Claude · Claude Code · MCP · Skills
Finance Team Internal Sharing · Presenter: Camille · July 2025
Before we build, let’s speak the same language.
Fastest and most lightweight. Great for simple, repetitive tasks where speed matters more than depth. Lower cost per token.
The balanced everyday model. Good accuracy, reasonable speed. Works well for most coding and document tasks.
The most powerful and accurate model. Best for complex reasoning, long tasks, nuanced writing — this is what we use for serious work.
For finance work, document generation, and any task where correctness matters — Opus is the right choice. The quality difference is very noticeable on complex or multi-step tasks.
/ to invoke any skill instantlyIn terminal, type / and a menu of all available skills and commands appears. It’s like an old-school pixel game — primitive, powerful, and satisfying. /brainstorming, /frontend-design, /sap-slides — everything is one slash away.
Watch the walkthrough video, follow the steps,
and reach out to Kristian if you get stuck.
Reach out to Kristian Schier — he’s the host of this session and the go-to person for all installation questions. There’s a dedicated Teams group chat where you can ask anytime — Kristian will happily help everyone get up and running!
My most-used skill when starting any project. Claude interviews you to clarify requirements, challenge assumptions, and map out the full scope before writing a single line of code.
Claude challenges your idea from every angle — pokes holes, stress-tests assumptions, and forces you to think through edge cases. Great for hardening your plan.
Generates polished, professional UI. The difference between a plain form and a beautiful interface — see the next slide.
Claude is your worker but it needs clear instructions. The /brainstorming skill ensures your worker understands exactly what you want — saving you 10x the back-and-forth and rework later.
Same functionality — same data — completely different user experience. One slash command makes the difference.
This MCP server connects Claude Code directly to SAP’s internal tools and platforms — giving Claude a “SAP badge” to access internal resources, data, and workflows your team already uses daily.
Setup guide and source code are on SAP’s internal GitHub. Clone it, follow the README, and you’re connected in minutes.
🔗 github.tools.sap / super-gc-saperWith SAP MCP installed, Claude Code can interact with SAP-internal platforms directly from your terminal — automating workflows that previously required manual navigation through multiple systems. This is where AI meets your actual day job.
Build your own projects, or stand on the shoulders of others.
GitHub is where your work lives and is shared.
Use Claude Code to build tools for your team. Push to GitHub and share with a simple link — no IT deployment, no waiting, no approvals needed.
Find a community project that fits your needs. Clone it, make it yours, adapt it for your workflow in minutes. No need to start from scratch.
Every project comes with a README. Browse, star, and learn from thousands of AI tools built by developers worldwide — all free, all open.
From real pain points — built with Claude Code.
One success, one lesson.
Upload .docx with {field} placeholders — images & stamps preserved
First-row headers match field names — one row per notice
Server fills & batch-converts to PDF — real-time progress bar
Word/ + PDF/ for every row — format identical to template
Many colleagues were still waiting to get Copilot access. So I built a “Fake Copilot” — an Office.js add-in powered by Claude that would deliver ~80% of Copilot’s features to anyone, for free.
The code was ready. The install package was ready. Everything was in place.
SAP’s IT policy blocks all three sideload channels for Office add-ins. The package existed but could not be installed on any SAP laptop.
Lesson: Always check company permissions before investing in the build. AI can build anything — but corporate constraints are real walls.
Failure is part of building. The instinct was right — identify a gap, solve it with AI. The limitation wasn’t the technology, it was the permission layer. As AI capabilities grow, so will the pressure on companies to open more access.
Host your frontend for free. Once deployed, just share the link — anyone on the team can use it instantly, no installation needed.
Free cloud hosting for your backend. One-time setup, runs forever. Essential for server-side features like PDF conversion.
📋 To check logs: Dashboard → your service → Logs
Use it every time you build a UI. The before/after gap is enormous — there’s no reason to ship an ugly interface when one slash command changes everything.
The first build takes a few hours. But once deployed on GitHub with a Live Demo link, everyone uses it forever with zero manual effort. Build once, benefit always.
In an AI-first world, the most valuable skill is your ideas. AI handles execution — you supply the vision, the problem, the “wouldn’t it be great if…”
Anyone who can imagine a better workflow and communicate it clearly will outpace those who just execute manually.
Needs company permissions — but the tech is ready today.
That’s exactly the kind of problem AI can solve. The bottleneck isn’t technical ability anymore — it’s permission and imagination. The future belongs to people who ask “what if?”
The entire presentation was designed and written inside Claude Code — the AI agent CLI we just introduced. All content, layout, and logic generated through conversation.
A custom SAP-branded slides skill packaged with SAP 72 font, Indigo palette, 20+ layouts, and animation patterns. One slash command — professional deck output.
Screenshots of the before/after comparison were captured automatically via Playwright browser automation — Claude navigated, scrolled, and screenshotted without any manual steps.
A Python script assembles the final HTML — embeds SAP logo, before/after screenshots as base64, and writes a single self-contained .html file. Zero dependencies at runtime.
No Figma. No PowerPoint. No designer. Every layout decision, color choice, and content edit was made by describing what I wanted in plain language — in the terminal.
From blank page to 20-slide polished presentation: a single session. The hardest part was deciding what to say — not how to make it look good.
This deck is itself a live demo. The same tools, skills, and workflow described in these slides were used to create these slides. That’s what “AI in your hands” actually looks like.
All content from this session is available online. You can revisit the slides, share the link, or fork the repo. 🔗 camille1223.github.io/ai-sharing-slides
What would you like to build first?
Ask away — no question is too basic!