
- Model diversity: Providing Anthropic’s Claude as an alternative to default OpenAI models lets teams pick an LLM whose output style and risk/accuracy trade-offs best suits a task (e.g., finance explanations versus broad summarization). Recent independent coverage of Claude’s Opus 4.5 demonstrates the model’s improved spreadsheet and agent performance — another reason enterprises may want a choice.
- Auditable, step-based outputs: Agent Mode surfaces the agent’s plan and edits as explicit steps, which helps auditing and reduces the “black box” feel for regulated workflows. This is a significant UX improvement vs. inserting a single opaque suggestion.
Risks and mitigations
- Data exfiltration risk. Any feature that routes workbook data to cloud models creates the risk of sensitive data leaving controlled environments. Enterprises must apply DLP, prompt filtering, and tenant-level policies to limit what data agents can send. Treat agent outputs as assistive and require a human sign-off for critical financial or regulatory deliverables.
- Hallucination and provenance. Even with web search, agents can misinterpret context or stitch together facts incorrectly. Always verify key numbers and maintain supporting source links for audit trails. Prefer ops where agents populate candidate worksheets that a human validates, rather than auto-committing to production reports.
- Licensing and quotas. Copilot features (agents, web calls, model choice) are subject to license limits and AI credits depending on the subscription level. Organizations should plan quota policies or purchase sufficient Copilot capacity for heavy agent workloads. Microsoft publishes AI credits and limits for Personal/Family/Premium plans and enterprise Copilot details for administrators.
Get Data dialog: a cleaner entry point to Power Query
What changed
Excel for Windows now exposes a
modern Get Data dialog to start Power Query flows. The dialog is a dedicated window with a search bar, categorized connector lists, and a Home tab for popular sources — replacing the old drop-down menu and making connectors easier to discover. Microsoft pairs the new dialog with OneLake catalog integration for organizations using Fabric/OneLake metadata. The feature is generally rolling out to Windows users in Current/Insider channels depending on your update ring.
How to access it (Windows)
- Open Excel and go to the Data tab on the ribbon.
- Click the Get Data dropdown and then select Get Data (Preview) or Get Data (depending on build).
- Use the dialog’s Home tab for popular connectors, the New tab to browse all sources, or the search bar to find a specific connector.
After selecting a source you continue into the familiar Power Query editor to shape, transform, and load data.
Why it matters — strengths
- Lowered friction for casual users. The search-first dialog reduces the cognitive load for people who only occasionally import data, and nudges users toward refreshable Power Query flows rather than ad-hoc copy/paste.
- Improved discovery and governance. By making connectors and OneLake catalog artifacts discoverable inside the dialog, organizations can promote sanctioned data assets and reduce shadow-data practices.
- Consistent Power Query experience. After you pick a source the flow still lands you in Power Query with the same shaping and M-language capabilities — nothing about your ETL steps changes, only the entry point does.
Caveats
- OneLake catalog integration is staged. The OneLake catalog option is still being tested in preview rings and may not appear for all tenants until later. If your organization expects OneLake integration, ensure the tenant Fabric configuration and Insider channel support are present.
- Windows-first. The modern Get Data dialog and its initial feature set are targeted at Excel for Windows. Mac and web parity timelines are not yet complete; expect Windows users to see it first.
Comment previews for protected files: faster triage, small governance trade-offs
What changed
If someone @mentions you in a comment on a
protected workbook, the email notification can now include the full comment text (or a readable preview) so you can triage without unlocking the file first. The email still links back to the workbook and a
Go to comment action for follow-up, but the preview speeds triage and reduces context switching. This behavior is available across Excel for Windows, web, and iOS where the update is active.
Why it matters
- Faster review cycles. Teams that review locked reports or protected templates will spend less time jumping through unlock steps just to read feedback. That accelerates minor editorial or QA tasks.
- Better mobile triage. Email previews let managers and reviewers decide whether a message requires immediate attention without opening a protected workbook on a shared device.
Risks and controls
- Preview surface vs. data exposure. Putting protected-file comments into email bodies is a design trade-off: email is more widely accessible than a protected workbook. Organizations worried about sensitive metadata or comments should treat this setting as configurable and update email DLP rules and retention policies accordingly.
- Partial visibility caveat. In some cases, the email preview will only include the full comment if it was made using specific Excel clients (for example, comments authored in Excel for Windows, iOS, or web). If your organization mixes clients, test the experience to avoid surprises.
Excel on iOS: Liquid Glass styling and bottom search (Insider)
What changed
On devices running
iOS 26 (and matching iPadOS/macOS Vision), Excel Insider builds (Version 2.102, Build 25101016 for Insiders) have adopted Apple’s
Liquid Glass translucent styling and moved the search entry to the bottom of the screen. The redesigned home experience aligns all Office mobile apps visually with the platform update and improves one‑hand usability on phones and Vision Pro. Template filtering and bottom navigation enhancements are included in the preview for Insiders.
Practical impact
- Cosmetic and ergonomic. This is primarily a UX update — it won’t change core workbook behaviors or formula logic, but it makes the app easier to navigate one-handed and modernizes the look to match iOS 26.
- Insider-only for now. If you depend on a stable corporate mobile toolchain, don’t force Insiders into production devices; wait for broader releases.
Administration, licensing, and deployment checklist
Start small, validate often, and put policy-first controls in place. Practical rollout steps:
- Inventory licensing: confirm which accounts have Copilot-eligible licenses (Personal/Family/Premium vs. enterprise Copilot licenses) and check AI credits and agent quotas for your tenant.
- Pilot Agent Mode with a limited group: pick a finance or operations team and run a 30-day pilot to collect validation processes and failure modes. Ensure Agent Mode edits are logged and saved to audit trails.
- Configure admin controls: enable or restrict external model providers (e.g., Anthropic) at the tenant level as needed, and set DLP rules to block sensitive fields from being sent to cloud LLMs.
- Test Get Data and connector flows: validate OneLake / Fabric artifacts discovery and Power Query refresh credentials in a staging tenant before promoting to production.
- Train end users: document how Agent Mode presents steps and require human verification on high-risk outputs. Provide quick guides for how to access the new Get Data dialog and the locations of template filters in the mobile app.
Practical tips and recommended workflows
- Use Agent Mode for repetitive, templated tasks (monthly close reports, standardized dashboards), not for ad-hoc single-use computations. Keep a “human validation” step as mandatory in your sign-off process.
- When pulling web-based facts, keep the original link in a provenance column; build a simple Power Query or formula that records the source URL next to the pulled data. That practice speeds audits and reduces “where-did-this-number-come-from” questions.
- Limit Claude or other third-party model routing to workbooks that don’t contain regulated data unless you have a documented data protection justification and the necessary contractual controls in place. Anthropic’s recent Opus announcement shows Claude’s spreadsheet strengths but also underscores general agent safety trade-offs — vendor-specific protections do not remove the need for enterprise governance.
- Use the modern Get Data dialog’s search bar to find connectors rather than inventing one-off import macros. Standardizing on refreshable queries reduces error and increases reproducibility.
Critical analysis — strengths and potential risks
Notable strengths
- Productivity gains are real. Agent Mode’s multi-step automation and Copilot-driven formula generation shorten common workflows that previously required deep Power Query or macro knowledge. The modern Get Data dialog reduces friction for non-expert users.
- Model choice is a pragmatic move. Allowing organizations to route prompts to different LLM backends acknowledges that no single model fits all enterprise needs. It’s a vendor-neutral approach to behavior and safety trade-off management.
- Incremental, auditable automation. Agent Mode’s step-by-step output helps maintain audit trails and makes AI actions inspectable rather than opaque—a crucial design choice for enterprise adoption.
Major risks
- Data governance complexity. The combination of cloud-based models, live web search, and connectors to organizational data creates a multi-vector attack surface for accidental exfiltration. Enforcement of DLP, policy-based model access, and tenant restrictions is mandatory for regulated organizations.
- Operational unpredictability. Agents can accelerate mistakes as fast as they accelerate work. Mistaken formulas, incorrect assumptions, or poorly scoped web searches can propagate bad numbers quickly unless the organization enforces verification steps.
- Staged parity across platforms. Windows-first rollouts (Get Data dialog) and Insiders-only mobile UI updates mean Mac and web users may not have feature parity immediately. That can complicate cross-platform teams.
Final verdict and recommendations
These four updates are practical and incremental rather than revolutionary. Agent Mode’s web search and model choice materially improve the usefulness of AI in real-world spreadsheet scenarios; the modern Get Data dialog cleans a long-standing UX pain point; email previews of comments speed collaboration; and the iOS interface refresh keeps Microsoft Office visually current on Apple platforms. But real benefit depends on disciplined governance: licensing clarity, DLP policies, tenant admin controls, stepwise pilots, and mandatory human validation for high-risk outputs.
For IT and Excel power users ready to adopt:
- Start with a limited business pilot for Agent Mode, with mandatory auditing and rollback plans.
- Confirm Copilot license entitlements and AI credits for your subscription type before relying on Agents for production workloads.
- Encourage use of refreshable Power Query flows via the new Get Data dialog; it reduces errors and improves repeatability.
- Treat email comment previews as a convenience feature and review email DLP settings accordingly.
These changes show Microsoft’s continued push to make Excel not just a calculation tool but a
coauthoring environment where humans and models collaborate. The productivity upside is large; the governance challenge is non-trivial. Adopt with care, instrument everything, and insist on human checks where stakes are high.
Source: How-To Geek
4 new Excel features to try in December 2025