McKinsey consultants sat down this year with executives at 15 companies widely regarded as the sharpest users of AI. The opening line of the report that came out of those interviews is almost deflating: nearly every one of these companies owns AI tools, but hardly any of them have learned to use those tools well. If you were hoping for a tidy playbook, that's a strange place to start.
What makes the report worth reading isn't a list of best practices — it's a list of shared mistakes. Even companies that get called "AI-native" trip over the same things. The consultants distilled the pattern into seven recurring failures, and what's notable is how many of them have nothing to do with the tools themselves, or the budget behind them. There's a real gap between stacking up tools and actually digesting them — and that gap is where most organizations are getting stuck right now.
Mistaking motion for speed
The first failure the report identifies is a speed illusion. Teams assume that once AI can produce a first draft, the whole workflow speeds up. In practice, it plays out differently. More time ends up going into reviewing what the AI produced, editing it, deciding which parts to keep, and communicating all of that within the team. Some parts of the process do get faster. But other parts get slower at the same time. If the gap between how fast the tool generates and how fast people can absorb it never closes, overall productivity can end up falling short of expectations.
The second pattern is know-how getting trapped with individuals. Whenever a new AI tool shows up, it's usually one or two enthusiastic people on the team who adopt it first. The problem is that their experience stays personal. Six months later, the same handful of people are still using the tool the same way. AI fluency ends up anchored to individuals rather than becoming a team capability — and the productivity gap between the people who use the tools and the people who don't just widens over time.
Third is verification avoidance. If a company never explicitly designs a process for checking AI output, people quietly start trusting that output by default. When an error surfaces, the instinct is to say "the AI got it wrong" — but the deeper problem is that no one designed a verification process in the first place. Among the 15 companies, the gap between the ones that built this structure explicitly and the ones that didn't showed up less in error rates than in how fast errors got caught.
The remaining patterns follow the same logic: data siloed by department, so every AI project requires renegotiating access from scratch; unclear ownership over who actually has the authority to act on an AI-generated recommendation; AI adoption read internally as a signal of headcount cuts, which starves the people who should be managing the tools of the time to do it properly; and tool fatigue, where so many new tools keep getting added that simply deciding what to use where becomes its own burden.
What's striking about all seven, taken together, is what's missing: no technical complaints. Nothing about GPT-4 being slow, APIs being unreliable, or costs running too high. Every single failure comes down to how people and organizations absorb the tools, not the tools' capabilities. Now that AI performance itself has cleared the bar, the bottleneck has shifted — away from the tools and onto the habits around them.
The skeptics
Among growth-stage startup CPOs who've read the report, a common reaction is that the seven principles are already written in big-company language. Verification frameworks and data-governance debates that assume an organization of hundreds of people, they argue, don't translate cleanly to a three-person team or a solo founder — applying them there just means spending more time designing process than actually doing the work. That criticism isn't wrong.
There's a more fundamental objection too: that codifying principles like operational gospel can itself suppress experimentation and iteration. There are real examples of organizations that skip structure-first design, fail fast, and adjust — and grow faster in their AI use because of it. The McKinsey report itself doesn't fully dismiss this. Some of the 15 companies said their AI culture emerged organically, without any formal guidelines. The principles didn't come first; the patterns formed after repeated experimentation.
Separately, there's a fair critique that none of the seven truths were actually measured. An interview-based report leans entirely on executives' self-perception. There's no controlled data in the report showing how much knowledge-hoarding or skipped verification actually costs in performance terms. A sample of 15 companies is also inherently limited, and the report never clearly defines what counts as "ahead" on AI adoption in the first place.
None of that skepticism amounts to an argument for throwing the report out, though. The seven patterns are more useful as a diagnostic after the fact than as a blueprint beforehand. Used as a reference point for asking "where is my team actually stuck right now," the big-company framing matters a lot less.
Is your AI routine actually doing anything?
It's true that the McKinsey report speaks in big-company language. But there are threads in it that connect directly to solo operators and small teams.
It's worth checking whether the AI tools you're using are actually shaping decisions, or just sitting there. If your use of AI stops at generating a first draft, you're not really using the tool — you've just installed it. Mapping out which judgment calls you're willing to hand to AI, and which ones you insist on making yourself, reveals a lot about your actual level of use. A good starting point: track how much you're editing an AI-drafted proposal, and on what basis you're making those edits.
Your verification routine is worth a check too. Without an explicit standard for reviewing AI-generated content, plans, or summaries, the review step tends to quietly disappear over time. For a solo operator, this doesn't need to be an elaborate internal system. Simply deciding in advance on "the three things I personally have to verify in this draft" is enough to count as a verification routine. One simple checklist can make a noticeable difference in the quality of a blog post or a proposal.
Your tool count is also worth auditing. Every time a new AI tool launches and you bolt it on, you spend a little more energy just figuring out what to use where. The tool fatigue the McKinsey report describes isn't a big-company-only phenomenon. Reviewing your current tool list once a quarter, and cutting anything that isn't actually contributing to a decision, helps sustain your level of AI fluency rather than eroding it.
There's a recurring theme in discussions about what capability remains distinctly human in the AI era. As a tool's feature set expands faster and faster, the gap widens around a different skill: judging in what context, and up to what point, that tool should be used. That's exactly why simply spending more hours using a tool doesn't close this gap. The ability to read context and know when to hold off is what actually creates the difference in capability — and that's a different layer entirely from how comfortable you are with any one piece of software.
What McKinsey observed across the 15 companies points the same direction. The differentiator wasn't which tool a company used — it was what the company decided to do before picking up the tool at all. The teams called "AI-native" weren't the ones that shipped every new feature the moment it launched. They were the ones that decided in advance which tool served which purpose, and had a routine for adjusting quickly when that decision turned out to be wrong.
