Writing
From the field
Notes on the substrate the lab is building: retrieval, context, protocol servers, and what it takes to run them inside a real operation.
NetSuite Alternatives for Manufacturers Who Want to Own Their Software
A working comparison of NetSuite alternatives for small and mid-market manufacturers. Cost, lock-in, on-prem options, and the path most lists miss.
RAG vs Knowledge Base: What a Manufacturer Actually Needs
A knowledge base is where answers live. RAG is how they get found. Which one you are missing, and why buying the wrong half is the common expensive mistake.
Lot Traceability Software: What It Has to Do Under Pressure
Traceability is only tested on the day a supplier calls. What forward and backward tracing actually require, where most systems break, and how to check yours.
Outgrowing QuickBooks: What Manufacturers Get Wrong About the Next Step
QuickBooks is rarely the problem. The operational load bolted onto it is. Which signals mean you outgrew it, and why replacing it is usually the wrong move.
ERP Data Migration: Where Projects Quietly Fail
The most underestimated line in any ERP project. What actually goes wrong, the validation step teams skip, and why moving all your history is the wrong default.
How Much Does a Custom ERP Cost?
There is no list price, and anyone quoting one before seeing your operation is guessing. What actually drives the number, and how to work out your own.
ITAR Compliant ERP: What the Requirement Actually Is
No ERP is ITAR certified. The regulation constrains who can read technical data and where it sits. What that means for software, and where vendors overstate it.
ERP for Job Shops: Why High-Mix Work Breaks Standard Systems
Most ERPs assume repeat production. Job shops and machine shops quote one-offs and reroute them every time. Why that breaks, and what to look for instead.
Why ERP Implementations Stall, and What Actually Prevents It
ERP projects rarely fail on the software. They stall on scope, adoption, and a go-live too late to correct. Five failure modes and how to catch them early.
AS9100 Software Requirements: What Your System Actually Has to Do
AS9100 does not certify software. It certifies your process. Here is what an auditor actually asks your system to prove, and where most ERPs fall short.
MES vs ERP for Mid-Market Manufacturers
ERP plans the work. MES runs it. Most mid-market shops evaluating an MES really need the floor data layer underneath. How to tell which you need.
Can I Build My Own ERP? What the Answer Looks Like in 2026
The honest answer for a mid-market manufacturer: cost, timeline, staffing, when a custom ERP wins, when it fails, and what the hybrid path delivers.
Manufacturing Data Sovereignty: What It Actually Means, and Why NetSuite Isn't It
For manufacturers, data sovereignty means 4 things: residency, ownership, portability, extensibility. Big ERP fails all four. Here is what passing looks like.
RAG for Manufacturing: A Knowledge Base Your Shop Floor Can Actually Use
An AI knowledge base for manufacturers, built on RAG. What to ingest, what to skip, how to deploy it on your infrastructure, and how to trust the answers.
Quality Substrates for AS9100 Shops: Turn Audit Prep from Weeks Into Hours
An AS9100 quality manager spends 3 weeks per audit hunting evidence across SharePoint, the QMS, and email. A quality substrate cuts that to hours.
The 6-Question Context Engine Audit (And What Your Score Means)
Six questions that tell you whether your AI system has a real context engine or a pile of half-built parts. Score in 15 minutes. What each band means.
Storage for AI vs Object Storage (S3, R2, GCS): The Comparison Nobody Else Writes
S3, R2, and GCS hold bytes. They do not know what is in them. A storage for AI layer holds the record an agent reads from. When to use each, and both.
Why Manufacturing AI Projects Stall (and the State Layer That Lets Them Ship)
CNC shops and fabricators carry deep state in BOMs, routings, and AS9100 evidence. Stateless LLMs cannot reason against it. The fix is a substrate.
Your AI Demo Worked. Your AI Project Failed. Here's Why.
Frontier LLMs are stateless. Every conversation starts cold. The gap between a demo and a shipped project is the state layer your business does not have.
Every AI Team Builds a Context Engine. Most Don't Realize It.
Every team building with AI rebuilds the same 6 components: connectors, retrieval, storage, review, protocol, drift detection. That is a context engine.
Storage for AI vs Vector Databases: When to Use Which (and How They Work Together)
A storage for AI layer holds the canonical record. A vector database indexes over it. Different problems, different layers. Most systems need both.
The 5-Question Substrate Audit (And What Your Score Means)
Five questions that tell you whether your business has an AI substrate. Score in 15 minutes. What each band means and what to fix first.
Vector DBs Aren't Storage. They're Indexes.
A vector database is an index. The storage sits somewhere else, usually a Postgres table nobody talks about. The missing category is storage for AI.
AI Substrate Glossary: Definitions for the Terms KoldOps Uses
Definitions for AI substrate, storage for AI, context engineering, decision-state, code-state, drift detection, and substrate audit. Refreshed quarterly.
Decision-State, Airlocked to Code-State: Defining the AI Substrate
The AI substrate isn't compute. It's the discipline that fuses business decisions to your codebase with the same git, review, and audit you use for code.
Why Your Manufacturing Floor Still Runs on Spreadsheets
Plenty of manufacturers still manage production data in spreadsheets. Why the spreadsheet survives, what it costs, and what the path out actually looks like.