AI in 2026

How it went

We covered the key 2026 trends in language models, vibe coding, and agentic AI, including both paid and open-source solutions and their impact on development, automation, and secure in-house deployment.

The workshop focused on the increasing release cadence of language models and major shifts in vibe coding and agentic AI. We compared 2026 progress across commercial and open-source ecosystems. Anthropic was presented as a provisional leader with Opus 4.7 as the strongest vibe-coding model, plus product additions such as Cowork and Design. The Mythos vulnerability model was also discussed, including its withdrawal and the immediate emergence of the open-source alternative OpenMythos. Google broke through with Antigravity, a vibe-coding tool that delivers one of the most realistic "agent team" development experiences, and Gemini improvements such as Scheduled Tasks were highlighted. OpenAI's shutdown of Sora was covered, alongside a brief demo of Seedance 2.0. We also reviewed the release of ChatGPT Images 2, positioned as a strong competitor to Nano banana Pro, the launch of Workspace Agents, and new quiz-generation features. Perplexity's Comet browser improvements were presented in the context of smoother shopping flows and deeper personalization with agentic AI, while Microsoft's ecosystem progress through Copilot integrations was also discussed. On the open-source side, MoE, Mamba, and MTP concepts were highlighted for enabling up to 4x larger models on the same hardware, making them more attractive for cost-efficient, secure in-house use. Open-source diffusion models were shown to have improved text rendering and reduced visual artifacts. The session also addressed Base44 as a no-code option for application building.

Event resources

• Full workshop recording
• Presentation
• LLM model release frequency

Event speakers

• Martin Čejka

  Workshop Lecturer and AI Researcher • Department of Information Engineering

  LinkedInCZU