The last three months of product updates have pushed business tools further toward an AI-first workflow: Excel’s Agent Mode makes multi‑step spreadsheet work feel like directing an assistant; Power Query’s web editor brings desktop+cloud data shaping together; Power BI’s new Input Slicer fixes a long-standing pain point for text filtering; and third‑party creativity and collaboration apps — from Canva to ChatGPT and Zoom — are closing the loop between idea, branded output and meeting action. This feature roundup pulls together the practical changes that matter to accountants, analysts and IT teams, explains the trade‑offs, and highlights governance, accuracy and deployment risks you need to plan for now. erview
Microsoft and several cloud vendors used the January–February 2026 release windows to accelerate the integration of generative AI into day‑to‑day work. Rather than merely adding isolated chat panes or image generators, vendors are shipping agentic tools — features that plan, execute and sometimes iterate multiple steps without a human writing each formula or macro. That shift shortens time to insight, but it also widens the attack surface for errors, data leakage and governance lapses. This article summarises the major functional changes, verifies specs with vendor documentation and independent reporting, and flags where IT and finance teams must act.
Power Query Online gives teams a visual, cloud‑first data‑prep environment with collaboration and scheduled refreshes that Desktop users can’t match out of the box. The addition of cluster values and fuzzy grouping reduces the manual workload for de‑duplicating and canonicalising lists (for example, turning “fRAnce”, “Frnce” and “France” into a single canonical value with an attached similarity score). The Diagram view and Schema view make it easier for non‑technical readers to understand step sequences and table schemas.
Source: ICAEW.com New tech features for Dec-Feb 2026: Excel, Teams, Google Sheets, Chat GPT, M365, Power BI, Canva, Zoom ++
Microsoft and several cloud vendors used the January–February 2026 release windows to accelerate the integration of generative AI into day‑to‑day work. Rather than merely adding isolated chat panes or image generators, vendors are shipping agentic tools — features that plan, execute and sometimes iterate multiple steps without a human writing each formula or macro. That shift shortens time to insight, but it also widens the attack surface for errors, data leakage and governance lapses. This article summarises the major functional changes, verifies specs with vendor documentation and independent reporting, and flags where IT and finance teams must act.
Excel: Agent Mode, data cleaning, and web Power Query
What changed (the headlines)
- Agent Mode landed in Excel as a multi‑step Copilot experience that can clean messy tables, diagnose and fix formula errors, build financial models, and scaffold dashboards from raw data. It’s accessible from Home > Copilot and — depending on tenant and region — is rolling out to desktop and web around January–February 2026.
- Clean Data and Formulas by Example: Copilot subscribers see a dedicated Clean Data workflow, while Formula‑by‑Example suggestions appear in Table editing flows. Some of these features are gated to premium Copilot licensing; Agent Mode however has been extended to a broader set of M365 subscribers (the commercial rollout window varies by region).
- Power Query for Excel Online: the full Power Query editor is now available in Excel for the web and exposes online‑only capabilities — cluster values, fuzzy grouping, diagram & schema views, and a script/formula bar — features that the Desktop UI does not yet surface in the same way. This materially changes how teams collaborate on data preparation.
Why it matters: productivity and new workflows
Agent Mode reframes work that previously required formula literacy (LAMBDA, XLOOKUP, complex nested IFs) as high‑level instructions. Example prompts reported in previews include “clean this table, standardise the country names and create a one‑page dashboard” or “build a 5‑year loan repayment model given principal in USD and show results in EUR.” The product now does planning → formula generation → validation steps inside the workbook, producing native Excel artifacts that are editable afterwards. That’s a major win for teams who cannot keep up with bespoke spreadsheet engineering demands.Power Query Online gives teams a visual, cloud‑first data‑prep environment with collaboration and scheduled refreshes that Desktop users can’t match out of the box. The addition of cluster values and fuzzy grouping reduces the manual workload for de‑duplicating and canonicalising lists (for example, turning “fRAnce”, “Frnce” and “France” into a single canonical value with an attached similarity score). The Diagram view and Schema view make it easier for non‑technical readers to understand step sequences and table schemas.
Practical guidance: how to use Agent Mode safely
- Start in a copy of your workbook. Agent Mode modifies live workbooks and can apply broad changes automatically.
- Use explicit, audit‑friendly prompts — include objectives, date ranges, and an acceptance test (e.g., “Ensure totals reconcile to the reported GL total in cell X”).
- Review the formulas and intermediate steps Copilot inserts; do not accept results blindly. Agent Mode can be wrong or produce sub‑optimal aggregation choices.
Risks and operational controls
- Accuracy risk: Agents sometimes create visually plausible but incorrect calculations (for instance, formula‑generated charts labelled “Series 1/2” or mismatched currency conversions). Always validate Copilot outputs against known controls or reconciliation points.
- Data governance: Agents can access workbook content and may make online calls to fetch internet data depending on Copilot settings. Ensure tenant Copilot policies, DLP and conditional access controls are set before widespread adoption.
- Regional availability & licensing: Some Copilot features roll out by channel and country; EU/UK customers have reported staggered availability. IT teams must track channel/version numbers (e.g., Current Channel 2602, Semi‑Annual Channel 2512) and Copilot license entitlements.
Power Query Online: why the web editor matters
New online-only features
Power Query in Excel Online exposes features not yet surfaced in Desktop UI, including:- Cluster values (add column → cluster) for fuzzy canonicalisation with similarity scores.
- Fuzzy grouping and improved Group By options for near‑match aggregates.
- Diagram view for visualising query dependencies, and Schema view for quick column management.
- Script view that exposes the entire M query script in a formula bar for advanced edits.
Governance implications
- Cloud‑first query editing increases the importance of workspace permissions, refresh credentials and tenant‑level data loss prevention. Make sure refresh credentials are stored in service‑managed secure stores and that people with edit access understand the difference between local and cloud refresh semantics.
- Query folding and performance characteristics can differ; long, cloud‑side transformations may incur refresh throttling or cost implications depending on connectors used.
Power BI: Input Slicer and Copilot ergonomics
What’s new
- The Input Slicer (previously “text slicer” in preview) is now generally available. It lets report viewers type or paste free‑form text to filter reports using operators such as Contains any, Contains all, Starts with and Is any. Creators can control single vs multiple entries and formatting. This closes a long‑standing usability gap when users only know part of a key or need fuzzy matching of text fields.
- Power BI also increased the Copilot prompt character limit from 500 characters to 10,000 across Copilo response to user needs for longer, more expressive prompts when dealing with semantic models and multi-step requests.
Why it matters
The Input Slicer makes report consumption more flexible for line‑of‑ten copy/paste lists of IDs or partial names. It’s especially useful for ad‑hoc investigations where the exact taxonomy isn’t known in advance. The Copilot prompt expansion reduces the friction for authors who craft longer instructions or paste in context blocks (for example, semantic model descriptions).Implementation notes
- When embedding reports or creating paginated reports, check whether the Input Slicer’s behaviour (especially “Contains any/all”) aligns with your report’s expected filter logic.
- The Input Slicer only accepts text columns currently; numeric or date use cases still require alternative filtering controls.
PowerPoint, Word, Teams: small changes with tangible gains
Presenter notes and editing in PowerPoint
PowerPoint’s Semi‑Annual Channel introduced the ability to edit notes in Presenter View, a pragmatic quality‑of‑life improvement for presenters who make last‑minute adjustments while speaking. PowerPoint also moved Find & Replace into a sidebar for less intrusive searches. These are workflow wins that reduce context switching during rehearsals and live presentations.Word: Agent Mode and Ink-to-Shape
Word now supports Agent Mode for batch document edits (for example, “make every header bold and centred”), and the Draw tab offers Ink to Shape conversions. That reduces manual formatting overhead in long documents and supports hybrid authors who mix handwritten notes with polished text.Teams: pane pop‑outs, drafts, and meeting UX
Microsoft Teams rolled out a raft of user features: pop‑out panes, a Drafts view for unsent messages, resizable meeting gallery when content is shared, and improved file discovery under Shared. Several AI features — custom meeting summaries and live interpreter functionality — remain gated to Teams Premium / Copilot tiers. For organisations, that means reconsidering whether meeting transcription, multilingual interpretation and tailored Copilot summaries are mission‑critical enough to justify premium seats.ChatGPT, Canva and the creative loop: brand safety meets speed
Canva inside ChatGPT and other AI assistants
Canva’s Brand Kit is now accessible directly inside ChatGPT (and previously Claude), enabling the creation of on‑brand, editable visuals from a single conversation. That means you can give a plain‑language brief to ChatGPT and receive layered, production‑ready Canva files that respect logos, color palettes and locked templates. The integration reduces the design handoff friction between marketing and AI.What this changes for teams
- Faster content creation cycles: idea → branded concept → editable file without manual re-branding.
- Lower reliance on designers for routine outputs, which can improve throughput but also increases the need for brand governance controls around brand kit access.
- Integration with multiple AI assistants via the Model Context Protocol (MCP) means teams can embed Canva design generation into broader agent workflows.
Risks to manage
- Brand leakage: controlling who can connect a Brand Kit to external AI apps is essential. Use SSO, conditional access and app‑level governance to avoid unauthorized use.
- Edit trail & approvals: generated assets may bypass existing content approvals. Update review workflows and, where necessary, require a designer’s sign‑off for high‑impact materials.
Google Workspace and Zoom: incremental but useful
Google Sheets and Docs
Google added the simple but useful =SHEET() and =SHEETS() functions (returning sheet index and number of sheets, respectively), plus the ability to edit password‑protected Office files in Google editors by requiring passwords on open — parity features that make cross‑platform collaboration safer and less error‑prone. Google Docs also offers an AI audio summary and longer document reading features via Tools. These are practical updates for teams that regularly move files between Office and Google environments.Zoom meeting enhancements
Zoom improved meeting experiences with a dynamic gallery that auto‑crops to faces, host‑pinnable chat messages, and a media add‑on for 1080p/60fps screen share — useful for high‑fidelity e‑learning and demo sessions. Zoom’s AI companion can create tasks from meeting summaries in premium tiers, and the new post‑meeting recording editor can insert polls/quizzes at timestamps for training content. Consider the licensing trade‑offs for add‑ons that unlock high‑frame‑rate sharing or AI summarisation.ChatGPT platform updates: privacy spaces and model currency
The ChatGPT product added a health‑focused sidebar space that isolates medical records and prevents that data from being used to train public models, reflecting a move toward compartmentalised privacy controls for sensitive contexts. ChatGPT also increased attachment capacity in prompts (from 10 to up to 20), and vendors report that GPT‑5.2 variants in ChatGPT are current through August 2025. These changes emphasise both usability (more attachments) and safety/segregation (dedicated clinical spaces), but they also require organisations to define acceptable‑use and retention policies for confidential data inside chat assistants.Security, governance and accuracy — the overriding checklist
When teams bring these features into everyday use, treat them as platform upgrades that change both productivity and risk profiles. Below are recommended actions for IT, finance and compliance teams.Immediate actions (first 30 days)
- Inventory Copilot access: who has Copilot seats, which features are enabled, and which users can run Agent Mode.
- Update DLP policies to cover Copilot and external connectors — include rules for attachments, external lookups and Brand Kit access.
- Require sandbox use: mandate that early Agent Mode workflows run in copies or staging workbooks.
- Train a small con verification checks: reconciliation points, formula review, and audit trails.
Ongoing governance (30–90 days)
- Set up automated alerting for suspicious Copilot or Power Query activity (large data exports, mass transformations, or cross‑tenant connectors).
- Review licensing and premium feature needs — Teams Premium, Copilot seats and Zoom add‑ons — to align spend with measurable productivity gains.
- Create a “trusted templates” library: vetted models/dashboards that Copilot can augment but not replace.
Data accuracy and auditability
- Always surface the Copilot/agent’s reasoning steps and intermediate artifacts as part of an audit trail. If an Agent Mode routine produces a dashboard, record the prompts and the list of formulas inserted. That evidence is essential for financial models and compliance checks.
Strengths: where these updates genuinely move the needle
- Speed of non‑expert productivity: Agent Mode and Formula‑by‑Example significantly lower the barrier for producing models and dashboards. For routine transformations, the time saved is real.
- Collaboration across cloud and desktop: Power Query Online’s visual maps and the Input Slicer in Power BI reduce the friction between analysts and report consumers.
- Design + brand automation: Canva’s embedded Brand Kit in ChatGPT collapses idea → on‑brand creative workflows into a single step, which is a major efficiency gain for marketing teams.
Weaknesses and open questions
- Model hallucination and numeric errors: Agents can present plausible answers that contain arithmetic or logic mistakes. The more automatic the change, the higher the chance of silent error. Human verification is still mandatory.
- Regional availability and licensing complexity: The same feature may be available in Current Channel but not Semi‑Annual, or free for some M365 subhind Copilot for others. Procurement and IT must plan rollouts centrally.
- Brand and data governance gaps: Tools that make it easy to create and publish content (e.g., Canva via ChatGPT) can also increase the risk of leaked brand assets and inadvertent sharing of confidential data if connectors are misconfigured.
Practical recommendations for finance and IT leaders
- Treat Copilot and Agent Mode as an application upgrade, not a feature toggle. Update risk registers, change logs and training plans before enabling broad access.
- Build a verification checklist for AI‑generated financial outputs: reconciliation with source ledgers, formula inspection, and sign‑off by an identified reviewer.
- Lock down Brand Kit connectors and require explicit, auditable approvals to connect corporate Brand Kits to third‑party AI apps.
- Monitor feature rollout channels (Current vs Semi‑Annual vs Insider) and align them to user groups: fast movers can be in Current/Insider; mission‑critical functions should remain on Semi‑Annual until validated.
Closing: what to watch next
The cadence of these changes shows a clear pattern: vendors are moving from prompts-as-features to agents-as-features — tools that can plan and act across multiple parts of the productivity stack. Over the next quarter we should expect:- Wider Copilot coverage inside classic desktop clients (Outlook, Excel, PowerPoint) and clearer admin controls for tenant governance.
- More cross‑platform integrations (Canva → Copilot → ChatGPT), which will accelerate production but require stricter policy controls for brand and data.
- Additional Power Query web features and closer parity with Desktop as online‑first data prep becomes the norm.
Source: ICAEW.com New tech features for Dec-Feb 2026: Excel, Teams, Google Sheets, Chat GPT, M365, Power BI, Canva, Zoom ++
