Today, I had the opportunity to attend Infoshare 2025, and I must say—it felt both familiar and refreshingly new. The event has evolved since my last visit, yet it still retains its vibrant spirit of innovation and community.

One of the highlights was, once again, the expansive Innovation Zone, where startups from various industries showcased their products and ideas. Each one brought something unique to the table, with compelling stories, missions, and visions for the future. It’s always inspiring to see such diversity and creativity in one place.

The career and recruitment section also stood out. While many companies from the Tricity area were present, I was excited to see a noticeable presence from Katowice and the Silesian region this year. It’s fantastic to witness companies from the south of Poland stepping into the spotlight and connecting with the broader tech community.

---

My Focus: The AI Track

This year, my main goal was to explore the cutting edge of artificial intelligence by attending as many AI-dedicated lectures as possible. Here’s a recap of the top 10 sessions that stood out:

---

1. The Productivity Trap: Meet the Perils and Promises of AI-Assisted Coding

This talk delivered a reality check on the excitement around AI-assisted development. Adam presented insights from analyzing over 100,000 AI-generated code refactorings, revealing that speed does not always equal productivity. He emphasized the risks of technical debt, code correctness, and provided strategies such as automated guardrails to make AI-assisted development sustainable.

---

2. GenAI bez hype’u – Designing AI That Makes Sense for Organizations

Presented by mBank, this talk was a refreshing break from AI buzzwords. Their philosophy? “Najpierw sens, potem technologia” — “Purpose first, then technology.” They focus on solving real business problems rather than adopting AI for the sake of trendiness. A pragmatic, business-driven approach to GenAI.

---

3. Human Robot Agent: Redesigning Work in the Age of AI

This session explored how AI is transforming leadership, work structures, and skills. Drawing from the book Human Robot Agent, it laid out strategies to thrive in a dynamic, AI-enhanced future — with an emphasis on multi-skilled talent, adaptive organizations, and ethical innovation.

---

4. Applying Agentic AI to Process Millions of Lines of Code in One Day Using Private AI

A glimpse into what’s beyond GitHub Copilot. This talk highlighted agentic AI systems that can:

• Write unit tests across millions of lines of code,
• Auto-generate user stories from business docs,
• Perform mass refactoring,
• Summarize codebases,
• Analyze logs at scale,
• And even prototype features.
  All done offline, with low resource usage and high privacy — an impressive use of autonomous AI agents.

---

5. RAGi, tagi i inne dragi – The Addiction to Useless Metrics

A humorous yet eye-opening critique of how Retrieval-Augmented Generation (RAG) systems are evaluated. While RAG combines search and generation, evaluating both accurately is a nightmare. This talk dove into why many existing metrics are misleading, and how to build benchmarks that reflect messy, real-world data — not just pretty slides.

---

6. The Rise of Vibe Coding and the Future of Programming

A fast-paced and entertaining exploration of how AI is changing coding itself. „Vibe coding” — where intuition, AI suggestions, and experimentation matter more than deep understanding — is on the rise. But it also raises big questions: Are junior devs being left behind? Are we truly productive, or just faster? A must-watch for developers navigating this new reality.

---

7. Vibehaking – Security in the Age of Automation

This technical and occasionally terrifying session explored the intersection of AI and cybersecurity. From tools that find and patch vulnerabilities to AI systems capable of writing insecure code, it covered the double-edged sword of AI in security. Full of practical tools, examples, and „gotchas” to watch for in an increasingly automated world.

---

8. AI-Powered Hacking: Deepfakes, Remote Access, and the Future of Cyber Threats

One of the most jaw-dropping talks of the day. Live demos showed how attackers use AI to:

• Create deepfakes on the fly,
• Automate social engineering,
• Launch remote exploits,
• And chain attacks with almost no manual input.
  A stark reminder that AI is no longer just a defense tool — it’s becoming a major offensive force too.

---

9. Avoid Common LLM Pitfalls

An extremely useful deep dive into the real-world limitations of Large Language Models (LLMs). Topics included:

• Reducing hallucinations with grounding and RAG,
• Dealing with outdated or inaccurate info,
• Securing LLMs against prompt injection,
• Managing cost and latency,
• And evaluating output beyond surface-level metrics.
  Great takeaways for anyone building production-grade AI systems.

---

10. Adventures of Integrating Multimodal Search in E-Commerce

This talk from the Vinted team shared lessons from deploying multimodal vector search at scale. It covered architecture decisions (including Vespa), real-world trade-offs, and how to meet tough business requirements while delivering semantic search at billion-scale. An advanced yet practical view into what’s happening behind the scenes of modern AI-powered search.

---

Final Thoughts

Infoshare 2025 delivered in a big way—especially for those of us passionate about AI. The talks ranged from deeply technical to strategically visionary, and they painted a complex picture of both opportunity and risk in the AI landscape.

I’m already looking forward to tomorrow’s sessions, eager to learn even more from this incredible community of innovators, researchers, and builders.

Stay tuned for part two!

---