Task Decomposition — References

Prompting work on breaking problems down, and the distributed-computing pattern this inherits its rejoining semantics from.

Decomposition in prompting

• Least-to-Most Prompting Enables Complex Reasoning in Large Language Models (Zhou et al., 2022), arXiv:2205.10625 — explicitly producing subproblems and solving them in order of increasing difficulty, with each result available to the next.
• Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (Wei et al., 2022), arXiv:2201.11903 — the baseline against which explicit division is usually compared, and the source of the intermediate-step framing.
• Decomposed Prompting: A Modular Approach for Solving Complex Tasks (Khot et al.) — subproblems dispatched to specialised handlers, including recursive handling of the same subproblem type. Cited by name.

Dividing and rejoining at scale

• MapReduce: Simplified Data Processing on Large Clusters (Dean and Ghemawat, 2004) — the canonical statement of split, independent processing, and deterministic combination, and the source of the argument that the combining function should be structural. Cited by name.
• AWS Step Functions documentation — dynamic fan-out with per-item retry, concurrency limits, and result aggregation, as a concrete implementation of bounded division in a durable engine.

No verified reference is recorded for choosing unit size in agent workloads specifically; the boundary criteria in current use are empirical and product-dependent.