Top view of smart business man put scrum board on table at meeting room while marketing team preparing marker and sticky notes for making task board for managing work flow in work place. Convocation.
getty
Almost everyone is using AI now, but very few have adopted it. We’re conflating usage with adoption and wondering why we’re still not seeing the ROI we expected by halfway through 2026. McKinsey reports that 88% of organizations now use AI in at least one business function, yet only 7% have fully scaled it. That’s an adoption problem, not a usage problem. Adoption won’t show up in usage metrics; it shows up in the quarterly business review.
The Usage-Adoption Gap
Usage is often conflated with adoption because it’s the simplest metric to track. We often assume that widespread use of the enterprise LLM signals success, but it’s only the first step. True organizational adoption occurs when teams turn usage and learning into fundamentally redesigned work. The McKinsey survey also found that, among 25 organizational attributes examined, workflow redesign had the greatest impact on AI’s ability to enhance the bottom line.
Dave Grow, CEO of the work acceleration platform Lucid Software, sees the gap from inside a company that is navigating exactly this. “We’re hearing all these anecdotes about ‘I’m twice as effective’ or ‘we’re 50% more productive within engineering,'” he told me recently. He then asked the question those anecdotes don’t answer: “But are we seeing that yet at an organizational level?” Most leaders can’t say yes. It isn’t because their people haven’t learned the tools or because usage numbers are low. It’s that teams aren’t returning to first principles and asking the right questions, starting with “What are we trying to accomplish?” And then: how might we rethink the work with AI?
This Shift Is Different
Every major organizational shift we’ve led has been both technical and cultural. What’s different this time is the pace, scale, and uncertainty: we are learning the technology, teaching it, and rethinking the work itself, all at once. Change is hard, and decades of it have taught us lessons most organizations still haven’t applied. BCG recommends that companies allocate 10% of their AI efforts to algorithms (the models themselves), 20% to technology and data, and the remaining 70% to people and processes. But the 70% is not a license to skip the technology. Leaders need to keep learning the tools alongside their employees. Deciding what should change in people and processes requires knowing what the tools can do. We have learned that it takes an understanding of why we’re changing, what we’re changing, and how we’re changing. Most organizations and leaders still aren’t practicing it. AI is exposing that gap.
Grow sees this gap from the CEO seat. The opportunity, he told me, is “not to just automate what we are doing today” but “to really fundamentally redesign those core workflows.”
Designing The Work
Every team workflow falls into one of three categories: automation, augmentation, or amplification. Across all three, humans own the output. That principle has always been important and is even more so with AI. Glean’s 2026 Work AI Index has a name for what happens when leaders don’t put in place systems and practices to ensure the right outcomes: botshitting, which means shipping AI-generated work you haven’t verified, don’t fully understand, and couldn’t defend if asked. Glean reported that 69% of AI users admit to botshitting. AI slop is what happens when the task or the owner is never clear. The fix is to intentionally design the work into one of these three buckets: automate, augment, or amplify.
Automate when AI reliably delivers the outcome, the cost of an error is low, and the task isn’t worth a human’s development time. Manual data entry is a skill; it’s just not one worth protecting a task for. Most organizations are structuring for automation first, even as their employees already exercise judgment in their work. Automating a workflow before asking whether it should exist can save time in the short run, but it makes a bad process harder to unwind, iterate on, or redesign in the long run. Kevin Scott, CTO at PGA of America, put it plainly when we talked: “Look critically at things before you automate them. If you kill unneeded workflows first, your agents will be simpler and therefore more reliable.” Once there’s tooling built around it, the workflow becomes more deeply embedded in the way of doing things. Changing it later could cost more than the workflow ever did.
Augment when AI speeds up the work, but the risk lies in judgment, or when reviewing and refining AI’s output is how someone develops. The phrase “human in the loop” undersells what augmentation is. In the loop, a person approves what AI has already done. In the workflow, a person shapes the work as it happens, questioning, redirecting, deciding. Anthropic’s Economic Index Report, which has tracked millions of real conversations since early 2025, has consistently found that augmentation accounts for more than half of AI usage.
Amplify when the outcome requires human skills such as empathy, judgment, discernment, and emotional intelligence. As a team manager, I might automate scheduling and metric updates, but I want to amplify coaching, collaboration, and decision-making, as well as the values-based behaviors my organization expects of every employee. That’s what ‘good’ looks like here, compared with other organizations. That’s why culture shows up most visibly in the “amplify” decision and why companies with clear values-based behaviors make that decision faster.
Lucid set its core values at 30 employees and has held to them for more than a decade, through $300 million in revenue and 100 million users. One of them is innovation in everything we do: exploring new and better ways to deliver value and distill complexity into clarity. When AI arrived, that behavior guided decisions about how, when, and why to use AI effectively. Teams already practiced experimentation and questioning, so integrating AI tools was a natural part of their everyday processes.
Keeping work human is also how you develop employees, building the skills and behaviors they’ll need to scale with the technology, culture, and strategy. So why would we automate away the tasks that build the people we’ll need?
The Hiring Workflow
Take hiring. Most companies have already automated parts of it, and the automation question is the easy one: should req approvals and interview scheduling be handled by AI alone? Yes. AI handles them reliably; errors are easy to catch; and scheduling never builds anyone’s skills.
The augment question is harder: which tasks need a human in the work, not just in the loop? Screening early in the funnel is a candidate for automation. But as roles become more senior and job descriptions more nuanced, screening becomes augment work. The recruiter isn’t approving AI’s shortlist after the fact; they’re shaping it as it forms, questioning why a candidate ranked low, redirecting the criteria, and testing the patterns the tool found. That’s judgment exercised in the work, and it’s also how a junior recruiter learns what a strong candidate looks like.
The amplify questions focus on the candidate experience. Based on the organization’s values, where do the human touchpoints belong? Is the first point of contact with your company a person or an agent? Who decides who advances, and on what criteria? And where does the next generation of recruiters learn to read a candidate if they never sit in the early interviews? These are technological and cultural decisions, best addressed when the whole team reviews the entire workflow, from the screener to the hiring manager.
Create The Environment to Do Things Differently
Intentional design, technology aligned with the strategy, and values-based behaviors reinforced in the work occur only when employees feel psychologically safe. Most organizations aren’t there yet: in a controlled experiment by Atlassian’s Teamwork Lab, colleagues rated identical work as coming from someone ten times lazier when AI use was disclosed, and were 24 percentage points less likely to recommend that person for high-visibility projects. In many organizations, transparency about AI still works against people. Decades of research by Harvard Business School professor Amy Edmondson,have shown that team innovation and real change depend on psychological safety, the shared belief that a group is safe for interpersonal risk-taking.
Scott highlighted this exact condition: “You have to create an environment where people have the psychological safety to think about doing things differently.” Questioning how work is currently done and how it could be done better is inherently risky. The organizations that get this right minimize that risk by setting clear guardrails, reinforcing open communication, and rewarding employees who actively integrate AI into their workflows.
The first half of 2026 was spent measuring usage. The second half belongs to the teams willing to question and redesign their work.