3 shown · 3 total
This report maps the content-management landscape in the LLM and agentic era, motivated by the structural ceiling a large (1,000+ page) WordPress site hits: runtime performance, editorial throughput at corpus scale, design drift, and the absence of any machine-readable AI surface. It distinguishes genuinely *AI-native* platforms — where content is structured first and AI assistants, actions, and agents operate *through* that structure with governance and attribution — from *bolted-on* "generate" buttons that paste prose into an HTML blob. Across 21 chapters it surveys every architecture and platform family (headless/composable, Git-based, modern WordPress, AI-native products, visual builders) and every AI-era capability (authoring, inline editing, agentic workflows, structured modeling, design systems, search, media, analytics, SEO/GEO, localization, governance, pricing). The central thesis: four habits — typed channel-agnostic content models, drafts-by-default AI, narrowly scoped MCP surfaces, and a clean agent-readable output pipeline — matter more than the vendor logo, and they are exactly what is worth stealing into a bespoke build.
This report answers one engineering question: how to add a natural-sounding text-to-speech feature that reads arbitrary text well in **both English and Czech** while staying **free or near-free**. The binding constraint is Czech — it eliminates most acclaimed English-only open models (Kokoro, Orpheus, Chatterbox, F5-TTS all lack official Czech), so every option is graded for Czech *separately* from English. **Headline recommendation:** build a swappable `speak(text, lang)` backend with a content-hash audio cache, then prototype on **edge-tts** (Azure-grade neural Czech voices — Vlasta/Antonín — free, no API key) or **Piper** (`cs_CZ-jirka`, MIT, fully local CPU) for English+Czech; for a licensed/commercial-safe upgrade route Czech to **Google Cloud Chirp 3 HD (cs-CZ)** or **Azure Neural** on their perpetual free tiers, with **ElevenLabs** as the premium "best-sounding Czech" path when budget allows.
This report is an end-to-end, stack-agnostic blueprint for replacing a 1,000+ page WordPress estate with a bespoke, owned CMS in which **content is structured data, AI is a first-class participant in authoring/delivery/operations, insight loops are built in, and every page is on-brand-not-identical**. It works through twenty-two chapters — from goals, requirements, and a nine-layer reference architecture, through content modeling, rendering, design-system guardrails, the AI/agentic content pipeline, retrieval (RAG + MCP), search, media, analytics, personalization, migration, SEO/GEO, i18n, security, and DevOps — keeping every layer vendor-neutral so you can substitute your own tools. Its honest closing verdict is contrarian to its own depth: for most organizations in 2026 the right call is to **buy a headless platform plus a thin AI/MCP layer and build only the 10–20% that genuinely differentiates**, because leading vendors have closed the AI-native gap and 5-year TCO punishes becoming your own CMS vendor.