Spreading work across agents
A task too big for one window - does the agent start helpers, what do they receive, and what comes back?
Why this is hard
A sub-agent is the most expensive answer to a full window: instead of cutting, a second window is opened. It pays off when the subtask is self-contained and its result small - a research pass, a test run, a review. It goes wrong when the helper needs context it did not get, or when three helpers each start three helpers.
The layer hangs on three others: on the loop that starts the helper as a tool or as a process; on the budget, which now has to count across levels; and on the isolation of the guardrails, because a helper either inherits the parent's permissions or gets its own - both have consequences the code shows.
From Pro: the matrix per challenge, all eleven topics with citations, and the architecture portrait of every harness. Freely visible are the six short portraits:
- OpenClaw - The multi-channel personal agent - a half-million-line estate of gateway, channels and tooling whose loop core shares its lineage with Pi.
- Hermes - Nous Research's Python harness - deterministic loop, a skill system, and an error path that escalates in stages instead of just giving up.
- Pi - The coding agent with a deliberately small core - every layer hand-built and kept readable, from the provider package to the session file.
- HybridClaw - The enterprise runtime - sandboxed execution, approvals, and credentials that travel through context as references instead of plain text.
- DeepSeek Harness - The plugin radical - “everything is a plugin” is meant literally here: providers, tools and the loop itself are modules on the Cordis core.
- NanoClaw - The minimalist - small enough to understand in full: the model layer moves behind the Claude Agent SDK, isolation into a container.