Future Direction: Harness Boundary Evolution (Where/When/What/With-Whom)

Description

Third of six open design directions (Section 12.3), building on Rajasekaran's observation that 'the space of interesting harness combinations doesn't shrink as models improve; it moves'. Four sub-axes: WHERE the harness runs (managed agents virtualizing session/harness/sandbox into replaceable interfaces); WHEN it acts (proactivity, KAIROS-style tick prompts); WHAT it acts on (vision-language-action beyond textual tool returns, physical affordances, humanoid control); and WITH WHOM it coordinates (role-differentiated multi-agent systems, debate, graph-structured workflows). Open whether a single harness can span all four extensions or will fragment into specialised stacks.

Key claims

Relations

Sources

src-20260423-0cff68d3291b