Wichtige Fragen & Klare Antworten
11 Fragen und Antworten, mit denen Sie bei jedem Gespräch souverän mitreden können.
How is a personal AI assistant like Instinct or Grokbot different from a standard chatbot?
A standard chatbot only answers when you open a window and type a prompt. A personal assistant or teammate works proactively: Grokbot acts as an autonomous colleague with its own cloud computer to write code for developers or research leads for salespeople 24/7. Meanwhile, the Instinct app acts like a real-world human assistant answering your phone calls and text messages to schedule appointments, negotiate bills, and coordinate life logistics without requiring screen time.
What's the difference between Skills and claude.md?
Think of Skills as special moves you teach an AI (like 'always write in this style'). claude.md is like an instruction manual file you give the AI at the start, telling it the rules of the game.
Give examples of 3 events where Hooks are activated
Hooks are like trip wires — when something specific happens, they automatically trigger an action. Examples: 1. When a user sends a message → automatically save it to a log. 2. When an AI finishes a task → send a notification to the user. 3. When an error occurs → automatically alert the developer.
What does MCP add beyond Skills?
MCP (Model Context Protocol) is like giving the AI a USB port — it lets the AI plug into external tools (like Google, databases, apps), not just do things it already knows.
What's the advantage of working with Sub-agents instead of one big Agent?
Imagine one chef trying to cook, serve, and wash dishes alone vs. a whole kitchen team. Sub-agents are like specialist team members — faster, smarter, and they work in parallel.
Whats the difference between an agent and a subagent?
While both help parallelize work, their structure differs: **Subagents** operate hierarchically, reporting results back to a main agent (the 'caller')—this is cost-effective and best for focused tasks. **Agent Teams** consist of independent agents that message each other directly and self-coordinate, making them ideal for complex work requiring discussion and collaboration.
What's the difference between an Agent and a Deterministic Workflow?
A Deterministic Workflow always follows the exact same steps (like a recipe). An Agent thinks and decides its own steps depending on the situation — it's flexible.
When is it better NOT to use RAG, but Fine-Tuning instead?
RAG is like giving the AI a cheat sheet to look up answers in real time. Fine-Tuning is actually teaching the AI new knowledge permanently. Use Fine-Tuning when the knowledge needs to be 'baked in', not just looked up.
Why is it a bad idea to stuff ALL the context into the prompt every time?
Imagine re-reading an entire book before answering every single question. It's slow and expensive. You should only feed the AI what it actually needs for that moment.
What's the problem with the LLM accessing the Database or Production API directly?
Letting AI touch your live database directly is like giving a new intern the master key to the building. Dangerous. One mistake = real damage. You need safety layers in between.
How do you measure the quality of an LLM-based system in production?
In production means in the 'real world', not just testing. You need to track things like: Did it give correct answers? How often does it fail? How fast is it?
How do you manage Memory in an Agents system — what's a Vector DB vs regular DB?
A regular database stores data in neat rows/tables (like Excel). A Vector DB stores data by 'meaning' — so the AI can find things that are similar, not just exact matches. Perfect for AI memory.