AI fluency assessment

Observe how people work with AI—not only what they know about it

twoprune uses timed, realistic knowledge-work simulations to create evidence of how a participant frames a task, navigates sources, works with AI, verifies outputs, exercises judgment and delivers a result.

An AI fluency assessment is a bounded work sample. It examines demonstrated behaviour in one assessment context and communicates the evidence, interpretation, confidence and limitations for human review.

What is an AI fluency assessment?

An AI fluency assessment examines how a person applies AI within realistic work, rather than relying only on self-report, multiple-choice knowledge or tool familiarity.

The participant receives an assignment, reviews workplace-style materials, uses the available tools and optional AI support, and submits a business deliverable. The assessment can consider both the final work and the available evidence of how that work was produced.

This distinction matters because a polished output does not reveal whether the participant missed a source contradiction, accepted an unsupported AI claim or used sound verification and judgment.

What the assessment measures

twoprune organizes observed performance into six connected pillars. The pillars describe applied work behaviours; they are not personality traits or a claim of universal professional competence.

Context Framing

How well the participant understands the task, audience, constraints, decision context and success criteria.

Evidence Navigation

How well the participant finds, reviews, weighs and uses relevant information from the available materials.

AI Orchestration

How effectively the participant directs, decomposes, iterates with and critiques AI support.

Verification & Risk Control

How well the participant checks outputs, manages uncertainty, catches unsupported claims and controls operational risk.

Judgment & Synthesis

How well the participant combines evidence, trade-offs and domain reasoning into a defensible recommendation.

Communication & Delivery

How clearly and usefully the participant produces a business deliverable for the intended audience.

How the assessment works

The assessment creates one connected evidence chain from the task context to the report.

Key points

  • 1. Enter a realistic simulation

    The participant receives a prioritized assignment with synthetic or anonymized workplace materials and a defined time limit.

  • 2. Work across familiar tools

    The workspace can include an inbox, documents, spreadsheets, research, notes, AI support and a structured deliverable builder.

  • 3. Submit a business deliverable

    The participant produces a decision memo, analysis, plan or other task-specific result for the stated audience.

  • 4. Review evidence-backed interpretation

    The report connects available observations to pillar-level interpretation, confidence limits, risks and development areas.

How a work simulation differs from other AI assessments

Knowledge quiz

Checks recognition or recall. It can show what someone knows about AI concepts but not necessarily how they work through ambiguity.

Self-assessment

Captures confidence and perception. It is useful context, but self-report does not independently demonstrate applied behaviour.

Conversational diagnostic

Explores reasoning through dialogue. It may be faster to administer but does not always produce a complete workplace deliverable.

Realistic work simulation

Creates a bounded work sample in which evidence use, AI collaboration, verification, judgment and delivery can become observable together.

What the result does—and does not—establish

A twoprune report is structured assessment evidence for human interpretation. It can support development planning, post-training review, capability conversations and scenario-bounded follow-up.

It does not predict future job performance, certify capability across an entire role, replace professional judgment or make an employment decision. Missing workflow evidence should reduce confidence; it should not be treated as proof of poor performance.

Where benchmark cohorts are immature or not comparable, percentile information should be labelled directional or unavailable rather than presented as precise.