Olaf Monien:
Hotel Hesselet, Christianslundsvej 119, 5800 Nyborg
About Olaf Monien
Worksthop Agenda
Tuesday: Day 1 - AI-Assisted Development with Delphi
- The Impossible Intro - What AI can do for Delphi developers
- Overview of Day 1 and Day 2
- Wi-Fi, Delphi toolchain check
- Git intro for beginners
- Clone workshop repo, build + tests green
- What is an LLM? (Transformer, tokens, context window)
- Model landscape: GPT-5, Claude, Gemini, Llama, Mistral, DeepSeek
- Provider vs. model vs. API
- Cloud vs. local: latency, cost, privacy
- API basics: prompt, completion, streaming, temperature
- From autocomplete to agents
- What works, what doesn't
- Opportunities for Delphi teams
- Risks: hallucinations, security, licensing, "silent breakage"
- Where does my data go?
- Markdown as the "glue" in AI development (README, CLAUDE.md, prompts, conventions)
- Live comparison: Claude Code, Kilo.ai, Chucki.ai
- Tasks: implement feature, refactor legacy, fix bug, explain code
- Steering agents effectively, avoiding "AI drift"
- Tools: Figma, Pencil
- MCP (Model Context Protocol): connecting AI to external tools and data
- Teams work through prepared backlog (6-10 tickets)
- Gates: build green, tests green, review, prompt log
- Compare cloud + local models in real repo
- Criteria: planning, refactoring, tests, debugging, Delphi idioms
- Outcome: task → model matrix
- Delphi prompt library
- Review checklist + AI safety rules
- Repo conventions
- Model/tool selection matrix
Wednesday: Day 2 - AI Inside Delphi Applications
- Perspective shift: not "writing code" but "solving user problems"
- Micro-exercise: same prompt, different models
- Evaluation: correctness, latency, format, cost, privacy
- Provider abstraction (cloud/local interchangeable)
- Streaming, timeouts, retries, cancellation
- Logging with redaction
- Cost awareness hooks
- Examples: DB query, log parser, validation, reports
- Flow: model → tool call → Delphi → result → model
- A) Hand-rolled: HTTP/JSON, full control
- B) MakerAI-style: Delphi-native components
- Refactoring: app code depends only on interfaces
- Backend swappable
- Smart Customer Search — "Show me all customers from Hamburg who didn't order last
- year"
- Invoice/Document Parser — PDF/scan → structured data (OCR + LLM)
- Feature Brainstorm — attendees pitch & implement their own AI feature idea
- Security & privacy
- Latency & UX
- Cost controls
- Observability
- Testing & rollout
