Last week, I asked you to take the AI Readiness Assessment and promised we’d dig into what it reveals and more specifically, why the gaps people find tend to be where the biggest leverage is. I’ve been combing through the early results, and there’s something in the data that’s worth sharing.
Here’s what I initially expected to find: people struggling with judgment. Unsure of when to delegate to AI and when to hold the line. Uncertain about whether to trust an output, whether to override a recommendation, whether the tool was making things better or just faster.
Interestingly, that’s not what the data showed.
The people who took the assessment — a self-selected group, which means they care enough about this question to stop and measure themselves — showed sophistication in the judgment dimensions. When asked what to do when an AI recommends a completely different project prioritization order than expected, the most common answer was to ask for the AI’s criteria, weigh them against team capacity and business goals, and make a hybrid decision. When asked how to handle an AI tool that’s right 80% of the time on customer service responses, the most common answer was to define clear boundaries. These are not the answers of people who are blindly trusting or blindly resisting.
A self-selected group of quiz takers will skew the data, yes of course. However, it’s a starting point for a conversation.
So where is the gap? The data seems to point at the operating layer that sits underneath the judgment.
What the AI Readiness Assessment Revealed
Three dimensions surfaced where even engaged, AI-forward people were struggling, and they’re not the ones that get talked about in most AI readiness conversations.
Cognitive Load Management
When asked what happens by Friday after a week of using AI for drafting, research, brainstorming, and summarizing simultaneously, nobody said they’d solved this. The most common response — half the group — was some version of “I notice the burnout but can’t figure out how to pace myself differently” or “I’m getting better at spacing it out, but I’m still figuring it out.” Only one person said they actively structured their week to balance AI-intensive work with solo, offline work. This is a real and under-appreciated drag. Cognitive fatigue doesn’t show up in any adoption metric, but it does show up six months later when someone attrits, or when output quality starts to erode, or when people start avoiding the tools altogether because they associate them with exhaustion.
Workflow Integration
Nobody — not one person — said they had clear boundaries and dedicated workflows that prevent their AI tools from overlapping with the rest of their work. The best answer most people gave was “some routines to compartmentalize,” and a third said they were aware of the fragmentation but hadn’t found a way to organize it yet. This matters because fragmented tooling creates fragmented thinking. If you don’t know which conversation informed which decision by the end of the week, you can’t learn from what worked, you can’t replicate what went well, and you can’t explain your reasoning to a colleague or a manager. That’s an infrastructure problem that AI has made more visible.
Policy Container
When asked how they make decisions about what’s safe to do with AI in the absence of a clear company policy, half said some version of “I try to be thoughtful, but I’m often unsure if I’m making the right call.” Only one person had developed their own principles based on an understanding of data risk and company culture. The rest were making judgment calls in a vacuum. And these, again, are engaged people. People who read newsletters about AI strategy. People who took a readiness assessment on a Thursday. If they’re operating without infrastructure, most of their colleagues almost certainly are too.
Why This Matters More Than the Skills Gap
Bloomberg Law released its 2026 State of Practice survey this week, and the headline is striking: 83% of lawyers are now using AI at work, and efficiency gains are lagging behind. Law is one of the most document-heavy, pattern-rich professions there is — in theory, a natural fit for AI. And yet adoption has outpaced measurable productivity lift.
The skills explanation doesn’t hold up well here. Lawyers are not necessarily struggling to learn new tools. What the data points toward instead is the same thing the assessment is showing: even high adoption without the right kinds of operational scaffolding kills efficacy and velocity.
The Census Bureau data published last month is where the rubber hits the road. According to an NBER working paper on AI diffusion using the latest BTOS survey data, 18% of firms are using AI in at least one business function, but among adopters, scope is narrow — 57% use AI in three or fewer functions. AI is spreading through organizations task by task, function by function, not as a systemic transformation. Which means the gaps are accumulating at the task level… bottom’s up, slowly, and without governance.
The companies that trained 80% of their team and then couldn’t execute weren’t necessarily failing because their people lacked judgment. Perhaps it was that these organizations hadn’t yet built the conditions in which that judgment could function effectively. Without thoughtful workflow infrastructure, an understood policy container to work with, and tools to manage the cognitive overhead, organizations are the ones that end up holding back individuals.
Where the Leverage Lives
Is today’s readiness conversation aimed at the wrong layer? The energy goes into skills training, tool procurement, and mandates. The scaffolding that would let those skills perform at scale gets treated as a downstream problem — something people will figure out on their own once they’ve adopted the tools — when in fact, it’s an organization and leadership problem.
They won’t. Or rather, some will, individually, which is how you end up with the 5x adoption gap between your highest and lowest AI users inside the same organization, with the same tools, the same training, and the same mandate.
Here’s where I’d focus if I were looking for leverage right now:
For individual contributors
The highest-value thing you can do in the next two weeks is build a decision log. It doesn’t have to be elaborate, a simple system where you are documenting everything you use AI for, the inputs/outputs, and outcomes should be sufficient. This does two things simultaneously: it closes the cognitive load gap (you can also use my open-sourced Chief of Staff/second brain Claude Code skill which I use every day) and it builds the audit trail that makes your AI-assisted work legible to others. Legibility is what gets you more autonomy. It also, over time, gives you real data on where AI is making you better and where it’s just making you faster.
For leaders whose mandates have stalled
Stop diagnosing the skills gap and start diagnosing the infrastructure gap. The question to ask your team is not “are you using AI?” but “when you use AI, can you trace the decision afterward?” and “do you know what’s safe to share and what isn’t?” Those two questions will surface initial operating conditions gaps faster than any adoption survey. And the fixes are easy wins: a clear data policy, a shared framework for when AI should draft versus when humans should lead.
For leaders who haven’t mandated yet
The assessment data suggests that the people most ready to move forward are not the ones who’ve been trained the most. They’re the ones who’ve developed their own operating practices, their own judgment about what to share and what to hold back, their own way of managing the cognitive overhead in the absence of organizational infrastructure to provide the same. That’s a culture signal, not a skills signal. If you’re waiting for your organization to be “ready” before mandating, you’re waiting for an infrastructure problem to solve itself. It won’t. The infrastructure has to be built alongside adoption, not after it.
The readiness gap is real and it’s fixable, but only once you’re diagnosing the right layer. If you want to know where your organization stands, start with two questions: can your team trace decisions that came from AI, and do they know what’s safe to share? Those are harder to answer than an adoption survey, but they’ll tell you more.
If you haven’t taken the assessment yet, it’s still live and still free — no email required. Take it here.
Leave a Reply