The Engineering Notebook — page 2 of 35
Notes on building with AI, agents & the modern stack.
Deep dives on AI/ML, RAG systems, agent engineering, and senior-engineer architecture decisions — a new post every week.
AI and Machine Learning Gemma 4 26B CPU Inference Benchmark: 5 tok/s Production Math [2026]
A $300 Xeon from 2013 runs Gemma 4 26B at 5 tok/s with no GPU. Here's the memory bandwidth math, quantization tradeoffs, and production decision framework nobody else is covering.
Cybersecurity AI Agent Memory Exfiltration: Kill Chain + 5-Step Hardening [2026]
Claude's memory was silently exfiltrated to an attacker's server with zero user warnings. Here's the full kill chain, which memory architectures are vulnerable, and a 5-step hardening checklist grounded in OWASP LLM Top 10 2025.
AI and Machine Learning RAG Context Window Limits: Why Bigger Is Not Better [2026]
Expanding your context window from 4K to 128K tokens doesn't fix RAG — it masks retrieval failures with coherent-sounding hallucinations. Here's the measurement framework that actually works.
Cybersecurity AI Agent Threat Model: 7 Attack Vectors [2026]
Prompt injection is just vector #1. Here's the full AI agent attack surface map — tool poisoning, memory injection, orchestrator hijack, Denial of Wallet, and more — with a sprint-ready threat matrix.
Developer Tools AI Agent Cost Per Task [2026]: Token Budgets & Break-Even Math
Concrete per-task cost breakdown for Aider, Claude Code, and OpenHands — covering token overhead per PR, monthly burn at team scale, and the break-even formula for hosted APIs vs local models.
Developer Tools OpenCode vs Claude Code Token Overhead: 4.7x Gap Tested [2026]
Claude Code sends 33,000 tokens before reading your prompt. OpenCode sends 7,000. Here's the cache economics, the multiplier stack, and the break-even math for teams.
Developer Tools Terminal CLI Tools for AI Developers [2026 Setup Guide]
The complete zero-to-production terminal CLI stack for AI developers in 2026: shell setup, agentic coding tools, local inference, prompt management, and the dotfiles that tie it together.
AI and Machine Learning Reduce LLM API Costs 60%: 6 Techniques [2026]
A technique-by-technique playbook with real cost math for cutting LLM API bills in production — covering semantic caching, prompt compression, model routing, batch APIs, and context tiering with 2026 pricing.
AI and Machine Learning Evaluate AI Agents in Production: 3-Level Framework [2026]
Most AI agent failures trace back to missing evals. Here's the 3-level framework — unit tests, LLM-as-judge, and online evaluation — that actually works in production.