Working Without Rate Limits on Fable 5: Model Routing and Token Budget Strategy
We unpack the Claude Fable 5 workflow tips shared by T3 creator Theo: effort levels, Codex orchestration, model priority in CLAUDE.md, and offloading token-h...
Development
AI Agents That Write Papers - Moving Your Writing Workflow to an Open-Source Skill Package
An open-source project bundles a professor’s paper-writing expertise into a Skill package that works across Codex, Claude Code, and Gemini. We examine how it...
vLLM vs Ollama - Local LLM Inference: When to Choose Which
The conventional wisdom that Ollama is easy and vLLM is fast gets re-examined with 2026 RTX 4090 public benchmarks. At a single user the gap is around 13%, b...
Building Your First AI Agent - Start with a Minimal Guarded Loop
Most introductory guides to AI agents cover four pieces: the LLM brain, memory, tools, and the agent loop. What actually breaks in production is the part tha...
Android Emulators in Containers - Building a Reproducible Device Farm on K8s with docker-android
docker-android packages an Android emulator into a single container and runs it headlessly. We verify image sizes and KVM requirements against the official d...
Building Sovereign AI for Air-Gapped Public Sector: On-Premises LLM Reference Architecture
A guide for government and public sector organizations that cannot use external cloud services to securely operate LLMs on internal GPU infrastructure. Intro...
How a Solo AI Engineer Operates a Stack of 1,620 Skills
1,620 skills, 55 sub-agents, nightly self-evolution, and cost guardrails. Exposing the full operational system that lets a solo AI team maintain a massive au...
Routing 1,600 Skills Without Noise - Operating an AI Agent Skill Ecosystem
Routing design principles learned from running 1,620 skills solo on Claude Code. Honestly sharing SRA (Skill Retrieval Augmentation) + BM25 gate design, desc...
The Real Engineering Behind ‘AI That Works While You Sleep’: A Deep Dive into Opus 4.8 Dynamic Workflows and Parallel Subagents
A viral X article about ‘40 workflows that make money while you sleep’ swept through developer timelines. The passive income framing is overblown, but the un...
GPU Cluster Cost Optimization: Kueue Fair-Share + Gang Scheduling + Scale-to-Zero
How to reclaim tens of millions of dollars annually wasted across three bottlenecks in a 1,000-GPU cluster using Kubernetes-native scheduling.
The Open-Weight Model That Matches GPT-5.5 at One-Sixth the Cost: A Self-Hosting Analysis of GLM-5.2
Z.ai has released GLM-5.2, a 744B MoE coding model under the MIT license. Reports indicate it surpasses GPT-5.5 on SWE-bench Pro and Terminal-Bench at roughl...
What Is Agent-Native Cloud: Skills and Policies as First-Class Resources
Why VM-centric cloud is ill-suited for autonomous AI agent operations, and the design principles of agent-native infrastructure that treats Skills, Tools, Po...
Treating Skills Like Software: A Hands-On Verification of yao-meta-skill v1.1.0
yao-meta-skill, an open-source meta-skill rumored to be more powerful than Anthropic’s official Skill-creator, gets cloned into the ThakiCloud environment an...
Controlling Image Style with Structured Prompts: The 5-Section Layered Prompt Technique
An analysis of a structured-prompt technique that turns a travel photo into a Studio Ghibli style animation. We break down how to separate subject, backgroun...
Nature-Grade Figures and Polishing as Code: A Hands-On Report on Running nature-skills
We cloned nature-skills, an open-source Claude skill package that bundles Nature-journal-grade scientific figure generation with academic polishing, and used...
PDF to Markdown Without an LLM: LiteParse, RAG Ingest Cost, and Data Sovereignty
LiteParse, released by LlamaIndex, is an Apache 2.0 open-source parser that converts PDFs to markdown without an LLM. We look at the cost and data-sovereignt...
Cutting Token Cost by 34-71% with Reversible Compression: A Headroom Field Report and ThakiCloud Context Hygiene
The biggest hidden cost of an AI coding agent is context. We ran Headroom (headroom-ai) directly against three real JSON tool outputs from the ThakiCloud rep...
Running Ollama on Kubernetes in Production
Practical patterns for operating Ollama as an LLM serving layer in a K8s cluster rather than as a local experimentation tool.
GPU Workload Preemption with Kueue: ClusterQueue Design and Priority Patterns
Kueue’s preemption mechanism and ClusterQueue design principles from the perspective of operating a real AI/ML platform.
AI Coding Era Cultural Innovation: Paradigm Shift from Resistance to Collaboration
How to embrace and develop the new development culture brought by Vibe Coding and Agentic Coding? A guide to building collaborative culture with AI, breaking...