---
id: "framework-ai-collaboration-modes"
type: "framework"
source_timestamps: ["§ Step 2. Collaborate with AI across multiple modes."]
tags: ["prompt-engineering", "critical-thinking"]
related: ["framework-four-step-ai-development"]
steps: ["Generate: ask AI to produce multiple versions with different approaches (breadth).", "Critique: ask AI to identify its own weakest assumptions; ask yourself what the AI doesn't know about the situation.", "Compare: ask AI to surface tradeoffs between alternatives; ask yourself which version fits the actual audience stakes.", "Simulate: ask AI how a specific stakeholder would react; ask yourself where the AI's simulation of that person is wrong.", "Challenge: ask AI where its data sources are weakest; ask yourself what non-public information contradicts the output."]
sources: ["reskilling"]
sourceVaultSlug: "hbr-seg-reskilling"
originDay: 10
articleStem: "hbr-edu-32-help-employees-get-better-with-ai"
sourceUrl: "https://hbr.org/2026/06/help-employees-get-better-not-just-faster-with-ai"
sourceTitle: "Help Employees Get Better—Not Just Faster—with AI"
---
# Five Modes of AI Collaboration

A taxonomy of five interaction modes that forces *both* the human's and the AI's reasoning into the open. Most users stop at **Generate** — precisely where human judgment is least visible. The other four modes pressure-test the AI's logic against real-world constraints. Used in [[framework-four-step-ai-development|Step 2 of the four-step model]]; see [[action-use-multiple-ai-modes|the action to use multiple modes]].

1. **Generate** — *AI:* produce multiple versions with different approaches (breadth). *You:* review the spread.
2. **Critique** — *AI:* identify its own weakest assumptions. *You:* ask what the AI doesn't know about the situation.
3. **Compare** — *AI:* surface tradeoffs between alternatives. *You:* decide which version fits the actual audience stakes.
4. **Simulate** — *AI:* predict how a specific stakeholder would react. *You:* spot where the AI's simulation of that person is wrong.
5. **Challenge** — *AI:* expose where its data sources are weakest. *You:* supply the non-public information that contradicts the output.

Each mode pairs an AI-facing prompt with a human-facing question — the pairing is what surfaces hidden flaws and missing [[claim-ai-lacks-context|context]]. This is where [[concept-ai-era-judgment|judgment]] becomes visible instead of hiding inside a single 'generate' step.