McKinsey's AI 'two-speed' race isn't about budget. It's about who says no.

McKinsey's State of AI 2026 survey, published this week from 1,719 respondents across 97 countries, found something that should worry every executive who greenlit an AI budget this year: adoption is up sharply, and the share of companies generating significant financial value from it is stuck at 6% — the same number as last year.

The adoption numbers alone look like progress. Among companies with more than $1 billion in revenue, 40% are now scaling AI agents, up from 27% a year ago. For coding agents specifically, 31% of large enterprises are scaling versus roughly 20% of organizations overall. Everyone is spending. Almost nobody is capturing value proportional to that spend.

McKinsey's own read on the split is the interesting part: the companies pulling ahead aren't the ones running more pilots. They're the ones redesigning how work actually gets done around AI, instead of layering another tool onto an unchanged process. Everyone else is running the same workflow with an AI assistant bolted to the side, wondering why the ROI never shows up.

Spend isn't the bottleneck. Selection is.

Here's the part that doesn't get said out loud in surveys like this: if only 6% of initiatives are generating real value while adoption climbs every quarter, the other 94% aren't failing quietly in a corner. They're consuming engineering time, review cycles, vendor contracts, and change-management bandwidth that could have gone to the 6% that actually mattered — before anyone had proof they'd work.

That's not a technology gap. It's an intake failure. Somewhere between "let's try an AI agent for this" and "we're now running twelve of them across four departments," nobody was scoring which ones deserved to scale and which ones deserved to be killed at week three. The default became yes, and yes doesn't discriminate between the initiative that redesigns a workflow and the one that just makes an old workflow look busier.

The organizations McKinsey found moving faster didn't get there with a bigger AI budget. They got there by being willing to not fund most of what crossed their desk — and by having a clear enough set of criteria that saying no didn't require a debate every time.

Triage is the unglamorous half of "AI strategy"

Nobody wants their AI strategy deck to say "we rejected 80% of proposals." It sounds defensive, not visionary. But the 6% number says the opposite framing is the honest one: the companies capturing value are the ones who treated every AI initiative as a request competing for scarce resources against every other request — not a foregone yes because the word "AI" was in the title.

That requires the same discipline as any other intake process: a defined set of criteria (does this redesign a workflow or decorate one?), someone accountable for scoring against it, and a willingness to kill projects that don't clear the bar, even after they've already started. Most organizations have none of that for AI initiatives specifically, because AI got fast-tracked around the normal approval process in the rush to "not fall behind." That rush is exactly how you end up spending more and capturing the same 6% two years running.

If your organization is trying to close its own version of this gap, the fix isn't a bigger AI committee — it's a scoring model applied consistently at the point where requests come in, before they've eaten a quarter of engineering time. That's the whole premise behind how Admisio handles intake: every request, AI-flavored or not, gets scored against the same weighted criteria before it earns a spot on the roadmap.

Sources: McKinsey Report: Enterprise AI Is Becoming a Two-Speed Race, AIwire, September 2, 2026; McKinsey & Company, "The State of AI: Global Survey 2026."