Notes on the Work
Analytics, business intelligence, sales automation, and the thinking behind how I work. Most of this is practical. Some of it is about why things break the way they do.
I Almost Lost a $350K Deal Because I Forgot to Text a Guy Back
I built 75 tools across three machines. None of them reminded me to text a prospect back. So I fixed that in two hours with shell scripts and jq.
Read more →Designing for Forgotten Context: Anti-Fragile Patterns for LLM-Native Codebases
A 5-part series on building systems that get stronger — not weaker — the more AI sessions touch them. Structural patterns, interchange formats, and architectural health from a production system with 75+ tools.
Read the series →What This Means for Your Team
Most companies adopted AI coding tools and got a productivity bump. A few companies redesigned their architecture for AI collaboration and got a compounding advantage. The gap between those two groups widens every month. Here's what the second group does differently.
Read more →The Session Problem: Living Documentation and Architectural Health
You built clean architecture. Then thirty AI sessions touched it over two weeks and nobody tracked what changed or why. The meta layer — session hooks, automated reviews, and responsibility monitors — is what keeps LLM-native codebases from decaying into the same mess you designed your way out of.
Read more →Interchange Formats: How Your Tools Should Talk to Each Other
The hardest part of an AI-modified codebase isn't the individual tools — it's what happens between them. Append-only event logs, declared contracts, and subscription-based routing turn integration from a minefield into something an LLM session can get right on the first try.
Read more →One File, One Job: Design Patterns for LLM-Native Codebases
Your AI coding assistant doesn't need better prompts. It needs a codebase designed so that modifying one thing can't silently break another. Four patterns from a production system with 75+ tools: single-responsibility files, config-driven behavior, dispatcher layers, and contracts registries.
Read more →Why LLMs Are Great at Greenfield and Terrible at Your Codebase
Every team using AI coding tools hits the same wall. Greenfield is magic. Edits break things. The problem isn't the model — it's that your codebase was designed for humans who remember, and LLMs don't. This is Part 1 of a series on designing codebases that work with AI, not against it.
Read more →Why Your AI Pilot Never Shipped
AI pilots die in the gap between demo and production. That gap is infrastructure, not intelligence. Here’s why most companies never cross it.
Read →I Built an AI That Writes LinkedIn Messages. Here’s What It Actually Does.
Coldstream writes genuinely personalized outreach and handles the conversation after someone accepts. 50% acceptance rate on Day 1. It’s running right now for paying clients.
Read →What an AI Readiness Assessment Actually Looks Like
Your CEO said “figure out AI.” You have budget and a mandate with no instructions. Here’s what the assessment that actually answers that question looks like.
Read →The Companies Winning With AI Aren’t Firing Anyone
The actual competitive advantage isn’t cutting headcount. It’s keeping the same team and multiplying their output. The gap compounds monthly.
Read →Overlap Isn’t Saturation: How to Read Your Outreach Signal in a Regional Market
When your outreach starts hitting people your prospect already knows, most people panic and switch markets. Here’s why that’s the wrong read — and what the signal actually means.
Read →Why Your Power BI Project Failed (It’s Not the Tool)
Most companies that come to me with broken Power BI environments made the same mistake at the start. It’s fixable — but you have to fix it at the right layer.
Read →Your Sales Team Isn’t the Problem. Your Prospecting System Is.
The most common symptom I see in mid-size B2B companies: salespeople who are good at closing but terrible at getting into the room. The problem is almost never the people.
Read →How to Fix a KPI System That Got Out of Control
Most companies don’t have a data problem. They have a definition problem. Here’s what a broken KPI system looks like and how to actually fix it.
Read →Data Democratization Isn’t a Tool. It’s a Decision.
Every BI vendor sells “data democratization.” Most companies that buy it don’t get it. Here’s what it actually requires — and why the technology is the easy part.
Read →What a Fractional Analytics Consultant Actually Does
The term gets used loosely. Some people mean a part-time hire. Some mean staff augmentation. What I mean is different — and it’s worth understanding the distinction.
Read →Why Your Salesforce Reports Are Lying to You
I’ve worked inside a lot of Salesforce environments. The reporting is almost always wrong, and it’s almost never Salesforce’s fault.
Read →If something here resonates, let’s talk.
Book a 30-minute call. No intake form. Tell me what’s going on.
Book a 30-Minute Call