Practical thinking on how companies should actually use AI — and what that means for investors.
Sharper essays, operating playbooks, and thesis notes at the intersection of AI agents, real companies, capital allocation, and physical constraints. The commodities notes aren’t a hobby drift: they trace the AI buildout downstream — agents → data centres → power → uranium and copper.
The useful question is not “can AI do this task?” It is: can a company redesign a workflow so that agents, people, data, and controls produce a better economic system?
I write from the operator-consultant perspective: where the technology really works, where adoption gets stuck, and where the second-order market opportunities appear. The Proof Index maps the whole system; the Hermes case study explains the agent workflow behind it.
The adoption framework below is the spine of the audit I run for clients → services.
Workflow selection, control points, review gates, and the org-design choices that separate compounding adoption from abandoned pilots.
What an AI operating system looks like for companies that do not have research labs: inbox, CRM, documents, finance ops, reporting, and knowledge workflows.
Why the investable question is not only model quality, but distribution, workflow ownership, proprietary data, trust, compliance, and switching costs.
A diligence checklist for separating wrappers from workflow ownership: user pain, data rights, gross margins, implementation burden, and defensibility.
A field note on copper’s structural demand shock, hidden supply bottlenecks, and why Ivanhoe Mines is a clean expression of the thesis.
A trial field note on uranium, energy security, supply fragility, and the danger of being right on a commodity thesis but wrong on the expression.
How regulation, distribution, banking rails, treasury management, and cross-border settlement could create European opportunities.
Start with a costly repeated workflow, not a model. Map inputs, decisions, exceptions, tools, and owners.
Agents need useful context: documents, CRM, inbox, calendars, databases, and permissioned company knowledge.
Define what the agent can do alone, what requires review, and where audit logs or approvals are mandatory.
Measure time saved, throughput, error reduction, sales conversion, working-capital impact, or faster decision cycles.
The winning companies will redesign roles around agent leverage rather than bolt chatbots onto existing processes.
Look for companies that own distribution, proprietary workflow data, trust, compliance, and the system of record.
This framework is productised as the AI adoption audit.
That’s the work I do. If you’re wrestling with adoption, diligencing an AI opportunity, or thinking through the market implications — I’d be glad to speak.