Work
Case studies and writing
Projects that shipped to production and notes on building AI systems.
Case studies
AI Keyword Classifier
Rules-first AI classification - 50 hours/month saved, <$1 per run
AI Translation Pipeline for Pharma
AI translation integrated with Veeva Vault - 2+ week turnaround reduced to hours
Enterprise Data Governance Transformation
€400K engagement, 25+ stakeholders, 7 data domains mapped across 13 countries
Multi-Agent FDA Document Review
6-agent AI system for FDA document review - 60-70% time savings, multi-million USD ROI
GSC → BigQuery Pipeline
60 customers, 53M+ rows, solving GSC's data retention limitation
When to Halt a Migration
400,000+ documents profiled, 92% scope reduction, migration halted - sometimes stopping is the right answer
Product-Led Growth Engine
0 → 1,000+ users, 100% retention, 8-system commercial stack built from scratch
Writing
The Harness or the Model?
I gave a cheap AI model a better harness and it stopped writing broken code. Then it told me the work was finished when it wasn't. Building the same feature four ways taught me which failures the tooling owns, which belong to the model, and which one you can never buy your way out of.
What an Agent-Hour Actually Costs
The claim going around says Fable 5 bills $43/hour, more than the average developer. I metered ten weeks of my own AI usage to check: $37,000 of work that would have cost $210,000 done badly, and $109,000 on the wrong model. The bill doesn't measure the model. It measures the operator.
Automating Work with AI: From Chat to Pipeline
The leap from using AI in a chat window to wiring it into a workflow that runs without you. Triggers, AI steps, actions, and where to keep a human in the loop.
Building Your Own AI Agents: Templates and Patterns
Once you delegate the same task more than twice, stop re-prompting and define an agent. The anatomy of a reusable agent template, and the patterns worth copying.
Using GitHub as a Knowledge Base for Your AI Tools
A well-structured GitHub repository gives your AI tools a searchable, versioned, plaintext knowledge base. Here's how to set one up.
Memory, Project Files, and Retrieval: Which One Do You Need?
There are three ways to get information into your AI tools. Each does something different. Here's how to pick the right one.
Why Your Prompts Aren't Working (And How to Fix Them)
Most AI prompts fail for the same reason: the person writing them already knows the context. The AI doesn't. Here's the framework that fixes it.
Scheduling AI Work: Recurring Tasks and Overnight Agents
The difference between AI that responds when you ask and AI that runs on a timer. Recurring digests, watchers, and overnight batch work - plus what's safe to run unattended.
What Are AI Skills (and Why They Beat Re-Prompting)
A skill is a named, reusable capability you invoke instead of re-explaining the same task every time. Here's the shift from prompts to skills.
What is Generative AI?
Generative AI explained for first-timers: what it actually is, how it differs from search, and the mental model that makes everything else click.
What is Git, and Why Does My AI Use It?
If your AI coding tool keeps mentioning branches and pull requests, here's the mental model you need - no terminal required.
Code Review Is Becoming Work Review
As agents generate more code, human judgment needs to move from reading every line to reviewing intent, constraints, permissions, verification, rollout, and outcomes.
Software inventory is cheap. Organizational context is not.
If AI makes software cheaper to generate, the durable value moves to distribution, trust, domain context, governance, and continuity.
How I ship with agents
Why serious agentic coding needs gates, independent review, shared work state, and handoffs that survive the chat window.
AI Brain Fry Is Real - But It's Not the Tools' Fault
Harvard Business Review says AI is frying workers' brains. My data - 1,986 commits and 1,900 AI sessions in 76 days - shows the opposite is possible. The difference isn't the tools. It's whether anyone invested in actually understanding them.
Your AI Strategy Is Collecting Dust
Most AI strategies fail before implementation starts. Not because the ideas are wrong - because the strategy was built for a board deck, not for the people doing the work. Here's how to build one that actually gets executed.
Your agents know exactly what you tell them
Most people's AI instructions are empty or stale. The ones getting consistently better results are transferring judgment, not just preferences - and they have a system for keeping it current.
Meta Prompting: Let AI Write Your Prompts
The most underused technique in AI: using the model to improve your inputs. How to generate, critique, and refine prompts using the AI itself.
What Are AI Agents, Actually?
An agent is a system where the model decides the next step. That's it. Most things called 'agents' aren't - they're workflows with LLM-powered steps, which is usually the right architecture anyway. Understanding the actual distinction helps you buy smarter and build better.
What Are LLMs, Actually?
Large Language Models explained without hype or jargon. The 4,096-dimensional mental model that explains hallucination, context windows, and how to use them well.
How I Learned to Build
From port forwarding at eight years old to building AI systems for pharma - a twenty-year journey of refusing to accept limitations.
What Are Workflows and Agents
The distinction that matters is simpler than the terminology suggests. In a workflow, you define what happens. In an agent, the model defines what happens. Understanding when to use each - and how to combine them - is most of what separates successful AI implementations from expensive experiments.