The Luca Berton Blog
1596+ expert articles on AI, Kubernetes, Ansible, Platform Engineering, and cloud infrastructure — practical, hands-on guides and industry insights from Luca Berton, AI & Cloud Advisor. Explore tutorials, deep dives, and the latest from every topic.
Fish Audio Raises $52M Seed for AI Voice Models
Fish Audio's $52M seed funds AI voice models for creators and enterprises — 8M+ users, $21M ARR, and an open-source Fish Speech lineage.
AI Builders Amsterdam: A Hands-On AI Meetup Recap
AI Builders Amsterdam recap: talks on validating side projects, docs for agents, and running local LLM agents at StartDock.
5 min read
Four UI & AI Tools I'm Watching in 2026
Watermelon UI, Motion Primitives, Manus, and Haikei: four tools I'm watching for building interfaces, motion, autonomy, and design assets.
5 min read
Add a Live QR Code in Word
Use the native DISPLAYBARCODE field in Word to embed a real, updatable QR code that links to lucaberton.com — no add-in required.
3 min read
YAGNI: You Aren't Gonna Need It
YAGNI — You Aren't Gonna Need It — an Extreme Programming rule against building features no current requirement demands. Less code, fewer bugs, faster delivery.
5 min read
Claude Code Usage Limits: Session vs Weekly
Claude Code's Usage page shows two limits: a rolling session and a weekly cap across all models, plus temporary boosts that revert to your plan baseline.
3 min read
Azure Linux 4.0: Microsoft's First-Party Linux in Preview
Azure Linux 4.0 is now in public preview: Microsoft's Fedora-derived, dnf5-powered, secure-by-default Linux built for Azure VMs, AKS, and containers.
7 min read
Model Soups: Better Fine-Tuning by Averaging Weights
Model Soups average N fine-tuned checkpoints into one. No extra inference cost or latency, often beats the single best run. Recipe + PyTorch code inside.
8 min read
Netris Raises $15M to Automate AI Neocloud Networks
Netris raised a $15M a16z Series A to automate AI neocloud networks: hardware-accelerated, deterministic, no AI in the product.
5 min read
Soup: Fine-Tune an 8B Model on a 4 GB GPU
Soup is an open-source CLI that fine-tunes LLMs from one YAML and streams the base model layer-by-layer so an 8B model trains on a 4 GB GPU.
7 min read
Meta Muse Glimmer: A 30B Agentic Model for Your GPU
Meta open-sourced Muse Glimmer, a 30B dense agentic model that runs under 20GB on a consumer GPU via 4-bit quantization and DFlash speculative decoding.
5 min read
Photonic Networks: the Standout Tech at KubeCon Japan 2026
From KubeCon Japan 2026: photonic networks could reshape the AI data center by moving data optically — Fujitsu leads this work in Japan.
2 min read
Hiroshima Peace Memorial Ceremony 2026: Attending in Person
I attended the Hiroshima Peace Memorial Ceremony on 6 August 2026. Here is what the minute of silence, 120 nations, and two hibakusha testimonies taught me.
5 min readCommunity & Takeaways from KubeCon + CloudNativeCon Japan 2026
From KubeCon Japan 2026: the growing CNCF platform-engineering community in Japan, hallway-track moments, and how AI helped me discover volcanic eggs.
3 min read
Kata Containers 4.0 Reinforces AI Agent Sandboxing with a New Rust Runtime
Kata Containers 4.0 ships runtime-rs, a Rust runtime replacing Go, hardening AI agent sandboxing with VM-grade isolation and lower startup latency.
3 min readCloud Native in the Real World: Optics, Subaru & Uber (KubeCon Japan 2026)
From KubeCon Japan 2026: an optics maker and Subaru running AI on cloud-native infrastructure, plus a walk through the Uber booth on the show floor.
2 min readOpen Source Building Blocks for AI Platforms (KubeCon Japan 2026)
From KubeCon Japan 2026: the AI conformance framework for portability, a CNCF+in-house research platform, and dynamic resource composition in CNCF sandbox.
3 min readSecurity, Isolation & Sovereign AI on Kubernetes (KubeCon Japan 2026)
From KubeCon Japan 2026: holding AI isolation without killing efficiency, the hybrid namespace/dedicated-tools pattern, and open-source sovereign orchestration.
3 min read
LocalAI LongCat-Video-Avatar 1.5: Local Talking Avatars
Run an open-source, audio-driven talking avatar on your own GPU with LocalAI 4.7.0 and LongCat-Video-Avatar 1.5 — MIT, Whisper-Large lip sync, no API.
5 min readTuning Kubernetes for AI: the Real Trade-offs from KubeCon Japan 2026
From KubeCon Japan 2026: why AI on Kubernetes needs tuning, not just more GPUs — LLMD caching/routing and the real GPU, memory and electricity cost.
3 min read
Hermes Agent Troubleshooting: Fix Model, Provider, Gateway & Credential Errors
Model 404, provider retries exhausted, silent Discord bot? Here is the diagnostic flow and exact commands to fix every Hermes Agent error.
10 min read
Cloud Native Telecom Meetup Japan 2026 at NTT DOCOMO Open Lab Odaiba: My Recap
My recap from the Cloud Native Telecom Meetup Japan 2026 at NTT DOCOMO: docomo SOP-grounded agentic AIOps and graduated autonomy.
7 min read
Claude Code login: Unified Auth Hub & Opus 5
Claude Code v2.1.219 turns `claude login` into a unified auth hub for GitHub, GCP, and npm, ships Opus 5 by default, and adds a sandbox network allowlist.
3 min read
Codex Device Code Auth: Enable It in ChatGPT Security Settings
Codex login fails until you enable device code authorization in ChatGPT Security Settings. Here is the exact fix, including managed workspace steps.
3 min read