---
id: "framework-three-necessities"
type: "framework"
source_timestamps: ["§ Three Necessities"]
tags: ["system-architecture", "implementation", "organizational-design"]
related: ["concept-continuous-assessment", "concept-continuous-sensing", "concept-in-workflow-coaching", "action-shift-capability-evidence", "action-analyze-task-level", "action-close-insight-loop"]
speakers: ["Sangeet Paul Choudary", "John Winsor"]
steps: ["\\\"Change what you treat as evidence of capability: stop relying on periodic reviews or self-reports; continuously monitor real-time signals like code commits", "customer calls", "and tool usage.\\\"", "\\\"Analyze work at the level of individual tasks: use continuous sensing to understand how work is distributed between humans and AI", "identifying which skills are being absorbed by tools and who is adapting well.\\\"", "\\\"Close the loop from insight to action: use the insights gained to allocate work", "redesign roles", "and provide in-workflow coaching", "adapting people inside the workflow rather than in separate training cycles.\\\""]
sources: ["tail1"]
sourceVaultSlug: "hbr-seg-tail1"
originDay: 1
articleStem: "hbr-tail-112-continually-assessing-performance"
sourceUrl: "https://hbr.org/2026/06/the-pros-and-cons-of-continually-assessing-performance"
sourceTitle: "The Pros and Cons of Continually Assessing Performance"
---
# Three Necessities for Continuous Assessment

The authors outline **three main actions** required to build a mature architecture for continuous assessment, transitioning away from periodic reviews toward continuously captured evidence from actual work.

**1. Change what you treat as evidence of capability.**
Stop relying on periodic reviews or self-reports. Continuously monitor real-time signals like *code commits, customer calls, and tool usage*. → Operationalized as [[action-shift-capability-evidence]]; enabled by tools such as [[entity-microsoft-skills-agent]] (powered in part by the [[entity-linkedin-skills-graph]]). This is [[concept-continuous-assessment]] made concrete.

**2. Analyze work at the level of individual tasks.**
Use [[concept-continuous-sensing]] to understand how work is distributed between humans and AI — identifying which skills are being absorbed by tools and who is adapting well. → Operationalized as [[action-analyze-task-level]]; evidenced by [[entity-stripe-minions]], [[entity-github-copilot-d1]] (with the [[entity-zoominfo]] deployment study), and [[entity-r-potential]].

**3. Close the loop from insight to action.**
Use the insights to allocate work, redesign roles, and provide [[concept-in-workflow-coaching]] — adapting people *inside* the workflow rather than in separate training cycles. → Operationalized as [[action-close-insight-loop]]; exemplified by [[entity-cresta-agent-assist]].

Critically, the enrichment underscores that this framework is *not just a measurement problem — it is also a learning and redesign problem*. The interpretive discipline of [[claim-contextual-performance-variation]] and the governance guardrails of [[claim-surveillance-backlash]] are what keep the architecture healthy.