Excel in 2026: Copilot Agent Mode, Dynamic Arrays, and Why Spreadsheets Still Win
Every year someone declares spreadsheets dead. Every year, 1.1 billion people open Excel. In 2026, Excel has an AI copilot that writes formulas, builds pivot tables, and generates charts from natural language. Here is what actually changed.
DevForge Team
AI Development Educators
The Spreadsheet Death Predictions Keep Not Coming True
Every few years, a new technology is declared the spreadsheet killer. Databases, business intelligence platforms, no-code tools, and now AI. Yet every year, Excel's active user count grows. There are an estimated 1.1 billion Excel users worldwide — more than any programming language, any database platform, and most mobile apps.
The reason is straightforward: nothing else combines zero-setup analysis, universal literacy, immediate visual feedback, and incremental complexity in the same package. You can open Excel with no training and do something useful. You can also spend years mastering it and still find new capabilities. That range is rare.
In 2026, Excel is not dying. It is getting an AI copilot that changes how you interact with it — not whether you use it.
What Agent Mode Actually Does
Copilot Agent Mode became generally available in Excel for Microsoft 365 in January 2026. It is the most significant Excel update in years, and it is genuinely different from the Copilot Chat features that came before.
Chat-based Copilot answers questions about your data. "What is the average sales amount by region?" → Copilot tells you. You still build the formula or pivot table yourself.
Agent Mode plans and executes. Tell it: "Build a sales analysis by region with a bar chart highlighting the top 5 performers." Agent Mode:
- Creates a pivot table grouping sales by region
- Inserts a bar chart from the pivot table
- Applies conditional formatting to highlight the top 5 bars
- Shows you each step with an explanation of what it did
You can edit any step's output, ask Copilot to revise a specific step, or approve the result and continue. It behaves like a junior analyst who shows their work — capable and fast, but requiring your review before it goes into a report.
The "shows its work" part matters. When Copilot builds a pivot table, you can click through the settings and see exactly what it configured. When it writes a formula, you can read and verify it. This is deliberate design — Agent Mode is not meant to be a black box.
For complex analysis, Copilot can enter a "Think Deeper" mode that uses reasoning models to create structured analysis plans, writing and executing Python in the background when needed for statistical operations that Excel formulas cannot perform natively.
Dynamic Arrays Changed Everything (And Most People Missed It)
Microsoft rolled out dynamic array functions starting with Excel 2019, with the full suite landing in Microsoft 365. These functions changed the fundamental behavior of how Excel formulas work, but the change happened gradually enough that millions of users missed it.
Before dynamic arrays, returning multiple values from a formula required Ctrl+Shift+Enter "array formulas" — a counterintuitive keyboard shortcut that created formulas with mysterious curly braces. Building a list of unique values required a pivot table or a complex helper-column approach.
After dynamic arrays, formulas can naturally return arrays that spill into adjacent cells:
=UNIQUE(A2:A100)This single formula returns every unique value in the column, automatically expanding to as many rows as needed. Add new data, the spill range expands. Delete data, it contracts.
The functions that ship with dynamic arrays deserve careful attention:
- FILTER — returns all rows matching criteria (replaces manual filtering + copy-paste)
- UNIQUE — extracts distinct values without pivot tables
- SORT / SORTBY — sort dynamically without touching source data
- XLOOKUP — the modern replacement for VLOOKUP, with built-in error handling
LET and LAMBDA bring programming concepts into the formula bar. LET assigns variables within a formula, making complex calculations readable:
=LET(
rate, TaxRate,
subtotal, SUM(OrderItems[Price]),
subtotal * (1 + rate)
)LAMBDA creates reusable custom functions without any VBA or macro knowledge. Define a calculation with named parameters, store it in the Name Manager, and call it anywhere in the workbook like a built-in function.
The gap between "Excel user" and "Excel developer" is narrowing. The cognitive distance between writing a formula and writing a function is now one LAMBDA.
The Honest Copilot Assessment
AI features in Excel deserve an honest assessment rather than hype or dismissal. After real-world use, here is where things stand in early 2026.
What works well. Formula generation from natural language descriptions is genuinely useful, especially for functions you use infrequently. "Write a formula that calculates the number of business days between the order date and today, excluding federal holidays" → Copilot produces a correct NETWORKDAYS formula with the right argument structure. For analysts who know what they want but do not remember exact syntax, this is a significant time saver.
Basic analysis questions against clean, structured data work reliably. "Which product had the highest margin last quarter?" against a properly structured sales table produces correct answers quickly.
Chart creation, pivot table setup, and basic formatting are the tasks where Agent Mode pays off most clearly. Mechanical tasks that take 15 minutes of clicking produce results in 30 seconds of describing.
What is mediocre. Complex financial models with custom logic, multiple assumptions layers, and intricate formula dependencies do not improve significantly with Copilot. The formulas Copilot generates are correct but not always efficient — it does not always find the most elegant solution.
Workbooks with inconsistent structure, merged cells, and mixed data types frustrate Copilot as much as they frustrate human analysts. Clean data is a prerequisite for good AI output.
What does not work. Replacing understanding. If you cannot evaluate whether a pivot table is configured correctly, you cannot catch Copilot's mistakes. AI-generated Excel output requires a competent reviewer. An analyst who has never built a pivot table manually will trust Copilot output they cannot evaluate, leading to dashboards that look correct but answer the wrong questions.
The honest summary: Copilot is an accelerator for people who already know Excel. It turns 30-minute tasks into 5-minute tasks. It does not turn no-knowledge users into competent analysts.
This is why foundational Excel training — formulas, lookups, pivot tables — matters more in the AI era, not less. You need the foundation to use the tools effectively.
Why Spreadsheets Still Win
Five properties make spreadsheets irreplaceable, and no current alternative has matched all five simultaneously.
Zero-setup analysis. No database to configure. No deployment. No permissions system. Open a file, start analyzing. The barrier to initial productivity is the lowest of any analytics tool.
Universal literacy. Everyone in most organizations already knows the basics of spreadsheets. A pivot table built in Excel can be shared with any executive, any investor, any client — they can open it and interact with it. The same analysis in a BI tool requires training or a managed link.
Immediate feedback loop. Change a number and see the result update instantly. This tight feedback loop makes spreadsheets ideal for modeling "what if" scenarios. No deploy step, no refresh button, no waiting for a query to run.
Incremental complexity. You can start with SUM() and be productive. Graduate to XLOOKUP when you need it. Add Power Query when your data cleanup takes too long. Move to Power Pivot when your analysis spans multiple tables. The tool scales with your skill level — you never have to switch to something else.
Persistent local analysis. A spreadsheet you build today works in five years without any platform changes. The file is self-contained. This long-term reliability matters for financial models, business cases, and analytical templates that get reused over years.
No other tool currently offers all five. Google Sheets comes closest but loses on desktop performance and advanced features. Power BI is more powerful for visualization but less flexible for formula-based analysis. Python + Pandas is more powerful for transformation but requires programming knowledge and a development environment.
Excel sits in the center of the Venn diagram between accessible and capable.
The DevForge Approach
The Excel Mastery pillar on DevForge Academy starts from foundations and builds toward AI-augmented analysis. The sequence is deliberate:
- Formula Architecture — understand how Excel thinks before using AI to write formulas for you
- Lookup & Reference Mastery — the modern lookup toolkit: XLOOKUP, FILTER, and when to use each
- Data Modeling — structure your data so formulas and AI tools work correctly
- Pivot Tables — the fastest path from data to insight
- Power Query — automate data transformation without writing code
- Dashboards & Copilot — combine everything with AI assistance
The goal is analysts who use Copilot as an accelerator — fast at the mechanical parts, deliberate and critical about the analytical decisions. That is the combination that produces reliable business intelligence.
Spreadsheets are not going anywhere. Learn to use them well, and AI makes you dramatically more productive. Skip the fundamentals, and AI gives you faster ways to produce wrong answers.