---
id: "framework-four-step-ai-development"
type: "framework"
source_timestamps: ["§ A New Development Model", "\\\"§ Step 1. Establish an initial point of view", "so you have a basis for evaluating AI's output.\\\"", "§ Step 2. Collaborate with AI across multiple modes.", "§ Step 3. Analyze the differences between your initial view (from Step 1) and AI's output.", "§ Step 4. Deliver the output with an explanation of how you and AI arrived at it."]
tags: ["workflow", "training", "methodology"]
related: ["framework-ai-collaboration-modes", "framework-difference-analysis", "concept-reasoning-trail"]
steps: ["Establish an initial point of view: scope the task and form a preliminary hypothesis of the answer before opening any AI tool.", "\\\"Collaborate with AI across multiple modes: move beyond simple generation to critique", "compare", "simulate", "and challenge.\\\"", "\\\"Analyze the differences: diagnose what AI added", "what it got wrong", "and what looks right but isn't", "comparing against your initial view.\\\"", "\\\"Deliver the output with an explanation: submit the final task alongside a reasoning trail detailing the starting AI output", "human changes", "and the observed jagged frontier.\\\""]
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"
---
# Four-Step AI Skill Development Process

A systematic process for developing the hybrid skill of combining human judgment with AI capability. It shifts focus from *generating output* to the meta-level work of reflection and interrogation, and culminates in a [[concept-reasoning-trail|reasoning trail]] for managerial review. It was piloted at [[entity-disruptive-edge-d32|Disruptive Edge]].

**Step 1 — Establish an initial point of view.** Scope the task (audience, utility criteria) and form a preliminary hypothesis of the answer *before opening any AI tool*. If the task is entirely unfamiliar, use AI strictly to ask what a strong deliverable looks like and what judgment calls are involved, then form your own view. This deliberately reintroduces [[contrarian-friction-is-good|friction]] — see [[action-establish-pov|establish a POV first]] and [[quote-friction-is-necessary|the friction quote]].

**Step 2 — Collaborate with AI across multiple modes.** Move beyond simple generation to critique, compare, simulate, and challenge — the [[framework-ai-collaboration-modes|five modes of AI collaboration]]. See [[action-use-multiple-ai-modes|engage AI in multiple modes]].

**Step 3 — Analyze the differences** between your Step 1 view and the AI's output using the [[framework-difference-analysis|difference analysis]]: what AI added, what it got wrong, and what [[concept-looks-right-but-isnt|looks right but isn't]]. Executing this step requires [[prereq-domain-knowledge|underlying domain knowledge]].

**Step 4 — Deliver the output with an explanation.** Submit the final task alongside a [[concept-reasoning-trail|reasoning trail]] documenting the starting AI output, the human changes and why, and a one-sentence read on the [[concept-jagged-frontier|jagged frontier]] for the task.

The model is the article's central deliverable. The enrichment overlay notes it is a *coherent original synthesis* rather than a canonical published framework — its steps align with established critical-thinking, reflective-practice, and human-in-the-loop traditions [4][6][7]. It grounds [[claim-reasoning-trail-accelerates-judgment|the apprenticeship-acceleration claim]]. The open question [[question-time-efficiency-tradeoff|whether its friction negates AI's speed]] remains unresolved.