How to build an Agent policy
Agents need a policy. Don’t reinvent governance. Apply what you already do with people: review, limit, escalate, grow trust. Reality begins not when the model speaks, but …
Section
Agents need a policy. Don’t reinvent governance. Apply what you already do with people: review, limit, escalate, grow trust. Reality begins not when the model speaks, but …
This is a new talk delivered in December in front of Data and AI leaders in Financial services.
The human is the UI is a useful mental model is this: the vendor’s humans are the user interface.
Most organisations treat governance as paperwork to be completed before an audit. Policies, slide decks, control libraries and “principles” circulate widely. Everything …
AWS Bedrock adds Priority, Standard, Flex. Priority: lower latency, ~60–90% pricier. Standard: predictable baseline. Flex: ~50% of Standard, slower. Anthropic stays …
Show the value, give it context, define good and bad, and explain why it matters.
Most architecture diagrams fail because they show components, not meaning. The FC method fixes this by mapping use cases and shaping components by cost, revealing how …
Governments can’t use cloud discounts that rely on upfront or multi-year commitments—and for good reason. These laws protect taxpayers from political overreach. Cloud …
AI’s future is cache-shaped: reuse answers to cut costly inference, deliver stable, fast, predictable responses.
In London with four young Italians at an AI event, our talk moved from models to Frantoio, olive presses, harvests and villages. It reminded me that culture guides how we …