The fastest train in the world still needs working rails. That is the central argument Sastry Durvasula, Chief Operating Officer at TIAA, and Manish Sharma, Chief Strategy and Services Officer at Accenture, make in a joint commentary published August 20, 2026 — and the numbers behind their own project give it weight.
Nearly two years ago, the two organizations partnered to modernize TIAA's recordkeeping infrastructure. TIAA is 108 years old and, by the authors' own account, carries the technical debt to prove it. Before any AI deployment could scale, the team had to clean data, retire outdated systems, and redesign how work actually flows through the organization.
The results are concrete. Plan sponsors can now change investment options for employees' retirement plans in days instead of weeks. Digital engagement across TIAA's millions of participants has risen 13%. Neither outcome came from a headline-grabbing demo. Both came from what Durvasula and Sharma call 'the unglamorous work most companies skip.'
The diagnosis is blunt. Most enterprises today are 'stacking agents on top of decades of legacy IT, siloed data, and broken workflows,' the authors write. The consequence is not acceleration — it is dysfunction automated at higher speed. Only a fraction of companies, they argue, turn AI investment into measurable business impact.
The prescription follows logically: treat AI transformation not as a technology project but as a business transformation, a change-management project, and an operating-model rebuild that happens to run on AI. Companies that bolt AI onto an unreformed core, they warn, will spend years chasing pilots that never scale.
Durvasula and Sharma identify clean data, a modernized core, and workflows rebuilt for how AI actually works — not how it is marketed — as the separating factors between enterprises pulling ahead and those stuck in 'perpetual pilot mode.'
Their closing frame is direct: 'In the AI era, complexity is a competitive disadvantage that can no longer be hidden behind a sizzling AI experience.'
---
The market lesson here is straightforward. Capital allocated to AI without first fixing the underlying operating model is not investment — it is expensive theater. The TIAA-Accenture case is a useful corrective to the vendor-driven narrative that more compute and more agents automatically translate into more productivity. They do not. Clean data and redesigned workflows are unglamorous line items; they do not generate conference keynotes. But they are precisely the kind of structural, disciplined work that free enterprise rewards over time — and that separates companies building durable competitive advantage from those burning shareholder capital on pilots that never leave the demo stage. The 13% engagement gain did not come from a press release. It came from doing the boring work first.
