Ryan Kashtan · August 2026
Cowork Is a Bad Name, and the Least of Three Confusions
Cowork, model family, and effort fix different failures. How to diagnose which dial a Claude problem needs, and why autonomy raises the model bar.
Read →Writing
Thinking out loud.
Notes on Claude implementation, what actually works, and why most AI initiatives stall.
Ryan Kashtan · August 2026
The Case for Believing the No-Training Promise
Anthropic, OpenAI, and Google don't train on enterprise data. The contracts, audits, and incentives that make the claim credible, and where the real risk lives.
Read →Ryan Kashtan · August 2026
Claude's Microsoft 365 Connector Now Has Write Access
Claude can now send email, manage calendars, and edit SharePoint files. How admins enable it, and the risks to weigh scope by scope.
Read →John Crowley · March 2026
What Is Claude Implementation (and What It Isn't)
Using Claude and implementing Claude are not the same thing. Here's the distinction that determines whether your team actually gets value from it.
Read →John Crowley · March 2026
The implementation gap
Most businesses have Claude. Almost none have workflows. Here's why that gap exists and what closes it.
Read →John Crowley · March 2026
What makes a good Claude workflow
Not all workflows are equal. The ones that stick share three characteristics.
Read →John Crowley · March 2026
Why we named this Shannon Advisory
Claude Shannon invented information theory. The name is one layer deep — it rewards curiosity without announcing itself.
Read →