Our Take
The infrastructure vendors are right to flag this, but they're also the ones selling you the technical fix while the organizational redesign stays someone else's problem.
Why it matters
Most AI investment to date has flowed toward model development and infrastructure. The hard part—retraining staff, rewriting processes, changing incentives, managing layoffs—requires a different skill set and budget line. This gap explains why many deployments stall after initial pilots.
Do this week
CTO/CRO: Schedule a work-design audit with ops and HR this week so you can separate which AI projects require process change from which ones don't, and resource accordingly.
The org design problem nobody wants to solve
Okta's President and Chief Operating Officer raised a claim at a corporate level: companies pursuing AI adoption are systematically underestimating or ignoring the organizational redesign required to make it work. The thesis is straightforward: deploying a model is the easy part. Restructuring roles, workflows, decision rights, and compensation to actually use that model is where most efforts falter.
This is not a new observation. McKinsey, BCG, and Deloitte have published versions of this argument for two years. What makes Okta's executive statement worth attention is that it comes from a platform vendor whose revenue depends on companies buying more software, not less. Okta has no incentive to tell customers that their bottleneck is organizational rather than technical.
Infrastructure can't fix process denial
The AI gold rush has been dominated by technologists and investors. Model capability, inference cost, latency, and security are measurable and vendorable. Work redesign is not. There is no SaaS product labeled "Org Redesign Platform." There are no benchmarks. HR and operations leaders are rarely given the budget authority that engineering receives for AI infrastructure.
This creates a predictable gap: companies hire ML teams, deploy models in controlled environments, see promising pilots, then attempt to scale without the supporting process changes. The model works in the lab. It fails in production because nobody changed the approval chain, retrained the staff, or removed the incentive for the old workflow.
Okta's statement is a warning sign that this pattern is visible at the customer level. If adoption is genuinely stalling because of org design, not capability, then the infrastructure vendors themselves will eventually face flattening demand. That framing makes the COO's candor credible.
Audit your org before you audit your model
Before committing to a production AI deployment, map the work it will change and the people who will be affected. Identify which roles will shrink, which will shift, and which will be created. Estimate the retraining cost and timeline. Get explicit budget sign-off from the executive sponsor for both the technical infrastructure and the organizational change management. Do not assume that the model deployment and the process redesign can happen on separate timelines.
If your organization is structured to move fast in engineering but slow in operations, that asymmetry will break your AI strategy. The technologists will be ready first, and then waiting. Align the paces upfront.