Latest

Lay of the land: capability is flat, but failure modes cluster by role

GenAI capability scores are broadly consistent across functions — no role is meaningfully ahead. But when you look at how people fail, clear patterns emerge. That changes where to invest first. Read more →

2 min readcapabilityenablement
2 min read

The confidence trap: why self-assessment fails and what it costs

Self-reported confidence is a poor proxy for GenAI competence. Nearly 1 in 4 users are overconfident — and they make 7x more verification errors than their capable peers. Read more →

capabilitydiagnostics
2 min read

Hidden failure modes: verification, data handling, and the blindspot

Verification is the #1 skill gap. Data handling errors are the most common error type. And the people who most need data handling training rarely ask for it. Three findings that change how enablement should be designed. Read more →

enablementdata-handling
3 min read

The training paradox: frequent use without training, and what actually works

Two-thirds of daily GenAI users have never been trained. Usage alone doesn't predict capability — but training does. Organizational training outperforms self-directed learning, and governance reduces errors by 34%. Read more →

trainingenablement
2 min read

Governance works: 34% fewer GenAI errors when tools and policy align

Respondents with both approved tool access and completed policy review score higher and make substantially fewer errors. Access without awareness still leaves gaps. Read more →

governanceenablement
2 min read

Organizational training outperforms self-directed learning in GenAI

Employer-provided training is associated with a 12-point capability lift over no training. Self-directed learning alone performs no better than the untrained baseline. Read more →

trainingenablement
2 min read

Training — not experience — predicts GenAI capability

In a multivariate regression, training exposure was the only statistically significant predictor of GenAI capability. Usage frequency alone does not predict competence. Read more →

trainingcapability
2 min read

Two-thirds of daily GenAI users have never been trained

The people using GenAI most frequently are overwhelmingly untrained. Without structured enablement, high-volume users amplify verification and data handling errors across every workflow. Read more →

training
2 min read

The data handling blindspot that won't self-correct

Among respondents whose weakest area is data handling, 93% prefer training in other topics. The biggest gap is invisible to the people who have it. Read more →

data-handlingdiagnostics
2 min read

One in three GenAI errors involves unsafe data handling

When GenAI judgement errors are categorized by type, data handling failures — sharing sensitive information with AI tools — are the single most common category. Read more →

data-handling
1 min read

Overconfident GenAI users make 7x more verification errors

Overconfidence is not just a perception problem. It translates directly into higher error rates — especially in verification and data handling, the two areas where mistakes carry the most operational risk. Read more →

capabilityverification
2 min read

Nearly 1 in 4 GenAI users are confidently wrong

Self-reported confidence is one of the most common proxies for GenAI readiness. It is also one of the least reliable. Nearly a quarter of users are high-confidence and low-competence. Read more →

capabilitydiagnostics
2 min read

Three enablement streams emerge when you look at error patterns by role

Overall GenAI capability is flat across functions. But the types of mistakes people make cluster sharply by role — and that changes how enablement should be designed. Read more →

enablement
2 min read

GenAI capability is role-agnostic — start with a shared baseline

Organizations often assume some functions are further ahead with GenAI than others. The data tells a different story — and it changes where to invest first. Read more →

capability