--- title: "Copilot learns your monthly Excel job, and checking becomes the work" description: "Copilot in Excel learns recurring work. AI agents make more, but checking takes longer. And Google teaches Gemini 4 Argon to stop and ask." image: "https://bombos.ai/assets/og-en-v2.png" --- # Copilot learns your monthly Excel job, and checking becomes the work Copilot in Excel learns recurring work. AI agents make more, but checking takes longer. And Google teaches Gemini 4 Argon to stop and ask. *Bombos team · 14 October 2026 · 4 min read* [Book a call](https://cal.com/piet-baudoin-q9lwaz/15min) [Scan my website](https://bombos.ai/en/scan/)[Leave your details](https://bombos.ai/en/contact/) _By Piet Baudoin · 14 October 2026_ **AI finishes more and more work, and the question becomes who checks it and when it stops.** Three stories from this week. ## Can Copilot keep your monthly Excel report up to date on its own? Microsoft announced new features for Copilot in Excel on 12 October. ICT Magazine, a Dutch IT trade publication, sums it up: “Microsoft breidt Copilot in Excel uit met nieuwe functies waarmee gebruikers spreadsheets eenvoudiger kunnen omzetten in interactieve rapportages” (Microsoft is extending Copilot in Excel with features that make it easier to turn spreadsheets into interactive reports). Such a report stays tied to the workbook. Microsoft: “it automatically updates as your data changes.” The most striking part is about the work that comes back every month. With a new feature you show a recurring task once, and “Copilot observes the process and turns those actions into a reusable skill.” Cleanup is part of it too, with an example everyone knows: “Sometimes “USA,” “US,” and “United States” all appear in the same column.“ Not everything is here yet: the report and the show-me feature are still in the early access program, while cleanup and getting data roll out to everyone this month. Microsoft sells Copilot itself. What it means for you: think of the spreadsheet that the same person updates every month at your company, with numbers that then still have to go to accounting or the client. That job becomes a skill you show once. What happens outside Excel, like your email or your accounting software, is not in the announcement. How those fit together is in [automating documents, Excel files and email with AI](https://bombos.ai/en/knowledge-base/how-do-you-automate-documents-excel-files-and-emails-with-ai/). Source: [Microsoft, Brian Jones, 12 October 2026](https://techcommunity.microsoft.com/t5/excel-blog/beyond-analysis-copilot-gets-to-work-in-excel/ba-p/4517581) ## Does an AI agent make your team faster if someone has to check everything? Two Harvard researchers, Fiona Chen and James Stratton, looked at more than 700 software companies with more than 700,000 employees between them. Ars Technica wrote about it on 9 October. After those companies started using AI agents, there was 30 percent more code and 23 percent more proposals to change something. But no more got finished. Checking such a proposal took 49 percent longer on average, and the share of proposals sent back nearly doubled. The researchers found “little evidence that firms increase software output or reduce employment”. What it means for you: this is about programmers, but you will recognize the pattern everywhere. If you let AI make more drafts, whether quotes, emails or overviews, the work shifts to whoever has to check them. What helps is a draft that says where it differs, so you don’t have to read everything again. That is also the question to ask when you [test an AI agent](https://bombos.ai/en/knowledge-base/how-to-test-an-ai-agent-before-automating-business-processes/). Source: [Ars Technica, Kyle Orland, 9 October 2026](https://arstechnica.com/ai/2026/10/ai-coding-agents-generate-more-code-but-not-more-software/) ## What does an AI agent do when it runs into a locked door? Google released its new model Gemini 4 Argon on 13 October, for paying users of the Gemini app. It came later than planned: Google first let security defenders work with it because of what the model can do in finding vulnerabilities. In its safety post, Google is candid about what goes wrong in practice: “an agent trying too hard to finish the job it was given. When a resourceful assistant is told to get a task done and runs into a locked door, it can treat that obstacle as just another puzzle to solve.” That is why Argon was trained to “respect system boundaries, report negative findings honestly, and pause to ask for human confirmation.” Google sells the model itself. Yesterday we wrote here about how an Anthropic agent submitted a form it was meant to leave alone. Now two big makers say the same thing: the agent wants to finish the job, and that is exactly why you put the limit in the work itself. Source: [Google, Anca Dragan, 13 October 2026](https://blog.google/innovation-and-ai/technology/safety-security/gemini-4-model-safety-security/) ## What does this mean for your business? The common thread: AI finishes more than ever, and what counts is what happens before it goes out the door. At Bombos, that sits in the work itself. Take [the monthly invoicing round](https://bombos.ai/en/use-cases/the-monthly-invoicing-round/): an AI agent pulls the hours and receipts from your spreadsheet, prepares the draft invoices and marks what differs from last month. You look at those few differences, approve and send. The invoices are ready at the start of the month, and you don’t have to read everything again. That one task is the beginning; after that your team teaches Bombos the next one itself. ## Sources Every source was opened on 14 October 2026 and every quote appears in it word for word. The ICT Magazine quote comes from the article’s introduction; the full page would not open. That Argon came later because of security concerns is in the Tweakers headline; the full page would not open, and what Google itself writes about early access for defenders is in the source below. [Microsoft on Copilot in Excel](https://techcommunity.microsoft.com/t5/excel-blog/beyond-analysis-copilot-gets-to-work-in-excel/ba-p/4517581) (12 October 2026) · [ICT Magazine, Copilot makes Excel data more visual](https://www.ictmagazine.nl/nieuws/copilot-maakt-excel-data-visueler/) (13 October 2026) · [Ars Technica on the Harvard study](https://arstechnica.com/ai/2026/10/ai-coding-agents-generate-more-code-but-not-more-software/) (9 October 2026) · [Google on the safety of Gemini 4 Argon](https://blog.google/innovation-and-ai/technology/safety-security/gemini-4-model-safety-security/) (13 October 2026) · [Google on Gemini 4 Argon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/) (13 October 2026) · [Tweakers, Google releases Gemini 4 Argon after delay over cyber concerns](https://tweakers.net/nieuws/253284/google-brengt-gemini-4-argon-na-uitstel-om-cyberzorgen-alsnog-uit.html) (13 October 2026) ### Curious what Bombos can do for your business? Together we look at which work keeps piling up and set your workers up for exactly that. [Book a call](https://cal.com/piet-baudoin-q9lwaz/15min) [Scan my website](https://bombos.ai/en/scan/)[Leave your details](https://bombos.ai/en/contact/) --- ```json {"@context":"https://schema.org","@type":"BlogPosting","headline":"Copilot learns your monthly Excel job, and checking becomes the work","description":"Copilot in Excel learns recurring work. AI agents make more, but checking takes longer. And Google teaches Gemini 4 Argon to stop and ask.","datePublished":"2026-10-14T00:00:00.000Z","author":{"@type":"Organization","name":"Bombos"},"publisher":{"@type":"Organization","name":"Bombos","logo":{"@type":"ImageObject","url":"https://bombos.ai/assets/bombos-bee.png"}},"mainEntityOfPage":"https://bombos.ai/en/blog/newsletter/2026-10-14/","inLanguage":"en"} ```