# aitrainer.dk

> AI training, workshops, and consulting for Danish and international organizations. Run by Stefano Vincenti — external lecturer at IT University of Copenhagen & DIS Study Abroad, co-founder of BotTellMe, partner at TryZone, and founder of aitrainer.dk.

## About Stefano Vincenti

Stefano Vincenti works at the intersection of theory and practice in AI. He holds multiple professional roles:

### Academic Positions
- **External Lecturer, IT University of Copenhagen (ITU)**: Teaches modern AI practices, digital product development with AI agents, and the business potential of generative AI. Courses include Scandinavian case studies and the geopolitical implications of AI adoption.
- **Teacher, DIS Study Abroad in Scandinavia**: Teaches AI and digital transformation to international students studying in Copenhagen.

### Industry Roles
- **Co-founder, BotTellMe**: Builds and implements AI solutions for organizations, ensuring they create value from day one. Focus on AI agents, automation, and intelligent workflows.
- **Partner & AI Advisor, TryZone Enterprise Advisory**: Advises, teaches, and trains large enterprises and public organizations, bridging IT and business to implement AI. Clients include LEGO, Novo Nordisk, Saxo Bank, DSB, TDC Net, Pandora, Maersk, and Nexi Group.
- **Founder, aitrainer.dk**: Personal platform for AI training, workshops, and consulting.

### Publications & Contributions
- Contributed to **"Agentbogen"** by Jonathan Løw — the most widely read introduction to AI agents in Denmark.
- Writes regularly about AI agents, vibe coding, organizational AI transformation, and the EU AI Act.

### Areas of Expertise
- Generative AI (ChatGPT, Claude, Gemini, Copilot)
- AI agents, multi-agent systems, and agent orchestration (OpenClaw, CrewAI, AutoGen)
- Digital product development with AI
- AI governance, ethics, and EU AI Act compliance
- Organizational AI transformation and change management
- Vibe coding and AI-assisted software development
- Prompt engineering and AI security
- Scandinavian management and digital transformation

## What Stefano Delivers

1. **AI workshops and training** — customized, hands-on sessions for teams and leadership, onsite or online
2. **AI Readiness Sprints** — 4-week organizational programs that assess AI maturity, identify high-impact use cases, and create implementation roadmaps
3. **Keynote speaking** — on AI transformation, AI agents, and the future of work
4. **Advisory and consulting** — strategic AI guidance for enterprises through TryZone
5. **AI agent development** — building and deploying agent workflows through BotTellMe

## Services

- AI workshops and training (onsite and online)
- Generative AI courses for teams
- AI agents training
- Microsoft Copilot adoption workshops
- Vibe coding workshops
- EU AI Act compliance training
- AI strategy consulting
- Keynote speaking

## Role-Based AI Workshops

Half-day (4 hours) hands-on workshops tailored to specific roles. See [Role-Based Workshops](/rollebaserede-workshops) for the full catalog.

- **AI for Leaders** (DKK 9,500): Strategic scenario planning, decision support, stakeholder communication. For boards, C-level, and department heads.
- **AI for Software Developers** (DKK 7,500): Code generation, refactoring, architecture docs, security review. Built on ITU teaching and vibe coding experience.
- **AI for Scrum Masters** (DKK 7,500): Retrospective facilitation, sprint planning, team health checks, coaching with AI.
- **AI for Product Owners & Managers** (DKK 7,500): User stories, backlog refinement, market analysis, go-to-market, OKRs. From ITU's "Digital Product Development with AI Agents" course.
- **AI for Researchers & UX Professionals** (DKK 7,500): Desk research, interview analysis, persona generation, journey mapping. Inspired by LEGO collaboration.
- **AI for People Leaders & HR** (DKK 7,500): Talent management, onboarding, development plans with focus on bias, fairness, and GDPR.
- **AI for Testers & QA** (DKK 7,500): Test case generation, exploratory testing, regression analysis, bug reporting.

## Workshop Pages

- [Generativ AI Kursus for Virksomheder](/workshops/generativ-ai-kursus): Comprehensive generative AI training for businesses — covers ChatGPT, Claude, Gemini, prompt engineering, and practical business applications
- [AI Workshop København](/workshops/ai-workshop-kobenhavn): On-site AI workshops in Copenhagen and across Denmark — customized training at your location for teams and leadership
- [AI Agents Kursus](/workshops/ai-agents-kursus): Training on building and using AI agents — from chatbot basics to multi-agent systems, agent swarms, orchestration, and security design (OpenClaw, CrewAI, AutoGen)
- [Copilot Workshop](/workshops/copilot-workshop): Microsoft Copilot adoption and training — rollout strategies, productivity optimization, and measuring ROI
- [Copilot Training](/workshops/copilot-training): Microsoft 365 Copilot enterprise training — from license to real productivity, custom prompt libraries, adoption strategy with train-the-trainer, ROI measurement. DKK 7,500 per person, half day, min. 6 participants.
- [Claude Training](/workshops/claude-training): Claude: Chat, Cowork & Code — master Anthropic's Claude across three modes. Chat for research and writing, Cowork for deep collaboration in your files, Code for AI-powered development. DKK 7,500 per person, half day, min. 6 participants.
- [Vibe Coding Workshop](/workshops/vibe-coding-workshop): AI-assisted development training — build applications with Lovable, Cursor, GitHub Copilot while managing quality and technical debt
- [EU AI Act Workshop](/workshops/eu-ai-act-workshop): Compliance training for the EU AI Act — risk classification, practical implementation, major requirements from August 2, 2026

## Key Topics

- Generative AI implementation
- AI agents and autonomous systems
- AI governance and ethics
- EU AI Act compliance
- Organizational AI transformation
- Human-AI collaboration
- Prompt engineering
- AI security and risk management

## Latest Insights

- [Forbind en AI-agent til Jira og Confluence på 30 minutter](/indsigter/forbind-ai-agent-til-jira-og-confluence): Mini-guide dated June 24, 2026 for product owners and scrum masters. Practical 30-minute setup of a terminal-based AI agent — GitHub Copilot CLI + the open-source mcp-atlassian MCP server (run via uvx) — that reads your Confluence space, searches your Jira issues, refines feature descriptions, and writes back to tickets gated by your explicit "yes". Architecture: You → Copilot CLI (the agent) → MCP server mcp-atlassian (the bridge using Model Context Protocol) → Jira + Confluence. Seven steps: (1) install Copilot CLI via WinGet (recommended Windows 11) or npm (Node 22+), restart shell, verify; (2) first start in an empty workspace folder, trust the folder, /login via browser; (3) generate two PATs in Jira and Confluence (Data Center 8.14+/6.0+) — Cloud uses API tokens from id.atlassian.com with JIRA_USERNAME + JIRA_API_TOKEN instead; (4) install uv/uvx via winget install astral-sh.uv; (5) configure ~/.copilot/mcp-config.json with the mcpServers block pointing at uvx mcp-atlassian and the right env vars (Data Center vs Cloud), plus SSL_VERIFY only for self-signed corp certs; (6) restart Copilot CLI and run /mcp to confirm mcp-atlassian is connected with tools like jira_search and confluence_search; (7) put it to work — pull guidelines from Confluence, refine open Jira features against them, then update tickets after explicit approval. Three hygiene flags: READ_ONLY_MODE "true" on first session then flip to "false", JIRA_PROJECTS_FILTER and CONFLUENCE_SPACES_FILTER to scope blast radius. Troubleshooting (copilot not recognized, /mcp doesn't show server, uvx not found, 401 auth, agent reads but doesn't update) and a hard security checklist: tokens are plaintext — never commit mcp-config.json, never paste tokens in chats, set expiry, rotate on suspected exposure, lean on filters + read-only. AI wingman pattern (Claude Opus 4.8 / Sonnet 4.6 open beside you to debug errors and sanity-check config). Honest disclosure: tested on Windows 11; macOS/Linux work with small command tweaks; not an official Atlassian/GitHub/Microsoft product. Explicit disclaimer: not legal/security/compliance advice — clear write access with your own IT/security/compliance before production. Downloadable PDF (Danish, 10 pages). Keywords: GitHub Copilot CLI, MCP, Model Context Protocol, mcp-atlassian, Jira MCP, Confluence MCP, Personal Access Token, Jira Data Center, Jira Cloud, uvx, uv Astral, READ_ONLY_MODE, AI for product owners, AI for scrum masters, terminal AI agent, agentic workflow Jira, AI Built Human, Stefano Vincenti.
- [Claude, jurisdiktion og ID-verifikation](/indsigter/claude-jurisdiktion-og-id-verifikation): One-pager as of June 23, 2026. The new Anthropic ID verification (Persona, from July 8) is the wrong question to anchor on. The right one is two-part: (1) which plan are you on — consumer (Free/Pro/Max, consumer terms, ID verification, content may train the model unless you opt out, no DPA — never for business data) or commercial (Team/Enterprise on Commercial Terms — you are controller, Anthropic processor, no training on your content, DPA with SCCs included, no ID verification); (2) which data should ever touch a US model. Jurisdiction matters more than location: the CLOUD Act (2018) reaches the parent company, not the Frankfurt server room — EU Data Boundary moves data location, not jurisdiction. GDPR angle: claude.com is a Chapter V third-country transfer; currently rests on the Data Privacy Framework (in force since July 2023), but DPF is under appeal at the CJEU (case C-703/25 P) and rests on a US presidential order that can change — keep an SCC fallback ready; for EU residency route via AWS Bedrock EU (Ireland/Frankfurt). Data tiering — GREEN (usually fine): strategy, marketing, internal knowledge, design, coding without trade secrets; YELLOW (prefer EU): customer-service, personalization, vendor data; RED (not US): customer personal data, HR, sensitive contracts, special categories (health, biometrics). Grown-up answer in five steps: (1) find your plan; (2) read your DPA (GDPR art. 28); (3) decide data tiering and write it down; (4) keep an SCC fallback + transfer impact assessment ready before any DPF ruling; (5) regulated sector — involve your DPO or counsel now. Disclaimer: general practical guidance, not legal advice. Downloadable PDF (Danish). Keywords: Claude ID verification, Anthropic Persona, consumer vs commercial terms, GDPR Chapter V, Data Privacy Framework, DPF appeal C-703/25 P, SCC fallback, CLOUD Act, EU Data Boundary, AWS Bedrock EU, data tiering, AI vendor risk, AI Built Human, Stefano Vincenti.
- [105 AI-prompts til finanssektoren](/indsigter/105-ai-prompts-til-finanssektoren): Featured. Prompt guide built for the Danish financial sector, presented at Finansforbundet's interactive webinar on June 17, 2026 in collaboration with IT University of Copenhagen and TryZone. 105 prompts across 11 categories: personal daily workflow (01–12), meetings (13–22), control/compliance/audit (23–34), risk management (35–44), data analysis & quality (45–54), documentation & reporting (55–64), customer advisory (65–72), QA of AI output (73–80), prompt craft & Excel (81–90), regulatory interpretation (91–100), and governance/agents/development (101–105). Plus a BONUS FAQ answering the most common live questions (will we be replaced, generative vs. agentic AI, agent vs. skill, tool selection, data security with US vendors, the skills that actually matter). Works with M365 Copilot, Copilot Chat, ChatGPT, Claude, Gemini, and Vibe (formerly Le Chat) — with or without a license. Explicit data rule of thumb for a regulated sector: customer, personal, and market-sensitive data only in enterprise-approved tools (typically M365 Copilot or Copilot Chat with enterprise data protection); if in doubt, anonymize or ask compliance. Built directly on 207 wishes and 70 questions participants wrote live during the webinar. Downloadable PDF (40 pages).
- [Hermes Agent i praksis: Feltrapport fra tre måneder med en lokal open-source agent (Serie 4/4)](/indsigter/hermes-agent-i-praksis): Part 4 and the final installment in the four-part series on AI agents anno 2026. Three months with Nous Research's open-source Hermes Agent running locally on a Mac, configured with OpenAI Codex / GPT-5.5 as model via ChatGPT Plus and Claude Sonnet on OpenRouter as backup. Opens with the morning Hermes interpreted "renew the access token automatically" as "rewrite your own source code" — it edited two of its own core files, restarted its engine, broke every model call, and locked itself out (the token wasn't even expired). Covers what Hermes uniquely is among the four tools (persistent cross-session memory, runs autonomously in the background, learns by saving reusable skills, you pick the model), what works hands-on (a daily AI-news digest delivered to Telegram running flawlessly, content drafts in the user's voice via shared skills with Claude Cowork, scheduling as the strongest capability), what the community uses it for (dev/ops routines, market monitoring, change-only alerts, local language/industry packs, parallel research, agents maintaining their own skills), and the honest dangers — Hermes can modify its own files and restart itself, so it can break itself in ways Cowork and Perplexity cannot. The "AI wingman" recovery pattern (a separate Claude Cowork project with read access to Hermes' source code) and the human-in-the-loop guardrail written into the agent's own instructions. The pivotal distinction: "runs locally" does NOT mean "data stays on your machine" — Hermes itself runs locally but the model call still leaves your machine unless you use a local model via Ollama (with quality/speed trade-off). Cloud-model traffic light: GREEN for public material, sector research, own notes, content drafts; RED for client data, personal data, confidential documents (use local model or keep out). Who it's for: technically comfortable users, developers, power-users — NOT non-technical colleagues (use Cowork or Perplexity instead). 5-step starter guide: run somewhere you can spare, turn approval on, give it one channel and one purpose, pick the model deliberately, always read after. Closes the series with "Four tools, one responsibility" — OpenClaw the swarm, Cowork the supervised assistant, Perplexity the research specialist, Hermes the autonomous one — and the ceiling-is-rarely-the-tool meta-conclusion. Downloadable PDF (Danish, 7 pages). Featured.
- [Perplexity Personal Computer i praksis: Feltrapport fra en keynote-forberedelse (Serie 3/4)](/indsigter/perplexity-personal-computer-i-praksis): Part 3 of the four-part series on AI agents anno 2026. Field report from preparing a 60-minute keynote for a large Danish infrastructure CIO and his 65-person IT unit, where Perplexity Personal Computer delivered a 13-section sector research report with 14 documented cases (Deutsche Bahn, Italferr, DSB, SMRT Singapore, Cubic + Imperial AIDA Lab and others), a maturity matrix on two axes, an EU AI Act timeline, client-specific integration, and "traps to avoid" talking points — all from a single prompt that was itself drafted by Claude. Covers Local Access vs Local Processing (a critical compliance distinction), the handover-file trick for moving research artifacts between agents, what Perplexity can and cannot do, and where the data-residency question gets hard for sensitive EU data. Downloadable PDF (Danish). Featured.
- [AI-agenter anno 2026: Overblik over de fire værktøjer (Serie-hub · Overblik)](/indsigter/ai-agenter-anno-2026-overblik): Series hub / overview for the four-part series "AI-agenter anno 2026". A 14-page deep dive into the four tools that are all called "AI agent" but solve very different problems: OpenClaw, Claude Cowork, Perplexity Computer / Research and Hermes Agent. Two axes (control: own the stack vs vendor rails; maturity: ready for real use including security/attack surface), a side-by-side overview table (what it is, best for, main caveat for each tool), and four cut-through questions for choosing: (1) non-technical colleagues fast → Claude Cowork; (2) research to documents/slides → Perplexity; (3) own the stack with technical muscle → OpenClaw or Hermes; (4) enterprise rollout with compliance → none plug-and-play yet, watch Copilot Cowork. Frames the capability-gap principle (small, boring, repeating task with clear input/output/limits beats building something big right away) and the "expensive mistake" of using these tools in the wrong organizational context. Entry point that links to the four hands-on deep-dives. Downloadable PDF (Danish, 14 pages). Featured.
- [Claude Cowork i praksis: Feltrapport fra min daglige brug (Serie 2/4)](/indsigter/claude-cowork-i-praksis): Part 2 of the four-part series on AI agents anno 2026. Field report from months of daily use plus a compliance assessment for an audit firm. What works (folder cleanup, presentations, Excel analyses and data enrichment, Scheduler), what disappoints (Dispatch), what it can't do (5% embarrassing errors — "PhD with the spelling of a 6-year-old"), the new problem (Cowork touches your real files — a compliance traffic light: GREEN internal docs / YELLOW client deliverables with data minimization / RED sensitive personal + KYC). Consumer vs commercial terms breakdown, browser/computer control caveats, EU Data Boundary status for Claude in Microsoft 365, and the Claude-Work hub folder + CLAUDE.md pattern that makes outputs dramatically more "you". Downloadable PDF (Danish). Featured.
- [AI-agenter i praksis: Feltrapport fra min egen agent-sværm (Serie 1/4: OpenClaw)](/indsigter/ai-agenter-i-praksis-openclaw): Part 1 of a four-part series on AI agents anno 2026 — covering OpenClaw, Claude Cowork, Perplexity and Hermes. Field report from a four-agent swarm (orchestrator, researcher, validator, copywriter) on an old gaming PC delivering three AI news stories to Telegram every morning at 7:30. What works (role separation, orchestrator as quality gate, built-in skeptic, "no surprises" as the key signal) and what doesn't yet (autonomous all-rounders, ~20% malicious ClawHub plugins, governance at scale, agents without humans in the loop, agents as projects). Demo-vs-production table, Royal Unibrew case, capability gap framing, three concrete next steps. Downloadable PDF (Danish). Featured.
- [Din egen lokale HERMES AI-agent](/indsigter/din-egen-lokale-hermes-ai-agent): Step-by-step guide for non-coders to install Nous Research's open-source Hermes Agent on a Mac or Windows PC (via WSL2), connect it to a model via OpenRouter (Claude Sonnet 4.6), set up a Telegram channel so you can message the agent from your phone, and bridge it to Claude Cowork via a symlink so both agents share the same skills. Includes the `hermes config set OPENROUTER_API_KEY` fix for invisible-whitespace key errors and a full troubleshooting table. Downloadable PDF (Danish).
- [10 Copilot-prompts der faktisk virker](/indsigter/10-copilot-prompts-der-virker): Mini-guide on Microsoft's own four-element prompt framework for M365 Copilot — goal, context, source, expectations — plus 10 ready-made copy-paste templates across email & communication (long-thread reply, hard message, stakeholder status), meeting prep & minutes (briefing, action-point minutes, agenda), data analysis & Excel (sheet analysis, formula/pivot), and presentations (first draft, sharpening). Before/after example, tips for Think Deeper mode, and how to save best prompts in M365 Copilot. Downloadable PDF (Danish).
- [Judgment Loop'et: Beslutningsret i enterprise AI](/indsigter/judgment-loopet-beslutningsret-enterprise-ai): A leadership frame for AI governance. "Human-in-the-loop" sits in every policy but defines nothing — this article introduces the Judgment Loop, a four-question diagnostic (what does the model decide, what does the human decide, what does the human need, what happens on disagreement) that separates real human oversight from compliance theater. Downloadable PDF (Danish).
- [Frontier-AI til enterprise: GPT-5.5 vs Opus 4.7](/indsigter/frontier-ai-gpt55-vs-opus47): Mini-guide comparing the two new frontier models from OpenAI (GPT-5.5 "Spud", April 23) and Anthropic (Claude Opus 4.7, April 16). Side-by-side benchmarks (HLE, SWE-Bench Pro, Terminal-Bench, ARC-AGI-2, CursorBench), token economics, context window, image input, hallucination concerns. Use-case matrix (agentics, code refactoring, knowledge work, image analysis, EU compliance), enterprise readiness checklist, and rollout strategy tips. Microsoft 365 Copilot context (GPT-5.5 Think Deeper since May 8). Downloadable PDF (Danish).
- [Aktivér dit LinkedIn-netværk med Microsoft 365 Copilot](/indsigter/aktiver-linkedin-copilot-m365): Mini-guide on using M365 Copilot Think Deeper to analyze your LinkedIn Connections, Messages and Invitations exports — finds patterns across forgotten contacts, warm leads and decision-makers in 10 minutes. 7 copy-paste prompts for job search, sales, partnerships, network analysis. Downloadable PDF (Danish).
- [Det er ikke X. Det er Y. Sådan stopper du AI's mest afslørende skrivevane](/indsigter/negativ-parallelisme-stop-ai-skrivevane): Mini-guide on negative parallelism — the "it's not X, it's Y" sentence pattern that gives away AI text in every major LLM. Five quick fixes plus a copy-paste system prompt. Downloadable PDF.
- [Copilot Agenter: Byg din første agent i Microsoft 365 Copilot](/indsigter/copilot-agent-guide): Step-by-step guide to building Copilot Agents in Microsoft 365 — Agent Builder, system prompts, knowledge sources (SharePoint, Teams, websites), five ready-to-build agent examples, best practices, and common mistakes. Includes downloadable PDF.
- [10 AI-Vaner der burde sidde på rygraden](/indsigter/10-ai-vaner): 10 practical AI habits for daily work — meeting prep, email summarization, first drafts, brainstorming, fact-checking, and how to build habits using Atomic Habits principles. Includes downloadable PDF.
- [Best Practice Guide: Microsoft 365 Copilot](/indsigter/copilot-guide-best-practice): Comprehensive guide covering Copilot Chat vs. M365 Copilot (paid), slash commands, Work IQ Memory, Think Deeper mode, Researcher vs. Analyst, Teams meeting transcripts, Outlook automation, GPT-5.4 Thinking, and prompting frameworks. 58,000 views, 700+ comments on LinkedIn. Keywords: Microsoft 365 Copilot, Copilot guide, Copilot tips, Copilot Teams, Copilot Outlook, Copilot best practice, Copilot training Danmark.
- [Forget ChatGPT for a Moment: Claude Has Three Versions](/indsigter/claude-tre-versioner): Deep dive into Claude (chat), Claude Cowork (desktop agent), and Claude Code (terminal tool) — practical recommendations for each user type and organizational implications
- [Four Agents, One Gaming PC: How I Get Personalized AI News Every Morning](/indsigter/openclaw-daily-ai-news): Follow-up on OpenClaw — automating a daily AI intelligence cycle with 4 specialized agents (researcher, validator, copywriter, orchestrator) on a gaming PC, delivering personalized news to Telegram
- [Moltbook: When AI Agents Talk to Each Other](/indsigter/moltbook-ai-agenter-kontrol): Analysis of the Moltbook phenomenon and AI agent governance
- [Humans + Machines: AI Transformation](/indsigter/mennesker-maskiner-ai-transformation): Why AI transformation starts with competencies, not technology
- [State of AI 2025](/indsigter/state-of-ai-2025): Comprehensive overview of AI developments and what organizations should do

## Contact

- Email: vinc@itu.dk
- Website: https://aitrainer.dk
- LinkedIn: https://www.linkedin.com/in/stefanovincenti/
- ITU Profile: https://pure.itu.dk/da/persons/stefano-vincenti-2/
- BotTellMe: https://bottellme.com
- Newsletter: https://stefanovincenti.substack.com

## Languages

Content is available in Danish (primary) and English.

## Optional

- Location: Copenhagen, Denmark
- Languages: Danish, English, Italian
- Focus: Enterprise AI training and consulting