AI-Powered Development

From Buzz to Build

A Learning Lunch about AI-Assisted Development
What changes. What stays. And why you should just try it.

Press Arrow Right to begin — or F for fullscreen

The Cambrian Explosion of AI Tools

Tools are popping up like mushrooms in the forest. We're living through the most rapid tooling evolution in software history.

What Actually Changes?

Yesterday

"Person who types code in IDE"

Focus on syntax, boilerplate, gluing APIs, scaffolding projects

→

Tomorrow

"Product thinker who validates hypotheses at the speed of thought"

Focus on context, requirements, architecture, business value

"AI doesn't replace developers who adapt.
It replaces the tasks, not the role."

The Agile Reality Check

20 years of agile adoption. We have the ceremonies. Do we have the values?

What we often see

Long ceremonies, longer iterations
Multi-PI roadmaps carved in stone
6-month "discovery phases"
Silver-bullet solutions designed upfront

What agile actually means

Customer-centricity over process
Fast feedback cycles
Build hypotheses, validate fast
Recognize misdirection, change path

Prototype at the Speed of Thought

AI is our best tool for fast validation. Build prototypes good enough to test business assumptions — without building the silver-bullet solution first.

📅

Family Planner

A friend built a full family agenda web app over a single weekend.

🏒

Hockey SaaS

Another friend built a scheduling SaaS for his spare-time hockey team. Side project to product.

☁️

Sovereign Cloud

Full IaC-based open-source cloud stack: multi-tenancy, management UI, carrier-grade network isolation. VMware replacement, deployable in 30 min.

How to Use AI Effectively

Not technically — methodically.

Start with context, not commands
Tell the AI what you're building and why. A system prompt is your project's constitution.
Iterate, don't generate-and-pray
Treat AI like a pair programmer, not a vending machine. Review. Challenge. Course-correct.
Set boundaries explicitly
Define what AI can do autonomously and where it needs your sign-off.
Version control is your safety net
AI writes fast but can break fast. Git is your undo button. Commit often.
You are the quality gate
The domain expert. The one who says "this doesn't make sense for our customers."

Know the Limits

AI is powerful. It's not magic. Understanding the boundaries makes you a better user.

AI is great at

Scaffolding and boilerplate
Exploring unfamiliar frameworks
Refactoring and code review
Explaining complex code
Writing tests and docs
Rapid prototyping

Humans stay essential for

Business domain judgment
Ethical and security decisions
Architecture at scale
Customer empathy
Knowing when to say "no"
Verifying correctness

AI optimizes for "plausible", not "correct".
You optimize for "right for the customer".

Where Things Are Heading

Agentic Coding

Tools like Claude Code don't just suggest — they read your codebase, run commands, create files, and iterate on errors. Qualitatively different from autocomplete.

Tool Use & MCP

AI that interacts with your actual systems: databases, APIs, CI/CD, infrastructure. Model Context Protocol is making this standardized and interoperable.

Local & Open Source

Decentralization and democratization of AI. Run models locally with Ollama, fine-tune for your domain, keep data sovereign.

Responsible Use

Don't paste production secrets into AI. Understand your data policies. Use enterprise plans or local models where sensitive data is involved.

Just. Try. It.

No clear "learning path". No "use tool X, don't use tool Y".
Just be curious. Start a pet project. Break things. Learn.

"The best way to predict the future is to build it."

It doesn't have to be for work. A hobby project. A family tool.
A weekend experiment. The barrier to entry has never been lower.

Live Demo: This Presentation

What you're looking at right now was built by an AI assistant (Claude Code) in a live coding session. Here's what happened:

Built with Astro — a modern web framework
Matrix theme with light-mode toggle — because why not
Keyboard navigation, fullscreen support, progress bar
Packaged as a Docker container
Deployed to a fresh Hetzner VM — right now, live

$ ssh root@<vm-ip> 'docker run -d -p 80:4321 ai-slides'

Thank You!

Questions? Ideas? Let's chat.

Start today: claude.ai | chatgpt.com | ollama.com

This presentation: built with Astro + Claude Code
Source & slides available after the session