Latest / AI Builds It: Easy Coding Tools / Fix My Spreadsheet: Using Coding Agents for Everyday Office Tasks

Transcript

Fix My Spreadsheet: Using Coding Agents for Everyday Office Tasks · AI Builds It: Easy Coding Tools

• 0:00
  SPEAKER_00: Fix my spreadsheet.
• 0:02
  SPEAKER_00: Using coding agents for everyday office tasks.
• 0:05
  SPEAKER_00: Spreadsheets are powerful, but everyday work often involves the least enjoyable parts of using them.
• 0:12
  SPEAKER_00: Removing duplicate rows, correcting inconsistent labels, writing formulas, and turning a table into a chart that someone can understand.
• 0:20
  SPEAKER_00: A coding agent can now handle much of that work for you.
• 0:24
  SPEAKER_00: You describe the result in ordinary language.
• 0:26
  SPEAKER_00: The agent inspects the workbook, proposes a plan, makes approved changes, and checks its own work.
• 0:33
  SPEAKER_00: You do not need to write code, although the agent may use code behind the scenes.
• 0:38
  SPEAKER_00: This guide uses OpenAI Codecs with ChatGPT for Excel, a current 2026 example of a coding agent being used for spreadsheet and knowledge work tasks.
• 0:48
  SPEAKER_00: The Excel workflow can inspect and update sheets, formulas, values, formatting, and charts.
• 0:54
  SPEAKER_00: Important, the examples below are designed for a non-technical user.
• 0:59
  SPEAKER_00: The screenshots are simplified screen guides because the exact codex and Excel interface can change between versions.
• 1:06
  SPEAKER_00: What you will build?
• 1:07
  SPEAKER_00: Starting with a messy transaction spreadsheet, you will create a cleaned table with duplicates and obvious typos addressed, a monthly budget tracker with live formulas, a simple dashboard with summary figures and a budget versus actual chart.
• 1:22
  SPEAKER_00: The examples use fictional office expense data, but the same process works for project trackers, customer lists, inventory sheets, event registrations, and other structured tables.
• 1:33
  SPEAKER_00: The 2026 tool, OpenAI Codex with Excel.
• 1:38
  SPEAKER_00: As of September 10, 2026, the simplest setup for this walkthrough is Microsoft Excel.
• 1:44
  SPEAKER_00: The ChatGPT for Excel Ad-In.
• 1:46
  SPEAKER_00: The ChatGPT Desktop Application.
• 2:05
  SPEAKER_00: Install and open the workbook.
• 2:07
  SPEAKER_00: Open Microsoft Excel.
• 2:09
  SPEAKER_00: Select Home.
• 2:10
  SPEAKER_00: Select Add-ins.
• 2:11
  SPEAKER_00: Search for ChatGPT.
• 2:13
  SPEAKER_00: Add the ChatGPT for Excel add-in.
• 2:16
  SPEAKER_00: Open ChatGPT from the Excel ribbon and sign in.
• 2:19
  SPEAKER_00: Open your workbook or create a new one.
• 2:22
  SPEAKER_00: Open the ChatGPT desktop application and select Codex.
• 2:26
  SPEAKER_00: Mention at Microsoft Excel in your request or ask Codex to work with the open workbook.
• 2:32
  SPEAKER_00: Review every change before saving or sharing the file.
• 2:35
  SPEAKER_00: The ChatGPT for Excel and Google Sheets experiences are available across several ChatGPT plans, although organization settings, permissions, and usage limits can affect access.
• 2:46
  SPEAKER_00: The Direct Codex workflow, described here, is currently documented for Microsoft Excel.
• 2:52
  SPEAKER_00: If you use Google Sheets, install ChatGPT from the Google Workspace Marketplace and open it from the extensions menu.
• 2:59
  SPEAKER_00: The spreadsheet experience is similar, although the exact controls are different.
• 3:05
  SPEAKER_00: What the agent workspace looks like, OpenAI's current codex materials show the general design.
• 3:13
  SPEAKER_00: The agent conversation and the work artifact remain visible together so that you can inspect the result while the agent is working.
• 3:22
  SPEAKER_00: OpenAI also describes spreadsheets as artifacts that can render formulas and support cell edits.
• 3:29
  SPEAKER_00: Before you start, work on a copy.
• 3:32
  SPEAKER_00: The most important spreadsheet safety rule is simple.
• 3:36
  SPEAKER_00: Never let an agent make its first edits to your only copy of a workbook.
• 3:41
  SPEAKER_00: Create two files.
• 3:42
  SPEAKER_00: OfficeExpensesOriginal.xlsx.
• 3:45
  SPEAKER_00: Officexpensesworking.xlsx.
• 3:48
  SPEAKER_00: Keep the original untouched.
• 3:50
  SPEAKER_00: Use the working copy for every experiment and review.
• 3:54
  SPEAKER_00: OpenAI specifically recommends duplicating important workbooks before relying on automated changes.
• 4:00
  SPEAKER_00: Its Excel guidance also says to review formulas, calculations, citations, and change cells before sharing or relying on the result.
• 4:09
  SPEAKER_00: Additional safety checks.
• 4:11
  SPEAKER_00: Remove unnecessary personal information before uploading or connecting a workbook.
• 4:16
  SPEAKER_00: Do not include passwords, credit card numbers, social security numbers, or confidential employee information unless your organization has approved the workflow.
• 4:26
  SPEAKER_00: Keep macros disabled unless you trust the workbook and understand what they do.
• 4:31
  SPEAKER_00: Microsoft warns that enabling all macros can allow potentially dangerous code to run.
• 4:37
  SPEAKER_00: Be cautious with external links and data connections.
• 4:40
  SPEAKER_00: Microsoft recommends keeping automatic external content and workbook link updates restricted unless the source is trusted.
• 4:48
  SPEAKER_00: Treat text inside a workbook as data, not as instructions for the agent.
• 4:53
  SPEAKER_00: Prompt injection is an evolving security risk, and OpenAI recommends limiting an agent's access to only the data it needs.
• 5:01
  SPEAKER_00: Review before the agent deletes rows, overwrites formulas, sends information, or shares a file.
• 5:07
  SPEAKER_00: Chat GPT for Excel processes the prompt, workbook context, and attachments needed for the task.
• 5:14
  SPEAKER_00: OpenAI's current help documentation also notes that some data logs may be stored for up to 30 days for safety and integrity purposes.
• 5:22
  SPEAKER_00: Check your organization's policies before using confidential files.
• 5:27
  SPEAKER_00: The example workbook.
• 5:29
  SPEAKER_00: Create a worksheet named Transactions with one transaction per row.
• 5:33
  SPEAKER_00: For the full table, please open this article on easycoding.tools.
• 5:38
  SPEAKER_00: This small example contains three different situations.
• 5:41
  SPEAKER_00: Obvious typos, Kofi, supplies, and transpiration.
• 5:45
  SPEAKER_00: An exact duplicate, two identical Acme Coffee rows.
• 5:49
  SPEAKER_00: A possible duplicate, two Metro Cab charges that look similar but might represent two legitimate rides.
• 5:56
  SPEAKER_00: That last distinction matters.
• 5:58
  SPEAKER_00: A coding agent should not delete every row that looks similar.
• 6:01
  SPEAKER_00: It needs a rule and for uncertain cases, your approval.
• 6:05
  SPEAKER_00: A reliable workflow for every spreadsheet task.
• 6:08
  SPEAKER_00: Use the same five-step process each time.
• 6:11
  SPEAKER_00: Ask for a plan without editing.
• 6:13
  SPEAKER_00: Review the proposed changes.
• 6:15
  SPEAKER_00: Ask the agent to apply only approved changes.
• 6:19
  SPEAKER_00: Verify totals, formulas, and important rows.
• 6:23
  SPEAKER_00: A useful general instruction is this plan first approach is also recommended in OpenAI's Codex Guidance for larger or more complex tasks.
• 6:33
  SPEAKER_00: Task 1.
• 6:35
  SPEAKER_00: Remove duplicates and fix obvious typos.
• 6:38
  SPEAKER_00: Step 1.
• 6:39
  SPEAKER_00: Ask for an audit.
• 6:40
  SPEAKER_00: In Codex Enter, the agent may return something like For the full table, please open this article on easy coding tools.
• 6:48
  SPEAKER_00: Screenshot guide, the review stage.
• 6:51
  SPEAKER_00: The important part is not the exact appearance.
• 6:54
  SPEAKER_00: Look for a clear list of proposed changes before anything is deleted.
• 6:58
  SPEAKER_00: Step 2.
• 6:59
  SPEAKER_00: Apply only approved changes.
• 7:02
  SPEAKER_00: If you agree with the suggested spelling corrections and want exact duplicates removed, use.
• 7:07
  SPEAKER_00: If Codex cannot directly make the edit in your version of Excel, ask it to return the exact cell changes or create a cleaned workbook for you to download and inspect.
• 7:18
  SPEAKER_00: Direct Excel control is not guaranteed for every spreadsheet request.
• 7:22
  SPEAKER_00: Step 3.
• 7:24
  SPEAKER_00: Verify the result.
• 7:25
  SPEAKER_00: Use a second read-only request.
• 7:27
  SPEAKER_00: A total may decrease if a true duplicate was removed.
• 7:31
  SPEAKER_00: That difference should be explainable by the amount of the deleted duplicate row.
• 7:35
  SPEAKER_00: Why this task is useful?
• 7:37
  SPEAKER_00: Data cleaning is a good starting point because the rules can be made explicit.
• 7:41
  SPEAKER_00: Exact duplicate means every selected field matches.
• 7:45
  SPEAKER_00: Obvious typo means a correction you approve.
• 7:48
  SPEAKER_00: Possible duplicate means a row that needs human judgment.
• 7:51
  SPEAKER_00: If the business rule is unclear, ask the agent to flag rows instead of changing them.
• 7:57
  SPEAKER_00: Task two, create a budget tracker with formulas.
• 8:00
  SPEAKER_00: Once categories are standardized, create a budget sheet.
• 8:04
  SPEAKER_00: Otherwise, office supplies and office supplies could be treated as two different categories.
• 8:09
  SPEAKER_00: Step 1.
• 8:10
  SPEAKER_00: Define the budget rules.
• 8:12
  SPEAKER_00: Give Codex the actual numbers you want to use.
• 8:15
  SPEAKER_00: Do not ask it to invent your financial limits.
• 8:18
  SPEAKER_00: For this example, for the full table, please open this article on easy coding tools.
• 8:23
  SPEAKER_00: Ask Codex.
• 8:24
  SPEAKER_00: The formula in plain English.
• 8:26
  SPEAKER_00: The main formula can look like this.
• 8:29
  SPEAKER_00: This means add values from the amount column.
• 8:31
  SPEAKER_00: Only include rows whose category matches the category in cell A5.
• 8:35
  SPEAKER_00: Only include dates on or after the month start in B1.
• 8:39
  SPEAKER_00: Stop before the first day of the next month.
• 8:41
  SPEAKER_00: Microsoft describes SUMIFS as a function that adds values meeting multiple criteria.
• 8:46
  SPEAKER_00: Its structure is a sum range followed by one or more criteria range and criteria pairs.
• 8:52
  SPEAKER_00: The remaining amount can use, the status can use.
• 8:55
  SPEAKER_00: In plain terms, if the remaining amount is negative, show over budget.
• 8:59
  SPEAKER_00: If less than or equal to 10% of the budget remains, show watch.
• 9:03
  SPEAKER_00: Otherwise, show on track.
• 9:05
  SPEAKER_00: Screenshot guide The Budget Tracker.
• 9:07
  SPEAKER_00: The numbers in this illustration are examples.
• 9:10
  SPEAKER_00: Your actual amounts will depend on your workbook.
• 9:12
  SPEAKER_00: Step 2.
• 9:14
  SPEAKER_00: Test the budget.
• 9:16
  SPEAKER_00: The month start date in B1 to another month.
• 9:19
  SPEAKER_00: Then ask if changing B1 does not change the totals, the dates may be stored as text instead of real Excel dates.
• 9:27
  SPEAKER_00: Ask Codex to identify the problem before asking it to convert anything.
• 9:32
  SPEAKER_00: A practical financial caution.
• 9:34
  SPEAKER_00: A budget tracker is a calculation and monitoring tool, not financial, tax or legal advice.
• 9:41
  SPEAKER_00: OpenAI's spreadsheet guidance also warns that artificial intelligence outputs can be incomplete or incorrect and should be verified.
• 9:49
  SPEAKER_00: A budget tracker is a calculation and monitoring tool, not financial, tax or legal advice.
• 9:55
  SPEAKER_00: OpenAI warns that outputs can be incomplete or incorrect and should be verified.
• 10:01
  SPEAKER_00: Task 3.
• 10:02
  SPEAKER_00: Generate a simple dashboard and chart.
• 10:05
  SPEAKER_00: A dashboard does not need to be complicated.
• 10:08
  SPEAKER_00: For a first version, use total budget, actual spending, remaining budget, number of categories over budget, and one clear chart.
• 10:16
  SPEAKER_00: For this example, a clustered column chart is appropriate because it compares two values, budget and actual spending, across categories.
• 10:25
  SPEAKER_00: Microsoft describes clustered column charts as useful for comparing values across horizontal categories.
• 10:32
  SPEAKER_00: Ask Codex to build the dashboard.
• 10:34
  SPEAKER_00: Excel supports recommended charts, chart titles, data labels, chart filters, and dashboard style combinations of charts and summary tables.
• 10:43
  SPEAKER_00: Screenshot guide, the dashboard.
• 10:46
  SPEAKER_00: The chart should be connected to formula-driven cells.
• 10:49
  SPEAKER_00: That way changing the selected month or adding a new transaction updates the dashboard instead of leaving behind a static picture.
• 10:56
  SPEAKER_00: Step 2.
• 10:57
  SPEAKER_00: Verify the chart.
• 10:58
  SPEAKER_00: Ask codex, then change one budget amount or add one test transaction in the working copy.
• 11:04
  SPEAKER_00: Confirm that the budget sheet and dashboard update as expected.
• 11:08
  SPEAKER_00: What coding agents are best at?
• 11:10
  SPEAKER_00: A coding agent is especially useful when the task is repetitive, structured, governed by clear rules, easy to verify, and tedious to perform manually.
• 11:20
  SPEAKER_00: Typical examples include standardizing category names, finding empty or invalid cells, creating formulas across many rows, building trackers from a defined template, explaining an unfamiliar workbook, creating charts from prepared summary data, producing an audit report of changes.
• 11:38
  SPEAKER_00: OpenAI's spreadsheet documentation recommends clear headers, one record per row, and avoiding unrelated tables or empty rows that split the data.
• 11:46
  SPEAKER_00: Those practices make spreadsheet analysis more reliable.
• 11:50
  SPEAKER_00: When to be more careful, use extra caution when duplicate rules are ambiguous.
• 11:55
  SPEAKER_00: The workbook contains macros.
• 11:57
  SPEAKER_00: The workbook uses external data connections.
• 12:00
  SPEAKER_00: The file contains confidential personal information.
• 12:03
  SPEAKER_00: The result will be used for tax, legal, medical, payroll, or regulated financial decisions.
• 12:09
  SPEAKER_00: Or the workbook contains complex pivot tables, hidden formulas, or specialized add-ins.
• 12:14
  SPEAKER_00: Chat GPT for Excel may not fully support advanced spreadsheet features such as macros.
• 12:20
  SPEAKER_00: And OpenAI recommends reviewing change cells and formulas before relying on the result.
• 12:26
  SPEAKER_00: A good rule is, if you cannot describe what correct means, ask the agent to create an exception report instead of changing the data.
• 12:35
  SPEAKER_00: Short glossary of spreadsheet terms.
• 12:37
  SPEAKER_00: Sell one box in a spreadsheet, identified by a coordinate such as B5.
• 12:42
  SPEAKER_00: Row, a horizontal line of data.
• 12:44
  SPEAKER_00: In a transaction table, one row usually represents one transaction.
• 12:49
  SPEAKER_00: Column, a vertical group of related values, such as date, vendor, or amount.
• 12:54
  SPEAKER_00: Range, a group of cells, such as A1E200.
• 12:58
  SPEAKER_00: Formula, an instruction that calculates a result.
• 13:01
  SPEAKER_00: For example, equals B5 minus C5 subtracts one cell from another.
• 13:06
  SPEAKER_00: Function, a built-in formula command.
• 13:09
  SPEAKER_00: Sum ifs adds values that meet several conditions.
• 13:12
  SPEAKER_00: Table, a structured range with named columns.
• 13:15
  SPEAKER_00: A table named TBL Transactions lets formulas refer to TBL transactions amount instead of a less readable range, such as E2E500.
• 13:25
  SPEAKER_00: Structured reference, a formula reference that uses table and column names, such as TBL Transactions category.
• 13:32
  SPEAKER_00: Duplicate, a row that repeats another row according to a defined comparison rule.
• 13:38
  SPEAKER_00: Two similar-looking rows are not necessarily duplicates.
• 13:41
  SPEAKER_00: Conditional formatting, automatic formatting that changes based on a rule, such as turning an over budget amount red.
• 13:48
  SPEAKER_00: Dashboard, a summary worksheet that combines important figures, tables, and charts in one view.
• 13:54
  SPEAKER_00: Chart, a visual representation of data.
• 13:56
  SPEAKER_00: A column chart compares categories.
• 13:58
  SPEAKER_00: A line chart is often used for changes over time.
• 14:01
  SPEAKER_00: Your first next step, take one small nonsensitive workbook and make a working copy.
• 14:07
  SPEAKER_00: Then paste this prompt into Codex.
• 14:09
  SPEAKER_00: This gives you a low-risk first result, an audit rather than an irreversible edit.
• 14:14
  SPEAKER_00: Once you have a workflow that works, save the prompt, rules, and verification checklist as a reusable template or skill.
• 14:21
  SPEAKER_00: OpenAI describes skills as reusable playbooks that can capture spreadsheet workflows, formatting rules, and review steps.
• 14:29
  SPEAKER_00: Conclusion.
• 14:30
  SPEAKER_00: A coding agent does not replace your judgment.
• 14:32
  SPEAKER_00: It removes much of the mechanical work so that you can focus on the rules that matter.
• 14:37
  SPEAKER_00: Which rows are genuinely duplicates, which spelling changes are safe, what budget limits should be used, which figures should appear on the dashboard, what needs human review before sharing.
• 14:48
  SPEAKER_00: The safest and most useful pattern is plan first, edit second, verify third.
• 14:53
  SPEAKER_00: Start with a copy of a small workbook, ask for an audit.
• 14:57
  SPEAKER_00: All links to sources are available in the text version of this article.
• 15:02
  SPEAKER_00: You can find the full article at easycoding.tools blog.
• 15:08
  SPEAKER_00: Thanks for listening.
• 15:10
  SPEAKER_00: For more practical AI coding guides, tool comparisons, and resources for builders, visit easycoding.tools.
• 15:18
  SPEAKER_00: And if it's easier to remember, just go to aibuilds.it.
• 15:22
  SPEAKER_00: It redirects to easycoding.tools.