You implement a multi-agent market-research workflow. Competitor pricing scrape, social sentiment analysis, and regulatory news retrieval are independent. A synthesis agent must wait for all three. Which orchestration approach best optimizes task duration?
Select an answer to reveal the explanation.
Short Explanation
Run the independent research agents concurrently, then wait for all of them before synthesis. That cuts total duration by overlapping work while still giving the synthesis agent complete inputs. Sequential pipelines, empty-input synthesis, and unbounded spawning all waste time or break control.
Full Explanation
The correct approach is to run the three independent research agents in parallel, then run synthesis after a join barrier once all complete. Multi-agent orchestration should use parallel patterns for independent subtasks and control concurrent execution, then sequence dependent steps after completion. This reduces end-to-end duration versus pure sequential execution. Running agents one-after-another in a fixed sequential pipeline is incorrect because unnecessary sequencing increases latency. Starting synthesis immediately with empty inputs is incorrect because synthesis without inputs yields low-quality or empty analysis. Spawning unbounded copies of every agent on each keystroke without batching or rate limits is incorrect because uncontrolled spawning risks rate limits, cost spikes, and instability. Pair parallelism with timeouts, cancellation, and aggregation rules for slow or failed branches.