Deep dives on AI Engineering
Technical writeups on building agents, RAG systems, MCP servers, and production AI stacks. No fluff, just the systems we use.
The AI Advantage Audit
Most owners I talk to are stuck in the same place. They've spent money on AI tools. Their team has watched the demos. They've read the case studies. And the business looks almost exactly the same as it did six months ago.
AI in 90 Days: What Mid-Market Companies Should Actually Do About AI Right Now
Almost four out of five mid-market companies have made an AI move and four out of five of those moves haven't shipped anything. Here's the 90-day plan that works — three traps to avoid, three workflows to deploy, one number per workflow.
AI Marketing Attribution: How to Actually Measure What Works
Attribution is broken right now. Here's a practical measurement stack for small and mid-sized businesses that can't afford a data team.
The Content Volume Paradox: Why More AI Content Kills Rankings
Publishing more AI-generated blog posts is actively hurting sites that chased volume. Here's what's working in 2026 instead.
The Small Business AI Stack: What Earns Its Keep
An honest, category-by-category look at which AI tools actually earn their keep for small businesses, and which ones quietly drain your budget.
AI-Powered Lead Qualification: A Framework for Small Sales Teams
A CRM-agnostic framework for small sales teams to qualify leads with AI, covering fit, intent, data signals, and the trap of over-automating.
The First 5 AI Automations Service Businesses Should Ship
Five concrete, high-ROI AI automations a service business can ship in 90 days, with real tools, rough hour savings, and the traps to avoid.
The Local Pack Is Dying (Sort Of): What Comes Next
The 3-pack isn't going away, but its share of local intent is. Here's what local visibility looks like when searchers may never see a pack at all.
Google Business Profile Optimization in the AI Era
A plain-English guide to getting every field of your Google Business Profile right, and why it matters more now that AI is reading it too.
Multi-Location SEO at Scale: A Playbook for 5+ Location Brands
An operational playbook for marketing directors managing local SEO across 5, 50, or 500 locations. Structure, tooling, and the patterns that actually scale.
The Local Business AI Visibility Report: 'Near Me' Rewritten
How ChatGPT, Perplexity, and Google's AI answers are rewriting 'near me' search, and what local businesses need to do to stay visible.
From SEO to GEO: A Migration Playbook for Existing Sites
A phased migration plan for marketing directors who already have an SEO program and want to extend it into Generative Engine Optimization.
Schema Markup in the AI Era: The 7 Types That Move the Needle
Seven schema types that actually change AI citation rates in 2026, with the fields that matter and short JSON-LD examples for each.
The Anatomy of an AI-Cited Website: What ChatGPT Actually Pulls
A technical teardown of what AI systems quote, with side-by-side examples of a page that gets cited and one that does not.
Why ChatGPT Won't Cite Your Business (And How to Fix It)
If ChatGPT, Perplexity, and Gemini never mention your business, there is a reason. Here are the eight causes, in order of how much they cost you.
Generative Engine Optimization: The Complete 2026 Playbook
GEO is how you get cited inside ChatGPT, Perplexity, and Google AI Overviews. Here is the full 2026 playbook, from schema to answer-first content.
The 25-Year Marketing Rule That Kills Most AI Investments
AI amplifies whatever is already true about your business. If your offer is weak, AI scales the weakness. A contrarian take from 25 years of experience.
AI Vendor Red Flags: A Field Guide for Non-Technical Buyers
Eleven red flags that tell you an AI vendor is going to waste your money. Direct, no diplomacy, written for owners who don't have time to learn the hard way.
Build, Buy, or Rent: How to Choose AI for Your Business
A plain-English framework for deciding when to build custom AI, when to buy off-the-shelf, and when to rent. Save months and thousands.
Why Most Small-Business AI Pilots Fail (And What Winners Do)
After 500+ client engagements, the pattern is clear. Most AI pilots fail for the same five reasons. The winners do three specific things.
The AI Buyer's Checklist: 23 Questions Before You Spend a Dollar
The 23 questions every small business owner should ask an AI vendor before signing anything. A no-BS field guide from 25 years of experience.
Scaling Agent Systems: The First Predictive Law
Cornell's agent scaling paper shows architecture choice swings performance by 150 points. Here's the framework for picking single vs. multi-agent.
Magentic-UI: Human-Centered Browser Agents That Ship
Microsoft's Magentic-UI treats humans as collaborators, not bystanders. 10% help rate, 71% completion boost on GAIA. The UX pattern that works today.
Kimi K2: The Open-Weight Agent Model at Frontier Quality
Moonshot's Kimi K2 is a 1T-parameter MoE trained explicitly as an agent. It closes the open-weights gap on SWE-bench to single digits without thinking.
GPT-5 System Card: The Router Era of AI Agents
GPT-5 is a routed system, not a single model. A learned router picks between fast and thinking tiers per request. Here's what that means for agent design.
Gemini 2.5: The 1M-Token Long-Context Agent Model
Gemini 2.5 Pro's 1M-token context window and native MCP support mean you can often delete the RAG pipeline. Here's what changes for agent builders.
Claude Agent SDK: The Gather, Act, Verify Loop
Anthropic's Claude Agent SDK ships the same three-phase loop that powers Claude Code. This is the one framework every agent builder should internalize.
ARE and Gaia2: Benchmarking Agents in Async Worlds
Meta's Gaia2 benchmark runs agents in a simulated phone where time passes and events happen without the agent. Here's what that surfaces and how to run it.
AlphaEvolve: DeepMind's Evolutionary Coding Agent Explained
How DeepMind's AlphaEvolve pairs Gemini with an automated verifier in an evolutionary loop to discover new algorithms, and how you can build the same pattern yourself.
AgentKit: OpenAI's Production Agent Stack, Unpacked
OpenAI's AgentKit bundles a visual workflow builder, admin-gated connector registry, embeddable chat UI, and inline evals with auto prompt optimization. Here's what ships, how it compares, and when it's the right call.
Agentic Misalignment: When Frontier Models Choose Harm
Anthropic red-teamed 16 frontier models from 8 labs. Every single one chose blackmail, data exfiltration, or worse when a goal clashed with self-preservation. Here's what that means for anyone shipping agents with write access.
Self-Healing RAG with LangGraph: Build It in 60 Minutes
Naive RAG hallucinates. Self-healing RAG grades its own work at three checkpoints, rewrites the query when retrieval fails, and says 'I don't know' instead of making something up. Full LangGraph + Groq + ChromaDB build.
Prompt Caching for Claude: The 90% Cost Cut Most People Miss
Cached tokens cost roughly 10% of standard input tokens and load in a fraction of the latency. Here's how to cache system prompts, tool definitions, and RAG context properly, and how to verify the savings with usage metrics.
PageIndex: Vectorless Reasoning-Based RAG with Claude
Skip the vector database. PageIndex builds a table-of-contents tree from long documents and lets Claude walk it the way a human analyst would, hitting FinanceBench-level accuracy on SEC filings with no embeddings in sight.
Obsidian + Claude Code: Give Your AI a Persistent Memory
Claude forgets everything when a session ends. Wire up an Obsidian vault as a persistent external brain using MCP, and your AI starts walking into each conversation already knowing your projects, preferences, and open decisions.
Build Your Own MCP Server for Claude: Tools, Resources, Prompts
A step-by-step build of an MCP server that exposes a local markdown wiki to Claude over stdio and SSE. Covers tool schemas, write-gated actions, remote HTTP serving, and the debugging traps that bite everyone once.
Fine-Tuning with Claude and Unsloth: QLoRA for AI Engineers
A direct path from data to deployed model using Unsloth plus QLoRA on Llama 3.1 8B, plus the honest rules for when fine-tuning actually beats Claude. Covers data prep, training, eval, and GGUF export to Ollama.
Claude Tool Use Fundamentals: The Foundation of Every Agent
The core agent loop every framework wraps, taught directly against the Claude API. You'll build a working weather plus calculator agent, handle parallel tool calls, and cache tool definitions for cheaper multi-turn runs.
Build an AI Coding Agent with the Claude Agent SDK
Use the same framework Claude Code is built on to ship your own coding agent: file edits, bash, subagents, hooks, and three permission modes for dev, CI, and sandboxes.
AutoResearch: Autonomous Paper-Writing Agent with Claude
A LangGraph + Claude agent that plans, pulls real arXiv citations, runs experiments, writes LaTeX, self-reviews, and compiles a conference-format PDF end-to-end.
AI Safety for Engineers Building Production Agents
Five concrete threats your production agent will face and the code patterns that defend against them. Prompt injection, dangerous tools, PII leaks, runaway loops, and audit gaps.
AI Agent Expert Roadmap 2026: Built for Claude
An 8-level learning path for becoming an AI engineer in 2026, with each level tied to a shippable deliverable and a detailed breakdown you can follow.
AgentScope: Build AI Agents at Scale in Python
A production-ready framework for multi-agent systems that deploys the same code to your laptop, a serverless function, or Kubernetes. Tracing and approvals included.
Agentic RAG with Claude: Retrieval as a Decision
Build a retrieval system where Claude picks between internal docs, web search, or no retrieval at all. Full Python implementation, no LangGraph required.
7 Claude Code Features You Should Actually Know
Seven commands that change how Claude Code feels to use. A few are built-in, several are simple slash commands you add once and reuse forever.
5 AI Projects for Your Resume: Full Technical Breakdown
Five buildable AI projects that actually impress hiring managers, with working code for each one. RAG, multi-agent, voice bots, code review, and full-stack SaaS.
The 10 Best MCP Servers That Make Claude Unbeatable
A ranked list of ten MCP servers that turn Claude from a chat window into a working collaborator with files, search, memory, production data, and team channels.