Module 5 • Lesson 39• 35 mins

Designing Reasoning Flow at the Product Layer

Select across 4 UI reasoning disclosure levels based on the 2-axis matrix: Task Latency and Cost of Failure.

Distinguish 4 Reasoning Disclosure Levels on the user interface

Calibrate disclosure levels on the Latency × Cost of Failure matrix

Designing Reasoning Flow at the Product Layer

In AI Literacy Lesson 12, you learned the ReAct loop (Thought → Action → Observation) — the internal execution mechanism powering AI agents. However, the core question for Product Managers is not how the agent computes under the hood, but: "Does the user see that reasoning trajectory, and to what level of granularity?" This strategic UI decision is known as the Reasoning Disclosure Level.

Running example: ClaimGenie — an agent automating travel insurance claim processing and compensation.

1. The 4 UI Reasoning Disclosure Levels

In Lesson 34, you explored Reasoning Traces as one of 3 core explainability modalities (alongside Citations and Factor Highlights). In multi-step agentic workflows, you must calibrate how much of this internal state machine is exposed to the user:

Level 1 — Fully Hidden: Displays only the final output, concealing all underlying execution steps. Ideal for instantaneous, low-stakes micro-tasks.
Level 2 — Step Progress Labels: Displays high-level status chips ("Checking flight records...", "Cross-referencing insurance policy...") without revealing raw thought outputs. Reduces perceived waiting friction (extending the streaming principles from Lesson 30 to agent steps).
Level 3 — Step + Approval Gate: Exposes critical intermediate reasoning steps and intentionally halts execution to mandate explicit user approval before proceeding (integrating with Human-in-the-loop patterns from Lesson 32).
Level 4 — Full Expandable Trace: Provides a "Show work / View detailed trace" drawer, allowing users or internal compliance teams to inspect the full Thought → Action → Observation sequence post-completion.

4 Levels of Reasoning Disclosure on the User Interface

Calibrate reasoning transparency against Task Latency and Cost of Failure.

Select a Disclosure Level:

Level 3: Step + Approval Gate

UI Mechanism / Pattern: Exposes critical intermediate reasoning and halts execution until the user explicitly approves.

Optimal Fit: High Cost of Failure (regardless of whether execution is fast or slow).

Real-world Example: Drafting commercial contract clauses, initiating wire transfers.

Reasoning disclosure is not just for waiting-time perception; it is a critical governance tool for trust and risk containment.

2. The 2-Axis Decision Matrix: Latency and Cost of Failure

Selecting a Disclosure Level is not a subjective aesthetic choice; it is calibrated across two objective dimensions: Task Latency and Cost of Failure.

	Low Latency (< 3s)	High Latency (> 5-10s)
High Cost of Failure	Level 3 (Step + Approval Gate) Example: Drafting legal contractual covenants, issuing wire transfers. Mitigates risk even when execution is instant.	Level 4 (Full Trace) or Level 3 Example: Adjudicating insurance claims. Reduces wait anxiety while preserving compliance auditability.
Low Cost of Failure	Level 1 (Fully Hidden) Example: Summarizing a short email, reformatting a markdown table. Exposing internal traces only creates cognitive noise.	Level 2 (Step Progress Labels) Example: Aggregating web news across 10 sites. Step labels confirm system liveness and progress.

3. Decoupling Latency Management from Risk Governance

A frequent product anti-pattern is conflating latency feedback with risk containment:

Latency Management (Lesson 30): When an agent experiences long inference delays, high-level step labels (Level 2) prevent user drop-off.
Risk Governance & Trust (Lesson 34 & Lesson 39): Even if an agent executes in under 2 seconds, high-stakes actions (disbursing funds, deleting records, signing documents) mandate Level 3 or 4 disclosure so users can inspect and verify the reasoning premise.

4. Analogy: Routine Prescription Refill vs. Multi-Specialist Medical Board

The disclosure spectrum mirrors two distinct healthcare interactions:

Routine Prescription Refill (Low Latency, Low Risk): The pharmacist hands you the standard package (Level 1). Explaining how many steps they took to fetch the bottle is annoying and irrelevant.
Complex Surgery Consultation (High Risk): The medical board presents the full differential diagnosis: why alternative illnesses were ruled out, what biomarker findings mean, and why surgery is necessary before asking you to sign the consent form (Levels 3 & 4). Patients require transparent reasoning to establish trust.

Exercise 39.1: Compare and specify the appropriate Disclosure Level for 2 products:

Product A — LegalDraft: An agent auditing and amending copyright clauses in enterprise vendor contracts (execution time: 3 seconds).
Product B — FlightClaimBot: An agent cross-referencing flight delay records across 3 airline databases and submitting a €200 passenger compensation claim (execution time: 15 seconds).

Specify the Disclosure Level (Level 1–4) for each product and justify your decision using the Latency × Cost of Failure matrix.