Skip to content

AI Status

Current focus

AI Foundations and the full Agentic Engineering sequence are complete at first-draft depth. All currently published AI learning material now has aligned English and Hungarian variants. There is no forced next workstream yet; RAG, MCP and Memory remain the main candidates.

Current state

  • AI Foundations — 14/14 detailed first drafts complete in EN/HU.
  • Agent Skills — 12/12 detailed first drafts complete in EN/HU.
  • Agentic Loops — 12/12 detailed first drafts complete in EN/HU.
  • Agent Architecture — 12/12 detailed first drafts complete in EN/HU.
  • English is the default document form; Hungarian uses .hu.md sibling files and is rendered through the MkDocs language switcher.
  • RAG, MCP and Memory remain separate deeper workstreams that integrate with the agentic architecture through explicit boundaries.

Decisions

  • Keep the progression Foundations → Skills → Loops → Architecture as the base mental model.
  • Keep English and Hungarian learning pages conceptually aligned when making future changes.
  • Treat AI components as parts of normal application architecture rather than exceptions to software-engineering rules.
  • Keep probabilistic semantic decisions inside deterministic runtime/application boundaries for state, authorization, budgets, side effects, retry and lifecycle.
  • Use modular monolith, Hexagonal/Ports & Adapters, Clean Architecture and bounded-context thinking where they improve maintainability and replaceability.
  • Organize systems around business capabilities rather than model/vector-store/MCP vendors.
  • Keep LLM, retrieval, MCP/connectors, databases, queues and other infrastructure behind explicit ports/adapters where practical.
  • Context is a purpose-specific projection of canonical sources; it is not the execution-state database.
  • Distinguish tools, capabilities, skills, workflows and agents instead of collapsing them into one runtime concept.
  • Prefer deterministic workflows with bounded agentic islands when control flow is known.
  • Multi-agent architecture must solve a concrete separation/scaling/trust problem rather than serve as an "advanced" default.
  • Use retrieval for reference knowledge and tools/application services for current operational state.
  • Evaluation, observability, security and reliability are architecture concerns from the start.

Completed

  • AI Foundations topics 1–14 — bilingual first detailed drafts.
  • Agent Skills topics 1–12 — bilingual first detailed drafts.
  • Agentic Loops topics 1–12 — bilingual first detailed drafts.
  • Agent Architecture topics 1–12 — bilingual first detailed drafts.

Next steps

  • Review/refine completed topics as questions arise.
  • Keep EN/HU variants synchronized.
  • Add implementation exercises/examples to validate the mental models.
  • Choose the next detailed AI workstream: RAG, MCP or Memory.