AI Tools at Work
🖱️ Interaktiv: Ziehen zum Drehen · Lektion 6
Learning objectives
After this lesson you will be able to:
- Use ChatGPT, Copilot and specialized AI tools safely in everyday work
- Detect and address shadow AI in your organization
- Write effective prompts and critically review results
- Apply quality assurance measures to AI output
AI tools at a glance
The most important AI tools in everyday work:
ChatGPT / Claude / Gemini
General-purpose AI assistants for text creation, research, translation, brainstorming and summaries. Not suitable for confidential data without an enterprise contract.
GitHub Copilot
An AI programming assistant that suggests code, generates boilerplate and writes tests. It boosts productivity but requires careful code review.
Specialized AI tools
- AI translators (DeepL, translate.google.com)
- AI image generation (Midjourney, DALL-E)
- AI meeting notes (Otter.ai, Fireflies)
- AI research (Perplexity, Consensus)
Shadow AI — the invisible danger
Shadow AI is the use of AI tools by employees without the knowledge or approval of the IT department. It is one of the biggest compliance risks.
Why shadow AI is risky:
- No GDPR review: is the data processed in the EU?
- No DPA: who is liable in case of a data breach?
- Data leakage: confidential information leaves the company
- No training: employees don't know the risks
- Liability: the company is liable for employee mistakes
💡 Practical measures against shadow AI
- Clearly communicate which AI tools are allowed (whitelist) 2. Create and communicate an AI usage policy 3. Survey employees regularly about the AI tools they use 4. Offer training on safe AI alternatives 5. Technical measures (browser plugins, DLP)
📝 Schnellprüfung
What is the biggest risk of shadow AI?
Prompting basics
A prompt is the input (request) you give to an AI system. Good prompts lead to better results.
The CRAFT method
| Letter | Meaning | Example |
|---|---|---|
| C Context | Provide context | "Our company is launching a new SaaS product" |
| R Role | Define a role | "You are a marketing expert in the SaaS field" |
| A Action | Clear instruction | "Write a social media post" |
| F Format | Output format | "As a table with columns A, B, C" |
| T Tone | Specify the tone | "Professional but easy to understand" |
Important: the prompt determines the quality of the output. Unclear prompts lead to unusable results.
Quality assurance for AI output
The four-eyes principle for AI
Every AI output that reaches customers or business partners must be reviewed by a human.
AI quality assurance checklist
- Fact check: are all numbers, data and facts correct?
- Bias check: does the content contain distortions?
- Completeness: is anything important missing?
- Tone: is the tone appropriate?
- Legal review: are there legal risks?
- Data protection: does the output contain personal data?
🏢 Praxis-Szenario: Unreviewed AI output
A sales employee has ChatGPT write a quote email. The email contains a serious spelling error in the company name and states the wrong price (€99 instead of €999). The email goes out to 50 customers.
Using AI tools safely
Data protection at work
- Never enter personal data (names, addresses, employee numbers) into public AI tools
- Never share trade secrets or confidential information
- Use enterprise versions (opt-out for data training)
- Review output before sharing it
💡 Rule of thumb for AI inputs
Never enter information into an AI tool that you wouldn't publish in a daily newspaper. What you enter into ChatGPT Free, Google Gemini or Claude Free may be used to train the next model.
✅ Wichtige Erkenntnisse
Haken setzen, um deinen Lernfortschritt zu markieren:
→ Go deeper: AI governance in the enterprise — AI policy and approval processes · Transparency obligations — AI inventory for documenting AI tools