An anonymized participant case note from a Prune work simulation. Some identifying details have been removed.
A senior executive with approximately 20 years of professional experience recently completed a Prune AI-fluency assessment.
Her final report demonstrated strong authorship. She did not simply accept an AI-generated answer. The structure, priorities and recommendation reflected her own judgment, and she maintained control over the final deliverable.
But the assessment revealed a subtler workflow gap.
The source pack contained a material contradiction. Two pieces of evidence could not both support the conclusion as written. Although the participant used AI effectively to synthesize the available information, she did not use it to systematically check the final deliverable against the underlying sources.
The resulting report was coherent, well structured and largely her own work—but one important contradiction remained unresolved.
The old workflow: generate, then review
A common model for responsible AI use is:
Use AI to generate or synthesize.
Review the output yourself.
Correct anything that appears wrong.
That is better than accepting AI output without review. But it can create an important blind spot: the human reviewer is still working from their own attention and memory.
If a contradiction was missed during the initial analysis, simply rereading the final answer may not expose it.
In this assessment, AI had been used as a synthesis tool, but not as a verification tool.
The workflow change
After receiving her assessment report, the participant agreed with the finding and changed her workflow.
Instead of using AI only to produce an initial synthesis, she began adding a separate verification pass. The new process asks AI to:
compare each material claim in the deliverable with the provided sources;
identify evidence that supports or conflicts with the claim;
flag contradictions between source documents;
distinguish sourced facts from assumptions or interpretation;
surface claims that cannot be traced to the available evidence.
The participant still owns the conclusion. AI does not determine whether the deliverable is correct. It provides a second, deliberately sceptical pass that helps the participant focus human attention where verification is most needed.
Why this matters
The assessment did not reveal that the participant lacked experience, judgment or authorship. It revealed that a familiar AI workflow was incomplete.
That distinction matters.
AI fluency is not simply the ability to produce good work with AI. It also includes knowing how to use AI in different roles: as a researcher, synthesizer, critic and verification aid—while retaining human responsibility for the final result.
A polished deliverable can coexist with a missed contradiction. Strong authorship can coexist with an improvable verification process. Realistic work simulations create an opportunity to observe both.
This is evidence from one participant completing one bounded assessment. It does not establish performance across every task or prove that the revised workflow will always prevent errors. But for this participant, the report identified a practical change she considered useful enough to adopt.
That is the kind of development insight Prune is designed to surface: not simply whether someone uses AI, but where a small change in how they use it may strengthen the work.
Practical prompt: “Review this deliverable against the supplied source material. For every material claim, identify its supporting evidence, conflicting evidence and any unsupported assumptions. Do not rewrite the deliverable yet.”
