Anthropic’s Claude has passed a milestone that can easily be measured practically: it can now generate real files — spreadsheets, slide decks, documents and PDFs — straight out of chat. Request a revenue model in Excel, a client-ready PowerPoint presentation derived from a report or a well-formatted memo, and Claude puts the package together without closing the conversation.
What Claude can issue now
The new functionality supports popular office formats such as Excel spreadsheets, PowerPoint presentations, Word files, and PDFs. You can send Claude raw data, an uploaded file, or a detailed prompt, and it will return downloadable files with charts, tables and narrative analysis.
Example use cases might include taking a CSV of transactions and using it to create a multi-tab financial model with assumptions, summary dashboards and toggles to switch between scenarios, or distilling a long white paper into a 12-slide deck with speaker notes, or turning a PDF into an outline, then a formatted report.
Anthropic frames this as a way to help automate the grunt work that nibbles away at the first draft, whether by wrangling data, constructing a chart, laying out a slide or baking boilerplate formatting. What you get in return is not just the file, but more often a reasoning for design and formula choices, that makes review and iteration easier.
How it works (under the hood)
Anthropic writes that Claude does the heavy lifting in its own private computing environment, where it writes and executes code and then packages output. In practice, that means that it can do the data wrangling, the visuals and finishing touches and export the final files in one continuous flow — similar to code-interpreter experiences, but squarely targeted at creation of documents.
This is a pretty explicit foray into agentic AI, where models are taking it upon themselves to plan and enact multimodal tasks for the user. Competitors are similarly going in the same direction: OpenAI has agents browsing with features for compiling reports; Google is piloting agentic browsing prototypes; and new browsers like Arc and Perplexity experiment with assistants shopping or summarizing across the web. It’s a bit more stringently focused on “give it inputs, get a file,” which removes ambiguity and keeps you in the captain’s seat when it comes to the source materials.
Availability and how to try it
The feature is now available for as a preview for Claude Max, Team, and Enterprise subscribers and will roll out to Pro users. You can turn it on by visiting Settings, then tapping on Features and turning on Experimental, then enabling “Upgraded file creation and analysis.” Then add your data or paste instructions and the format you would like your output.
Helpful command patterns might be: “Create a three-statement model with base, best and worst cases from this CSV,” “Summarize this 40-page PDF report into a 10–12 slide PowerPoint deck with charts” or “Build a project tracker in Excel with conditional formatting and a burn-up chart.” Claude usually sends back the done file together with a brief explanation of methodology and assumptions.
Why it Matters for Workflows
Decks and spreadsheets are the currency of communication for many teams. Automate the first 60–80% of that effort — collect the data, structure the sheet, draft the slide — and you can shrink cycles from hours to minutes. Microsoft’s Work Trend Index has found massive demand for AI to take over routine tasks, while McKinsey forecasts generative AI could contribute trillions of dollars in annual economic value by streamlining knowledge work. File-generating agents are a straight shot at those gains.
Early adopters among those in finance, consulting, and operations will arguably see the fastest payback: repeatable models, standardized reporting templates, weekly performance packs — all can be spun up on demand, and then massaged by humans who understand the business context.
Limits and good practices
Likewise with any AI assistant, trust but verify. Verify formulas, pivot logic and chart ranges before distributing. If you want quote formatting or strict compliance with footnoting regulations, give examples at the beginning and lock them into the prompt. But for sensitive data, check your organization’s policies; Anthropic says that conversations can be used to improve models unless users opt out in privacy settings.
Concrete instructions yield the best results: Provide clean source data if you are able, indicate the file type and audience, note the key measures you want to emphasize and describe the story line you want your slides to follow. Think of Claude as an efficient junior analyst (he’s great with structure and speed but the final call still belongs to you).
The bottom line
And Claude can now’s output finished spreadsheets and slide decks […], moving AI from the conversational to the deliverable. It won’t replace experts, but it can strip away the drudgery between raw inputs and a credible first draft — freeing teams to focus on insight, not formatting.
