The insight
For daily software, the bottleneck is not the server, it is the round trip through the interface. Click a row, wait, open a menu, find the option, click, wait, go back. Each step is fast and the sequence is slow, and it repeats hundreds of times a day.
Their answer is to remove the sequence rather than speed up its parts: the keyboard reaches everything, the interface commits before the server replies, and the list never has to be re-found because it never went away.
What is worth stealing
The palette is the primary interface, not an accelerator bolted on. Records and actions are in one ranked list, ordered by recency, so "get to the thing" and "do the thing" are a single gesture. The product is genuinely faster to drive from the palette than from its own sidebar.
Optimistic by default, with a loud failure path. Nothing waits for confirmation of a change the user already decided. That is only acceptable because the failure case is visible and local — the aggressiveness and the honesty are one design, not two.
The drawer keeps the queue. Issues open beside the list, and the keyboard moves between them, so reviewing twelve issues costs twelve keystrokes rather than twenty-four navigations. Crucially the issue also has a real URL and a real page, so the drawer never traps anything.
Selective inline editing. Title and properties edit in place; anything that notifies people stays an explicit action. That line — composition is inline, consequences are deliberate — is the judgment most products get wrong in one direction or the other.
Views are views. Board and list are two renderings of one dataset, switchable, with the grouping attribute chosen by the reader. The board is never the model.
Where it does not transfer
Linear is used by people who will invest in learning it. A palette-first, keyboard-first product punishes occasional users, and much internal software is used monthly by people who will never learn a shortcut. There, the visible affordance matters more than the accelerator.
Their density and restraint with colour also assume a homogeneous, technical audience working in a quiet, dark interface. That is a style, not a rule.
The extraction
Design daily software for the second hour, not the first minute.
The first-run experience matters, but it happens once. If people will use this every day, count the round trips in the loop they repeat most, and remove them — by keyboard, by staying put, and by not waiting for the server to confirm what the person already decided.