Generative AI
Practical guide to integrating generative AI into systems — LLMs, RAG, agentic patterns, and the changing role of engineers.
2026
Everyone's Building With AI. Who's Actually Using It?
Building software became nearly free, so we build constantly — and ship into a void. The cost that didn't disappear is everything after the build: users, maintenance, support, attention. A filter for deciding what's worth building, and permission to stop.
From Backend or Frontend Engineer to AI Engineer: A Realistic Transition Guide
You do not need to learn machine learning from scratch. Most of what makes an AI engineer good is engineering you already have — APIs, data pipelines, system design, production discipline. Here is what transfers, what is genuinely new, and a concrete plan to close the gap in months rather than years.
Small Models and Inference Cost Engineering: When a 7B Beats a Frontier API
Frontier models are the right default until you measure your bill. Quantization, vLLM and Ollama, distillation, and honest total-cost math — when a small or local model genuinely wins on cost, latency, or privacy, and when self-hosting is a trap dressed as a saving.
AI Agents in Production: What the Demos Don't Show You
The agent demo takes two minutes and works perfectly. The production version takes two quarters and fails in ways the demo never hinted at — compounding step failures, tool calls that lie, cost blowups from loops, and no way to tell a good run from a bad one. Here is what actually breaks and how teams fix it.
Model Context Protocol (MCP) Explained: How AI Agents Actually Connect to Your Systems
MCP went from an Anthropic open-source release in late 2024 to the industry standard for connecting LLMs to tools and data. Here is how the protocol actually works — hosts, clients, servers, tools, resources, prompts — why OpenAI and Google adopted it, and what to watch out for before exposing your systems to an agent.
Long Context vs RAG in 2026: Why 'RAG Is Dead' Keeps Being Wrong
With million-token context windows in Gemini, Claude, and GPT-4.1, 'just stuff everything into the prompt' sounds simpler than building a RAG pipeline. Here's why retrieval survives — cost, freshness, access control, and the physics of attention — and how long context actually changed RAG design instead of killing it.
Precision vs Recall: The Essential Difference That Shapes RAG Systems
Understanding precision and recall is critical for building effective RAG systems. Learn how these metrics determine retrieval quality, the precision-recall trade-off, and why a RAG system with poor recall misses critical information while poor precision drowns the LLM in noise.
AI Coding Assistance and Agentic Development: Why Engineers Must Evolve Beyond Writing Code
Tools like Claude, Cursor, and Codex are reshaping how software is built. The engineer's real value is shifting from writing code to setting boundaries, questioning AI output, and holding the wider context no assistant can match.