Task statement 1.6 · Agentic architecture and orchestration

Design task decomposition strategies for complex workflows

Before the research coordinator delegates, it splits the job into steps. Some jobs have steps you can list before reading any source, while others reveal theirs only as findings arrive. The wrong shape yields a report that reads as complete but never looked at part of the topic.

Why it's tested

The tempting move is to write the full subtask list before any research starts, because a fixed list is easy to review and to budget. For an open-ended topic that list holds only what the coordinator already knew, so whatever the first searches uncover has no subtask to land in. For a large review that comes back uneven, the tempting fix is a larger context window, which fails because the input already fits.

KnowledgeTwo ways to split a job

Prompt chaining is a fixed sequential pipeline: the steps are written before the input is read, and each step processes the previous step's output. Each step has one narrow job, so each call is simpler and more accurate. You can also put a programmatic gate between steps, which is code that checks the intermediate output before the next step runs. The gate keeps the chain fixed rather than making it adaptive. Chaining suits predictable multi-aspect reviews. In the research system, every source the document analysis subagent receives goes through the same steps: extract its claims, match each claim to a supporting quote, then record the author and date.

Dynamic adaptive decomposition generates subtasks from intermediate findings instead. The coordinator starts with a step that maps the territory, picks the next subtasks from what that step found, and adds or reorders subtasks as dependencies appear. This suits open-ended investigation, where the useful subtasks are unknown until some of the work is done. Anthropic's research system works this way: its lead agent reads the results and "decides whether more research is needed".

Interactive · which shape does this job need?

Two decomposition shapes side by side, both flowing downward. Left: prompt chaining, four steps written before the input is read, with a gate between two of them. Right: dynamic adaptive decomposition, where a mapping step's findings choose the next steps and a discovered finding revises the plan. prompt chaining steps written before the input is read 1 · extract claims pass 1 · file 1 alone list of claims next file 2 · match claim to quote pass 2 · file 2 alone gate: all quoted? next file 3 · record author, date … 14 local passes cited claims all local findings 4 · hand to synthesis integration pass same steps, every source cross-file data flow Same steps, whatever the input says. dynamic adaptive decomposition steps generated from findings 1 · map: broad search 1 · map the structure finds six sectors import graph 2 · rank high-impact areas 2 · find high-impact areas music, film first billing: 31 callers 3 · prioritised plan new lead: court case dependency found 4 · plan revised subtask added currency helper first Each step exists because of a finding.

Choose a job to see which shape it needs.

Check each source → prompt chaining. The same three steps apply to every source, so they are written before any source is read.

Research an open topic → dynamic adaptive decomposition. Which sectors matter is known only after the mapping search.

Review a 14-file PR → prompt chaining. One local pass per file, then one cross-file integration pass, all known in advance.

Add tests to a legacy codebase → dynamic adaptive decomposition. The high-impact areas appear only once the structure is mapped.

Both panels read downward. Every step on the left was written before the input was read, and every step on the right exists because the step above it found something. The gate is code between steps 2 and 3. A claim with no supporting quote stops the chain there instead of reaching synthesis. The dashed arrow is a finding nobody planned for. It adds a subtask that no up-front list could have contained. Each local pass carries one file, so depth stays even. Only the integration pass sees how the files connect. The billing tests reveal the untested currency helper every charge passes through, so the plan changes to test it first.

SkillMapping an open-ended topic first

Asked about AI in the creative industries, a coordinator that plans up front assigns digital art, graphic design and photography. Every subagent succeeds, and the report never mentions music, writing or film. The plan was written before any finding existed, so nothing could add to it. The adaptive version maps first and lets each finding change the plan:

// coordinator plan log, adaptive
step 1  map    broad search → art, design, photography, music, writing, film
step 2  rank   music and film have the most sources → first
step 3  plan   one subtask per sector, music and film first
step 4  found  music licensing turns on one pending court case
        plan += "analyse the court case"   // not in any up-front list
Scenario switchThe guide's own examples for this skill come from its code scenarios. The 14-file pull request and the legacy codebase are walked here as the guide writes them, and the principle is unchanged.

SkillA review split into passes, and a legacy codebase

One review pass over a 14-file pull request gives detailed feedback on some files and superficial comments on others, misses obvious bugs, and flags a pattern in one file while approving identical code in another. Attention dilution is this failure: one pass carries more than it can attend to evenly. The fix is a prompt chain of fourteen per-file local analysis passes, then a separate cross-file integration pass that examines data flow between files. Keeping only findings that two of three runs share would suppress bugs that are caught intermittently.

"Add comprehensive tests to a legacy codebase" is open-ended, so it gets dynamic adaptive decomposition. First map the structure, such as which modules import which. Then identify high-impact areas: a billing module with no tests and 31 callers comes first. Then create a prioritised plan that adapts as dependencies are discovered. When the billing tests show that every charge passes through an untested currency helper, that helper moves to the top.

The decision test Can you name the steps before you read the input? Yes → prompt chaining: write the sequence, one concern per step. No → dynamic adaptive decomposition: map first, then let each finding generate the next subtask.
Takeaway
  • Uneven, contradictory findings from one big pass mean attention dilution: split into per-file passes plus an integration pass.
  • A plan written before the first finding covers only what you already knew.
  • A gate between chain steps checks output. It does not make the chain adaptive.

check yourself

0 / 3 answered

Every Monday the research system writes a pricing brief on the same five competitors, filling the same four fields for each: price, plan tier, date and source URL. Auditors compare each week's brief with the last one and need every week to follow the same steps. An engineer proposes letting the coordinator plan the brief from scratch each week.Which decomposition best fits the brief while keeping the weeks comparable?

Select one.

Asked why heat-pump adoption grew faster in Norway than in Sweden, the coordinator writes eight search queries before any search runs and delegates them. The report is well cited but never mentions a 2019 subsidy change, although one source retrieved in the first round names it as the main cause.Which change to the decomposition most directly fixes this?

Select one.

Before report generation, one review pass checks a 40-page synthesis draft. Late sections get one-line notes, and an uncited statistic is flagged in section 2 but passed in section 7. Nobody notices that sections 3 and 9 give different figures for the same market. The team wants the same checks run on every draft.Which TWO changes best restructure the review?

Select two.

sources

  1. Anthropic Engineering, "Building effective agents": prompt chaining "decomposes a task into a sequence of steps, where each LLM call processes the output of the previous one", with programmatic gate checks on intermediate steps, for tasks that decompose cleanly into fixed subtasks. Checked 2026-09-30.
  2. Anthropic Engineering, "How we built our multi-agent research system": research cannot follow "a fixed path" because it is "dynamic and path-dependent", and the lead agent "decides whether more research is needed". Checked 2026-09-30.
  3. The two decomposition patterns, the 14-file review symptoms and fix, attention dilution, and the four steps for "add comprehensive tests to a legacy codebase": Claude Certified Architect — Foundations Exam Guide v1.0, Domain 1.6 and Sample Question 12. The creative-industries sectors: the guide's Sample Question 7. The court case, the billing module and its 31 callers are this course's illustrations.