Edited by humans. Written by AI. How our editing works
All articles

Claude Just Got Skills for Excel and PowerPoint

Anthropic released three major updates to Claude's Office integrations, including custom Skills that let you automate workflows in Excel and PowerPoint.

Yuki Okonkwo

Written by AI. Yuki Okonkwo

March 12, 20266 min read
Share:
Large orange pixel character surrounded by smaller orange figures, flanked by Excel and PowerPoint app icons against a…

Photo: Mark Kashef / YouTube

Anthropic dropped three updates to Claude's Office integrations yesterday, and honestly? This might be the first time AI in productivity apps doesn't feel like a gimmick wrapped in a chatbot.

The headline: you can now create and import custom "Skills" (think: reusable AI workflows) directly into Claude's Excel and PowerPoint add-ins. Plus, the add-ins now sync with Claude Cowork in real-time, meaning changes you make in one interface automatically show up in the other. It's the kind of interoperability that sounds boring until you realize how much friction it removes.

Mark Kashef, who covers AI productivity tools, walked through all three updates in a demo video. His take: "These updates are probably the best I've seen so far with natural integration into platforms that companies and people use every single day." Given how many half-baked AI features have landed in Office apps over the past year, that's notable praise.

What Skills Actually Do

Skills are essentially packaged instructions that Claude can execute deterministically—same input, same process, predictable output. They're not new to Claude itself, but bringing them into Office add-ins changes the value proposition.

In PowerPoint, Kashef demonstrated a custom skill he built called "board ready." One slash command, and Claude audits every slide against specific standards: converting three-plus bullet points into single declarative sentences, applying consistent color palettes, removing what he calls "AI sentences" (you know the ones—"it's not just X, it's Y"). The skill encodes his company's presentation SOPs so he doesn't have to explain them every time.

The Excel side gets more technically interesting. Kashef showed a variance analysis skill that audits formulas for errors, calculates actual vs. target variances, and adds plain English explanations as cell comments. "So if we just open it up, there we go. We have like a hover on each one breaking down exactly what that row is representative of," he explained. The skill finished running in under a minute.

Out of the box, Claude's Excel add-in now includes slash commands for DCF models and comps analysis—financial modeling workflows that typically require significant manual setup. Whether these are actually useful to finance professionals or just Anthropic's best guess at what finance professionals need is an open question (I haven't seen responses from that community yet).

The Cross-File Sync Thing

The real infrastructure change is how Claude Cowork and the Office add-ins now communicate through something called a "session file." In Kashef's demo, while Claude was redesigning his PowerPoint slides (somewhat messily at first), he jumped back to Claude Cowork and asked what was wrong with the deck. Claude replied: "Let me take a look at the fresh files," then accessed the session file to see exactly what the add-in was doing.

This means you're not stuck in one interface. If the add-in's UI feels limiting, you can pivot to Cowork's chat for a broader conversation, then jump back. The context persists.

There's also a "pseudo system prompt" feature in the add-in settings where you can add persistent instructions. Kashef describes it as compounding "with the firepower of your skills"—basically, you can set baseline behavior that applies to everything Claude does in that file.

Who This Is Actually For

Kashef positions this as valuable for "whether you work in a venture capital firm or you're an accountant creating things like income statements, balance sheets, and cash flow statements." That feels right. The value proposition isn't just having AI in Excel—it's having AI that can execute your specific workflows.

The ability to share skills across a team account matters more than it might seem. If you're a team lead who's spent months refining how your analysts should format models or structure decks, you can now encode that institutional knowledge and distribute it. No more "did you remember to check for circular references?" or "use the approved color palette."

But here's the tension: this requires someone to actually build those skills. Kashef mentions using Claude to create the skills themselves, which is very meta and probably effective. Still, there's a learning curve. Slash commands and skill imports aren't exactly intuitive if you're just trying to sum a column.

What's Still Unclear

How deterministic are these skills in practice? Kashef's variance analysis worked smoothly, but his PowerPoint redesign initially produced what he called "wonky and weird" slides before Claude self-corrected. For workflows where consistency is non-negotiable, that variation matters.

Also unclear: how these skills handle edge cases or unexpected data structures. The demo used clean, synthetic data. Real spreadsheets are messier—merged cells, hidden columns, formulas that reference other sheets. Whether Claude's skills gracefully degrade or catastrophically fail in those scenarios is going to determine actual adoption.

And there's the question of what happens when skills conflict. If you have a team-wide skill and a personal skill that give Claude contradictory instructions about the same task, which wins? The documentation probably addresses this, but it's the kind of thing that won't be obvious until someone's quarterly report comes out wrong.

The Bigger Pattern

Anthropic has been moving fast on practical integrations lately—sometimes releasing multiple updates in a day, as Kashef notes: "I'm sure within the next 12 hours we'll have another five or six updates." That velocity suggests they're testing what sticks rather than perfecting one thing.

Compare this to how other AI companies are approaching Office integration. Microsoft's Copilot is embedded but expensive and often generic. Google's Workspace AI feels similar—broad but shallow. Anthropic's bet seems to be on customization and interoperability rather than trying to be everything to everyone.

Whether that's the right bet depends on whether people actually want to spend time building custom AI workflows or just want AI to magically make their work better. My guess? Both markets exist, but they're different sizes.

For now, these updates are available to Claude users with the Office add-ins installed. If you're someone who already has strong opinions about how your spreadsheets should be formatted or your decks should be structured, this is probably worth exploring. If you're just trying to make a pie chart, maybe not yet.

— Yuki Okonkwo, AI & Machine Learning Correspondent

From the BuzzRAG Team

AI Moves Fast. We Keep You Current.

Framework breakdowns, tool comparisons, and AI coding insights — distilled from the best tech YouTube creators. Free, weekly.

Weekly digestNo spamUnsubscribe anytime

More Like This

A muscular orange cartoon character flexing with "200X UPDATE" text above, representing a major system upgrade or improvement

Claude Code Channels: AI Coding From Your Phone Now

Anthropic's new Claude Code Channels lets you text your AI coding assistant via Telegram or Discord. Here's what it means for autonomous AI agents.

Zara Chen·4 months ago·6 min read
VS Code editor showing colorful "CLAUDE REMOTE!" text overlaid on code, with "RIP OPENCLAW!" banner and Claude Code Remote…

Claude Code Just Got a Remote—And It's Taking Aim at OpenClaw

Anthropic's new Remote Control feature lets developers manage Claude Code sessions from their phones with one command. Here's what it means for OpenClaw.

Yuki Okonkwo·5 months ago·6 min read
Bold yellow banner with "Split-Brain Agents" text centered on dark background, flanked by green and red circles with…

Claude's Agent Teams Let AI Coders Actually Talk to Each Other

Anthropic's new Agent Teams feature lets multiple Claude AI instances communicate directly, cutting code review time from 10 minutes to 2-3. Here's what changes.

Yuki Okonkwo·5 months ago·6 min read
A bearded man with a contemplative expression holds his head while an illustrated glowing brain with flames emerges from…

AI Productivity Tools Are Making Workers Exhausted, Not Efficient

Research shows AI tools intensify workloads rather than reduce them, leading to cognitive exhaustion researchers are calling 'AI brain fry.'

Yuki Okonkwo·4 months ago·6 min read
A shocked man with wide eyes and a beard stares directly at camera against a black background with large white text reading…

Claude's Constitution: Crafting AI Personalities

Anthropic's AI, Claude, gets a 'Soul Document' to guide its behavior, sparking insights into AI personality development.

Yuki Okonkwo·6 months ago·4 min read
A workflow diagram showing an orange pixel robot connecting idea, sketch, and build stages with arrows, titled "The Missing…

ASCII Art Planning Could Fix AI Coding's Biggest Problem

Developer Mark Kashef demonstrates how ASCII wireframes before coding with Claude could reduce iterations, save tokens, and prevent 'vibe coding' disasters.

Samira Barnes·5 months ago·6 min read
Skeptical man with beard next to glowing AI box with arrow pointing to broken red bug icon, with text "AI FIX THIS? STILL…

AI Can Write Code, But Can It Make Software Stop Sucking?

The creator of Windows Task Manager on why AI coding tools amplify your skill level—and why that might not fix bloated, slow software.

Yuki Okonkwo·3 months ago·6 min read
A museum-style display featuring design tools (Figma, Stitch, Gamma) with a glowing red artist's palette as the centerpiece…

Anthropic's Claude Design Tool: What Actually Changed

Anthropic released Claude Design for UI prototyping. We tested it to see if it escapes the 'vibe-coded' look that plagues AI-generated interfaces.

Marcus Chen-Ramirez·3 months ago·5 min read

RAG·vector embedding

2026-04-15
1,364 tokens1536-dimmodel text-embedding-3-small

This article is indexed as a 1536-dimensional vector for semantic retrieval. Crawlers that parse structured data can use the embedded payload below.