Search⌘ K
AI Features
Home
Courses
Master Claude Certified Architect Foundations Exam (CCAR-F)

Parallel Work and Result Synthesis

Explore how to implement parallel dispatch of specialists in a multi-agent system and handle diverse outcomes including access failures and empty results. Learn to validate results and synthesize them into a coherent final report that flags gaps or conflicts, enhancing your ability to build robust, coordinated AI architectures.

We'll cover the following...

In the previous lesson, we designed the coordinator-specialist split and the handoff packet. This lesson covers the runtime: dispatching specialists simultaneously, handling the full range of outcomes (success, partial result, access failure, empty result), and building the synthesis step that turns multiple structured results into a single coherent output. By the end of this lesson, we will be able to:

  • Dispatch multiple specialists in parallel and collect all results before synthesis

  • Distinguish access failures from empty results and handle each correctly

  • Merge specialist results into a synthesis payload

  • Build a coordinator synthesis step that is complete only when all expected fields are accounted for

Dispatching specialists in parallel

When specialist subtasks are independent, the coordinator dispatches all of them before waiting for any results. This is the primary performance benefit of a multi-agent architecture: total wait time equals the slowest specialist, not the sum of all specialists. In Python, this pattern uses concurrent.futures:

from concurrent.futures import ThreadPoolExecutor, as_completed
def run_specialist(system: str, prompt: str) -> dict:
    response = client.messages.create(
        model="claude-opus-4-8",
        max_tokens=1024,
        system=system,
        messages=[{"role": "user", "content": prompt}]
    )
    for block in response.content:
        if block.type == "text":
            return json.loads(block.text)
    return {"error": "no_text_block", "data_gaps": ["specialist returned no content"]}
def dispatch_all_specialists(company: str, fiscal_year: int) -> dict:
    handoffs = {
        "financial": build_financial_handoff(company, fiscal_year),
        "legal":     build_legal_handoff(company),
        "news":      build_news_handoff(company, days=90),
    }
    results = {}
    with ThreadPoolExecutor(max_workers=3) as executor:
        futures = {
            executor.submit(
                run_specialist,
                SPECIALIST_SYSTEMS[name],
                packet_to_prompt(packet)
            ): name
            for name, packet in handoffs.items()
        }
        for future in as_completed(futures):
            name = futures[future]
            try:
                results[name] = future.result()
            except Exception as exc:
                results[name] = {
                    "error":      "specialist_exception",
                    "message":    str(exc),
                    "data_gaps":  [f"All {name} data unavailable — specialist raised an exception"]
                }
    return results
Parallel specialist dispatch using a thread pool. All three specialists start simultaneously. Results and exceptions are both captured into the results dict — the coordinator sees every outcome, including failures.

  • Lines 3–13: run_specialist runs one specialist. It calls the API with the specialist’s system prompt and handoff-packet prompt, parses ...