The $100M Deal
~ Our global team was partnering with a global system integrator. We were shortlisted at a major healthcare provider for their end user compute transformation.
On the surface, we were swapping platforms with minimal behavior change with a nearly identical end user experience. Of course, our platform had the technical edge. In reality, we had seven layers of enterprise coordination across a small army or IT, operations, finance, security, project management, executive sponsors and end users.
Fast forward to today's AI adoption initiatives. The complexity is deeper, with end users left to imagine the 'art of the possible' with a toolset mistaken for a super search engine. However, now the stakes are higher, building foundations for enterprise reinvention.
Anthropic's CEO Dario Amodei even changed his narrative on the dynamic, shifting to ‘we still need white collar roles to reengieneer enterprise workflows’. Using Jevons Paradox, he cited efficiency gains expand demand rather than contract it.
ie. Automate 90% of the job, and the remaining 10% expands to fill all bandwidth. Between Jevons Paradox and actual productivity gains sits the ‘messy middle’: The workflow redesign, governance, trust, training, executive alignment, procurement, risk, and behavior change.
The recent investments signal commitment to this shift as well. OpenAI announced their $4B joint venture Deployment Company, with 19 PE investors, valued at ~$10B. Anthropic announced their move developing a JV enterprise AI services company with Blackstone, Goldman Sachs et al contributing ~$1.5B.
But who owns the messy middle?
Many enterprises will get the call from these transformation delivery organizations to solve the challenge. Here's how to be ready:
1. The coordination tax is real: Enterprise AI is a change management problem, not technology. Map stakeholders, governance, and decision paths before you buy.
2. The 10% to 90% principle is permanent: When AI automates 90%, the remaining 10% balloons. A legal team deploying contract review AI is expecting freed time. Instead, 40% of their day is reviewing edge cases, 30% handling exceptions, 20% training the model.
3. Get your data ready: Improve data quality, lineage, provenance tracking, and cleanliness standards. But don't let this become a massive governance exercise that delays rollout another six months.
4. Your decision making speed is your limiting factor: AI generates insights in real time. Refine your AI governance policies, implement ethic frameworks. Flatten approval hierarchy, push decisions closer to the work.
5. Uplift human roles before eliminating headcount: Enable existing FTEs to become prompt engineers, auditors, edge case handlers, training specialists.
Saanya Ojha, captured this shift -- "The irony is rich: AI was supposed to automate knowledge work. Its next act is hiring knowledge workers to help knowledge workers use tools that automate knowledge work."
![]()
![]()
11
4 Comments
