Learn what we build
:Slide decks and guides from our practice. Clearly explained, technically grounded.
Agents Without a Chat Window
When work starts on its own: an agent writes an incident briefing without anyone asking in a chat. The talk demos three start signals live — a schedule, a Sentry webhook and a request from ChatGPT via MCP — and when each one fits. (Slide content in German.)
Mission Possible: An Apartment in Munich
How a chat with an AI turned into an agent that applies for apartments on its own. The build in four steps with prompts to reuse, what went wrong in operation — and a checklist for your own automation project. (Slide content in German.)
Memory under Attack
An AI assistant reads its project knowledge from a note file and works from it. What happens when someone edits that file? The talk runs the same development task three times and shows where the assistant asks back — and where it doesn't. (Slide content in German.)
Ich wiederhole mich. Ich wiederhole mich.
How to stop explaining the same things to an AI every day: shared knowledge lives in plain text files that both sides may read and write. A field report from ten weeks of daily practice, including the problems left unsolved. (Slide content in German.)
Building, attacking and securing MCP servers
One tool, one prompt injection, three security boundaries. 26 slides from the Agentic AI Meetup: build a minimal MCP tool, attack it through untrusted tool data, then secure it with a server-side policy. With a live demo, a defense-in-depth table, and four executable tests that define what “secure” actually means. (Slide content in German.)
Artifacts in Claude
How a conversation becomes a result you can actually work with. 19 slides for business teams: what an artifact is, the six formats (Markdown, PDF, Mermaid, SVG, HTML, interactive app), local vs. cloud, three everyday cases, and a four-step start. With an interactive format gallery and a mini quiz. (Slide content in German.)
Prompting is over
Why single prompts don't scale — and how teams turn them into repeatable AI workflows. 17 slides, six-step blueprint, three live demos: business projects, Claude Code with subagents, and Claude Managed Agents wired to Notion and Slack. (Slide content in German.)
omadia: an agentic os
Launch deck for omadia, byte5's self-hostable agentic OS. How individual agents become a team — plugin architecture, knowledge graph, hallucination verifier, all on your own infrastructure, no model lock-in. (Slide content in German.)
Prompts and Agents
A look behind the scenes of Claude. 37 slides, nine chapters, ~40 min: from Eliza and the anthropomorphism trap to context windows, the session illusion, agents, and where LLMs go wrong. Two interactive stations: stateless animation and context-window visualiser. (Slide content in German.)
OpenClaw Hackathon
Hackathon format for building your own OpenClaw AgentSkills. Live demo of a reference skill, skill-anatomy check, level selection (Entry / Intermediate / Expert), two-hour build sprint, Show & Tell. Three reference skills under `demos/` to fork from. (Slide content in German.)
How LLMs actually work
A 25-slide deck that walks through modern language models step by step — from tokens and embeddings through transformers and training to agents. Clear, technically grounded, hype-free. Includes a few interactive elements for workshops and self-study. (Slide content in German.)
Claude & Claude Code
From chat to agent stack — a 60–75-minute walkthrough of the Claude ecosystem for developers. Model family, Cowork, MCP stack, Skills/Hooks/Subagents, and Spec-Driven Development. (Slide content in German.)
Spec-Driven Development with GitHub Spec-Kit
Sub-deck from the Claude Code talk: vibe-coding doesn't scale. Live demo with GitHub Spec-Kit — from .constitution through .specify, .plan, .tasks to .implement builds a streaming Next.js chat app. (Slide content in German.)
Introduction to OpenClaw
A local AI assistant — reachable on WhatsApp, Telegram, Slack, iMessage, Signal, Matrix. 90 minutes live: gateway, markdown-based skills, free choice of LLM (Anthropic, OpenAI, Ollama). (Slide content in German.)