The deep end. on Atlas of Work

Introducing Atlas of Work

Mon, 02 Nov 2026 09:00:00 -0800

Ask whether AI can do a job, and the most useful answer often starts with another question: which part of the job?

A title can hide an extraordinary variety of work. The same person may gather information, interpret an exception, coordinate with a colleague, make a judgment, and document a decision. Those activities depend on each other. They also depend on a setting that a short job description rarely captures.

Work is more than a job title

Mon, 02 Nov 2026 08:00:00 -0800

“Accountant,” “engineer,” and “designer” are useful names. They tell us where to start a conversation. They do not tell us everything a person does, what information they need, or how they decide that the work is finished.

When we want to understand where AI might help, those details become the subject. We need a description of the work that is specific enough to examine and connected enough to preserve context.

What does an AI benchmark actually tell us?

Mon, 02 Nov 2026 07:00:00 -0800

A benchmark can make a capability claim more concrete. It gives us an assessment to inspect: a task, a setup, a set of outputs, and a way of judging them.

The result becomes more useful when those details stay attached to it. In Atlas, we think about evidence as something to follow back to its source, with enough context to understand both what it supports and what it leaves open.