88% AI Adoption, 6% Value Creation: The Enterprise Agent Thesis Has a Maturity Problem
textak sits at 77% on autonomous agents being widely deployed in enterprise workflows — our highest-conviction enterprise AI position. Today's McKinsey data introduces genuine friction: 88% of organizations use AI in at least one business function, yet only 6% qualify as high performers capturing more than 5% of EBIT. Nearly two-thirds remain stuck in pilot purgatory. This isn't a weak signal we can wave away. It's a direct challenge to what 'widely deployed' means, and we owe readers an honest accounting.
Our 77% is grounded in observable deployment momentum — GitHub processing 275 million commits per week driven by autonomous agent activity, SpaceX acquiring Cursor for $60 billion in a deal that only makes sense if AI coding agents are genuinely central to infrastructure, cloud providers shipping agent frameworks at scale. These are production signals, not pilot signals. The GitHub infrastructure story is particularly telling: Microsoft routing traffic through AWS because AI coding agents overwhelmed the platform isn't a pilot program stress-testing a concept. That's a capacity crisis caused by production-scale agent deployment.
But here's where the McKinsey data lands a real punch: 'widely deployed' and 'widely generating value' are different claims, and we may have been conflating them. Our forecast target says 'widely deployed in enterprise workflows' — which technically could resolve YES even if most deployments are underperforming. A Fortune 500 company running agents in production for code review and IT ticketing has cleared our bar, even if those agents aren't yet showing up in EBIT. The 6% high-performer figure is damning for the economic thesis, but may not directly falsify the deployment thesis. We should be cleaner about which we're actually forecasting.
The honest tension is this: if two-thirds of enterprises are stuck in pilot purgatory, then the agents that ARE in production are concentrated in a relatively small number of sophisticated organizations. 'Widely deployed' starts to look like 'deployed at the vanguard' — which is a materially different claim. Sophisticated readers will notice this. The McKinsey data suggests the deployment wave is real but narrow, not broad-based. Our forecast language doesn't fully account for that distinction, and that's a genuine gap.
What we're watching: the Gartner projection that 40% of agentic AI projects will be canceled is the figure that keeps us up at night more than McKinsey's 6%. Cancellation is a harder stop than underperformance. If Q3 enterprise earnings show broad agent project cancellations rather than continued investment, we'd move from 77% toward the low 60s. What would push us toward 85%: if the GitHub commit trajectory holds — 14 billion annual commits versus 1 billion in 2025 — that's a 14x increase that isn't explained by anything other than agent deployment at scale. We'd want to see that confirmed in the next quarterly GitHub metrics release.