Thoughts on AI in chemical distribution: what's working, what to skip, and how operators are putting it to use.
AI tools are starting to blur together. NotebookLM still stands out for turning a chemical brochure into a two-host podcast you can interrupt with questions, and the other tools haven't caught up.
I dropped Stepan's surfactant selector guide into NotebookLM and it turned the PDF your reps already ignore into a chat, a quiz, an infographic, and an audio overview I could play in the car. NotebookLM is free.
I asked Gemini to chart five years of EBITDA for the biggest software companies against the biggest chemical companies, and software margins keep widening while chemical margins compress. Software subscriptions, ERP, CRM, analytics, AI pilots, are a growing line item behind that gap.
A reflection on what it feels like to be shaped by forces you don't control, told through the image of a lump of clay under a sculptor's hands. For me the sculptor is Jesus Christ, and the job each day is just to stay soft instead of hardening against it.
When a distributor is tight on margin, I look past pricing to where the data lives, since manual rekeys and confirming stock by phone hide real costs inside line items that all look normal on the P&L. Cleaning up those workflows can move EBITDA 50 to 100 basis points.
Dow cut 4,500 jobs and leaned on AI to cover the gap, and I think that cut goes too deep on technical depth. The formulation knowledge that technical sales reps carry moves to whoever is closest to the customer, so distributors who hire and train real chemistry depth will take accounts from the ones who don't.
Air Canada's chatbot gave a customer wrong bereavement fare information, and a tribunal held the airline responsible for every word on its website, awarding an $812 refund. If you're using AI in sales, support, quoting, or scheduling, every output is a legal exposure to treat like one.
An AI police report tool in Heber City, Utah wrote an official report claiming an officer transformed into a frog, after picking up movie dialogue playing in the background. A reminder that every AI output has a frog hiding in it, so never publish one without a human checking it first.
A few people asked what the old site looked like, so I posted a before-and-after. The new site took three hours to build.
Claude rebuilt my entire website in about three hours of back and forth, copy and code, and saved me roughly $10,000 versus paying a developer. The cost of designing, writing, and deciding has collapsed, and that applies to every industry, chemical distribution included.
For a client research task I ran eight Claude agents in parallel on market data, supplier trends, and pricing, with a ninth agent supervising to flag contradictions and cut anything it couldn't source. The output was clean enough to act on.
A Lincoln quote about spending most of your time sharpening the axe before you swing it, and a quick progress update, one down, nine to go.
