All insights

The model labs just told you where the bottleneck is.

OpenAI and Anthropic are building deployment organizations to help companies actually implement AI. That's not a product decision. It's a diagnosis of the market.

OpenAI and Anthropic both made a notable move this quarter: they're no longer just selling AI software. They're building deployment organizations, embedding engineers, standing up implementation teams, partnering with consultancies, and effectively helping rebuild operating models inside client companies.

That's an important signal. For the past two years, the narrative around AI has largely been: "The models are here; enterprises just need to adopt them." But enterprise adoption turns out to be much more operationally complex than that. Most organizations don't have clean workflows, clear ownership structures, standardized processes, or the internal capability required to integrate AI effectively at scale. And AI alone doesn't solve any of those problems.

This isn't a software rollout

Instead of waiting for companies to figure it out on their own, the major labs are moving closer to the implementation layer. What they're doing there looks less like a traditional software rollout and more like operational transformation work, the unglamorous business of workflows, handoffs, and decision rights.

The challenge was never just "getting the model." It was integrating AI into real processes, real teams, and real decision structures. That's slower and more labor-intensive than the early narrative implied, which is exactly why the companies selling the models are now investing in the layer where adoption actually stalls.

Why process work matters more, not less

As AI capabilities accelerate, organizations still need clear workflows, defined decision ownership, standardized operations, strong change management, and intentional system design. Without that foundation, AI tends to amplify inconsistency rather than eliminate it. Point a model at five versions of the same process and you get five automated versions of the same disagreement.

The next phase of enterprise AI adoption probably won't be won by whoever has the best model. It will be won by whoever best operationalizes the technology inside real-world systems.

For operators, there's a practical takeaway: readiness is buildable before the vendors arrive. The workflows, the ownership, the map of how work actually happens, that's yours to own. Own it first, or someone else will draw it for you and bill accordingly.