Blog
Practical insights on AI/ML deployment, product management, and advanced analytics.
Shifts and Frictions in Enterprises in the Era of GenAI Transformations
Field observations on change management and mindset shifts as enterprises navigate GenAI transformation — the efficiency-vs-innovation gap, technical expertise needs, job security anxiety, and why agile companies have the edge.
Casper Hsieh · 5 min read 
Vibe Jam 2026: A Fun Exploration of AI-Assisted Game Development
Reflections from building Bug Detective for Vibe Jam 2026, including where AI-assisted game development worked, where it struggled, and what long-running coding agents still miss.
5 min read
The Economics of Coding Agents: Does Composer 2 Level the Playing Field for Cursor?
How Cursor’s Composer 2 changes the cost-quality trade-off versus API-priced IDEs, with a weekend stress test across a 50-file refactor, a real API integration, and a Pacman terminal build.
4 min read
The Shift Toward Autonomous Coding: What I'm Seeing in 2026
Six major trends reshaping software development — from workflow optimization and cloud agents to the Ralph Loop and the end of janitorial coding.
Building a Technical Support Bot: Lessons from the RAG Design
Practical lessons from building a production-grade internal technical support bot, covering chunking strategies, hybrid search, rerankers, hierarchical retrieval, query expansion, and instructed retrieval.
Fine-Tuning Lessons from the Trenches: What 30+ Experiments Taught Me About Training LLMs
Hands-on lessons from 30+ fine-tuning experiments with Qwen 2.5-7B, comparing SFT, DPO, PPO, and GRPO across safety, math, coding, and conversational warmth.
From Sequential to Router Agents: Three Architectures for Agent AI Systems
A production journey through three agent architecture patterns -- sequential, multi-agent, and orchestrator (router) -- with real cost-quality trade-offs from building an endurance coaching AI.
How Smart Is a Local LLM, Really? My Weekend Experiment on the MacBook M4 Max
Testing local LLMs on the M4 Max — comparing Ollama and MLX, benchmarking models, and figuring out when local inference actually makes sense.
Claude Code: How Subagents, Tools, and Skills Fit Together
Breaking down the layered architecture of Claude Code — how sub-agents, tools, MCP, and the new skills system work together.
The Next Evolution of AI-Assisted Coding: Specialized Agent with Claude Code Plugins
How Claude Code plugins enable specialized agent workflows — solving context limits, domain knowledge gaps, and quality control in AI-assisted coding.