Notes from the margins.

A newly deployed claims-triage agent keeps a diary of its first eight hours, and discovers that the thirty-step process map was being optimistic.

The long road from demo to "production-grade."

Four claims-triage steps, each about 95% accurate. Run them in sequence: how accurate is the whole loop?

Transformation, as drawn versus as observed.

The authors of Reengineering the Corporation reconvene on a 2026 group chat. Posthumously.

An AI cuts one underwriting step from hours to minutes, but the loan still takes 11 days. Why didn't the process get faster?

When the agents are, technically, on it.

An out-of-office reply from your AI copilot, currently buried under 47,000 tokens of someone else's context.

Five questions to gauge how ready your team really is to put an AI agent into production.

On mistaking the model for the durable advantage.

Token costs are like office snacks: technically free, somehow $40,000 a month.

A cheaper model would save 40% per token but cost $2M to migrate to. When is the switch actually worth it?

A candle, an oil lamp, a fireplace, and a single match. What do you light first?

The surgeon says, "I can't operate, he's my son." How? A classic riddle about the assumptions we don't notice.

What Frank Gehry's buildings teach about leadership: the structures that endure are the ones with foundations no one sees.

Three guests, a $30 room, a $5 refund, and a dollar that seems to vanish. Where did it go?

Three environments compete for the same servers, each with a floor and a ceiling. Can every demand be met at once?

Julia Child's fearless experimentation: a lesson in learning in public and getting comfortable being a beginner.

Uptime falls as users climb, and downtime has a price. When does growth start costing more than it earns?

Why pure LLMs hit a wall, and how agent frameworks connect them to the tools, data, and systems that make them useful.

A quiz on the human side of AI adoption: how people really respond when new technology arrives.

Three departments contribute data and each wants a bigger share of the insights. How do you split it fairly?

Record-and-play mimics your exact movements; observation-based telemetry understands the work. Why the difference matters.

What captain Sourav Ganguly understood about leading a team of strong, very different personalities.

A quiz on process mining and task intelligence: what they find, and what they're for.

Five agents, a web of dependencies, some work in parallel. What's the fastest the project can finish?

A quick quiz on what actually makes enterprise data valuable for AI. How many can you get right?

Model drift by the numbers: how fast your freshly deployed models decay, industry by industry.

Asked who reached Everest's summit first, Tenzing Norgay refused to say. A lesson in sharing the credit.

An AI is more accurate the more confident it is. How should that change when you decide to trust it?

Socrates changed thought not by giving answers but by asking better questions. A leadership method, not just a philosophy.

Four digital twins update on different cycles. When do they all sync at the same moment again?

A quiz on how AI is reshaping jobs and skills by 2030. Test your assumptions against the data.

A quiz on why most digital transformations fall short, and the sobering numbers behind the failures.

How Lin-Manuel Miranda built Hamilton by collaborating early and often: creativity as a team sport.

One company waits for perfect automation; the other ships 60% now. Who comes out ahead?

Beaten by Deep Blue, Kasparov didn't sulk. He invented a new kind of chess. A lesson in losing well.

A quick quiz on what really separates an AI agent from ordinary automation.

AI handles 90% of decisions; humans are meant to review 10% but can only reach 8%. Where does the gap go?

Leadership lessons from the cricket pitch: why calm, trust, and resilience decide the close games.