Two categories, two assistants
The cognitive seam
Knowledge work splits along a seam that is more useful than the usual managerial categories. On one side sits the cognitively demanding work that produces durable value — writing that holds up, code that survives scrutiny, decisions that absorb a real amount of context. On the other side sits the logistical work that keeps an organisation alive — scheduling, drafting status replies, formatting, triage. Both are necessary. Only the first kind rewards uninterrupted attention. The risk in the present moment is that the second category has become so cheap to perform with an assistant that the first category is allowed to evaporate without anyone noticing.
Four philosophies of how to fit deep work in
There are four well-described patterns for accommodating the deep category. The monastic philosophy withdraws almost entirely; the bimodal alternates long stretches of seclusion with periods of full availability; the rhythmic philosophy sets a fixed daily block at the same time each day; the journalistic philosophy slots deep work into whatever opening appears. Each suits a different working life. The rhythmic version tends to be the most realistic default for someone with regular obligations; it requires no act of will because the block is already in the diary. The choice is less important than the act of choosing.
Where an assistant genuinely helps and where it intrudes
Assistants are unusually well-matched to shallow work — bulk replies, calendar arithmetic, formatting, triage, first drafts of routine artefacts. The boundary is more interesting than the rule. Inside a deep-work block, the same assistant can become an active source of distraction, especially in tools that surface notifications, autocomplete suggestions, and review nudges. The honest practice is to use one mode for the shallow work outside the block — fast, conversational, frequent — and a more austere mode inside the block, with autocomplete and notifications muted, used only for retrieval rather than as a partner in the thinking itself.
The flow condition the assistant can quietly collapse
Flow research suggests that durable engagement requires three conditions: clear goals, immediate feedback, and a challenge level that sits a little above current skill. An assistant tightens feedback loops, which usually helps. It can also collapse the challenge-skill gradient by autocompleting the productive friction that flow needs. The work feels easier but the engagement evaporates and the output becomes thinner because the worker stopped being asked anything difficult. The design move is to choose deliberately where the assistant should lead and where it should be silent, and to defend the second category.
Look at yesterday in honest detail. Of the work that mattered, how much happened in long uninterrupted blocks, and how much happened in fragments between meetings?