Ending index or upper bound: k

The passage around this formula

…it relocates where that computation’s output lands: $T_{\mathrm{total}} = T_{\mathrm{parent}} + \sum_{i=1}^{k} T_{\mathrm{subagent}_i}, \qquad T_{\mathrm{parent}} \ll \sum_{i=1}^{k} T_{\mathrm{subagent}_i}$. The parent conversation’s occupied share of its own window, $T_{\mathrm{parent}}$ , stays small and bounded even as the total work performed across k delegated subagents grows without bound. That inequality is the real content behind the phrase “context savings” — the saving is measured in the one window that has to keep attending to everything else in the task, not in total tokens billed.

Read this part in the article →