Aniesh PrepArchitecture Lens demo
From an answer to an architectural tradeoff
See the difference between the authored explanation and an optional question-specific AI response. This published example came from a synthetic test answer and was reviewed on 2026-10-08 using claude-haiku-5-5. It is not a live request or learner feedback. Opening this page sends nothing to Anthropic.
1. The question
A renewal-summary service always retrieves usage from one API, calculates adoption using a fixed formula, then drafts a summary. Every run must record the same steps and reproduce the numeric result. No exploratory investigation is needed. Which starting architecture best fits?
- Run retrieval and calculation in a fixed workflow, then have Claude draft from validated results.
- Let one tool-using agent choose the retrieval, calculation, and drafting sequence for each account.
- Give retrieved usage to one Claude call that calculates adoption and drafts the final summary.
- Use separate retrieval, calculation, and writing agents, with a coordinator choosing their order.
2. The selected answer
B. Let one tool-using agent choose the retrieval, calculation, and drafting sequence for each account.
This example selection is incorrect under the stated constraints. It does not show what the learner was thinking.
3. The authored explanation
Correct answer: A. The sequence is known and the formula must be reproducible. A fixed workflow records each stage, keeps arithmetic deterministic, and uses Claude where language generation is useful.
Compare all alternatives
- A: The sequence is known and the formula must be reproducible. A fixed workflow records each stage, keeps arithmetic deterministic, and uses Claude where language generation is useful.
- B: A tool-using agent is reasonable for variable investigation, but choosing the sequence dynamically adds autonomy that this fixed, auditable task does not need.
- C: A single call is economical for drafting from supplied data, but asking it to calculate the figure does not provide the required deterministic arithmetic stage.
- D: Specialist agents can help with independently complex tasks. Here the stages are simple and dependent, so dynamic coordination adds overhead without meeting an unmet need.
4. Lens: when would option B fit?
Key takeaway
A tool-using agent that chooses its own sequence for each account fits only tasks where the next steps depend on evidence; this fixed, auditable summary does not need that autonomy.
The actual constraint
The scenario requires every run to record the same steps and reproduce the same numeric result, and states that no exploratory investigation is needed. Letting the agent choose the sequence per account makes the path variable and weakens the audit trail.
What would need to change
Option B fits only if the task changes so that evidence determines the next step, such as investigating unexplained adoption changes by choosing which source to inspect next. Even then, keep arithmetic in deterministic tools and log the path taken. Under the original constraints, a fixed workflow is the better starting point.
The extra value is a concrete change in circumstances: investigation can need dynamic tool choices, while arithmetic stays deterministic. Lens also offers simpler concept explanations and feedback on the submitted selection. It does not infer your reasoning, score your confidence or predict exam success.