AI Investment Is ‘Harder to Justify’ as Productivity Returns Lag, Uber COO Says

The AI Productivity Paradox: Why Massive Spending Isn’t Always Equal to Faster Growth

For the last few years, the corporate playbook for AI has been simple: spend aggressively, integrate everywhere, and hope for a productivity miracle. But as the honeymoon phase fades, a sobering reality is setting in for some of the world’s biggest tech players. The gap between AI investment and tangible output is becoming a chasm.

Take the recent situation at Uber. Despite an aggressive push into AI-assisted coding, the company found itself in a financial bind. Uber’s CTO, Praveen Neppalli Naga, revealed that the company exhausted its entire 2026 AI budget in just four months. The culprit? The explosive adoption of Anthropic’s Claude Code among its engineering staff.

While the tools are being used at scale, Uber’s COO Andrew Macdonald has pointed out a critical missing link: the direct correlation between those “burned” tokens and actual features shipping to users. This is the AI Productivity Paradox—the phenomenon where companies see a surge in activity (more code written, more emails sent) but no proportional increase in business value.

The “Token Trap” and the Future of Enterprise Budgeting

Traditional corporate budgeting is built on predictability. You know the cost of a software license or a yearly salary. But generative AI introduces a variable cost model based on tokens—the fragments of words that LLMs process. When a tool like Claude Code becomes a staple for thousands of engineers, costs can spiral in ways that traditional finance teams aren’t equipped to handle.

We are likely moving toward a new era of AI Financial Operations (AIOps). Future trends suggest that companies will shift away from “all-you-can-eat” AI access toward more disciplined strategies:

• Tiered Access: High-cost, “frontier” models (like Claude 3.5 or GPT-4o) will be reserved for complex architectural tasks, while smaller, cheaper, distilled models will handle routine boilerplate code.
• ROI-Gated Spending: Instead of blanket budgets, AI spend will be tied to specific KPIs, such as “reduction in bug reports” or “increase in deployment frequency.”
• Custom SLMs: To avoid the “token tax” of third-party providers, more enterprises will build Small Language Models (SLMs) trained on their own proprietary data, hosted on internal infrastructure.

The goal is to move from spending for the sake of adoption to spending for the sake of outcome. As Forbes recently highlighted, token pricing is fundamentally breaking enterprise finance assumptions.

Will Agentic AI Actually Disrupt the Platform Economy?

There has been a lingering fear among platform-based businesses like Uber and DoorDash: the rise of Agentic AI. The theory is that users will stop opening apps entirely and instead tell a personal AI agent, “Get me a ride to the airport and order a healthy lunch for 1 PM.”

If this happens, the “front end” of the business—the brand, the user interface, and the loyalty loops—disappears. The platform becomes a mere utility, a “pipe” that the AI agent plugs into. This could lead to a race to the bottom on pricing, as AI agents optimize for the lowest cost rather than brand preference.

However, the reality is proving slower than the hype. Uber’s leadership noted that while they are working with large model companies, the “disaggregation” of commerce hasn’t happened yet. The transition to an agent-led economy will likely be iterative rather than an overnight collapse.

The Human Cost: AI Investment vs. Headcount

One of the most contentious trends is the trade-off between AI spending and human hiring. When AI costs skyrocket, some companies are slowing down hiring to balance the books. The logic is that AI should eventually replace the need for those new hires.

But this creates a dangerous tension. If the productivity gains aren’t tangible—as Andrew Macdonald suggested—then cutting headcount to pay for AI is effectively trading a proven asset (human talent) for an unproven one (AI efficiency). This “productivity gap” could lead to a correction in the labor market where companies realise that AI is an augmenter, not a replacer.

For more on how this affects the broader tech landscape, check out our analysis on AI workforce trends and the evolution of the software engineer.

AI ROI FAQ

What do you think? Is your organization seeing a real productivity boost from AI, or are you just seeing a higher cloud bill? Join the conversation in the comments below or subscribe to our newsletter for weekly insights into the intersection of AI and business strategy.