The Invisible Tax: Why AI Costs Are Reshaping Enterprise Strategy

In this conversation, Gregor Vand and Sean Falconer map the shifting dynamics of the AI industry. They move past the hype to reveal a hidden economic reality: we are entering an era of AI compute taxes. While companies currently treat AI spending as a necessary investment to avoid appearing obsolete, this phase is temporary. The real consequence is a transition from growth at all costs to a rigorous, seat based SaaS model where token usage becomes a line item scrutinized as closely as headcount. For leaders, the advantage lies not in adopting AI, but in mastering the agentic harness, which is the architecture that manages context. This allows them to avoid the trap of paying for high end models when lower cost alternatives suffice. This analysis is for CTOs and engineering managers who need to prepare for the shift from innovation fueled spending to operational margin optimization.

The Hidden Cost of Good Enough

The competitive landscape of AI has shifted from a race for the best model to a race for the best agentic harness. As Falconer notes, the underlying models have reached a plateau of performance where, for many tasks, the differences between top tier and mid tier models are negligible. Yet, teams often default to the most expensive model out of habit or a lack of oversight.

The true moat has what to do with through the context management of the information environment around the model rather than just the base model.

-- Sean Falconer

This creates a systemic inefficiency. Developers and non technical users alike are running high cost models for low value tasks, compounding monthly bills into the tens of thousands. The downstream effect is a compute tax on every employee, which changes the unit economics of hiring.

Why the Obvious Fix Makes Things Worse

Conventional wisdom suggests that AI will lead to massive headcount reductions. However, the systems level response is more nuanced. As Vand points out, when tools make individual output faster, the system simply raises the expectation for total output. Rather than a reduction in work, we see a marathon of sprints.

Everything that I output has to be taken in by somebody else on the team somewhere. And so we have got this debate going on right now of like well, where is the line between you cannot just keep churning out documents expecting a human to keep reading them?

-- Gregor Vand

This creates a feedback loop where AI generates content that requires more AI to summarize, effectively bloating the information environment without increasing net value. The immediate benefit of individual efficiency is offset by the hidden cost of collective information overload.

The 18-Month Payoff: From Innovation to Optimization

We are currently in an early innovation phase where enterprises are willing to absorb massive AI bills to signal progress. But as these companies move toward IPOs or face tighter margins, this behavior will flip. The history of cloud computing serves as a blueprint: initial lack of oversight eventually gives way to dedicated teams, and eventually, built in platform features, focused solely on cost optimization.

The companies that will win are those currently building the routing infrastructure. These systems dynamically shift prompts to the most cost effective model based on task complexity. This requires upfront investment in infrastructure that offers no immediate shiny feature, but creates a significant competitive moat by protecting margins when competitors are forced to reckon with unsustainable token bills.

Key Action Items

• Establish Token Visibility (Immediate): Start tracking token usage per department or role. You cannot manage what you do not measure, and the blind spending phase is nearing its end.
• Implement Model Routing (Next Quarter): Move away from a default to best model policy. Invest in infrastructure that routes simple tasks to cheaper models and reserves high end models for complex reasoning.
• Audit AI Generated Workflows (Next 6 Months): Evaluate if your AI driven output is actually creating value for the recipient or just adding to the noise. If the output requires another AI to summarize it, the process is likely flawed.
• Prepare for Compute Tax Budgeting (12-18 Months): Factor AI compute costs into your long term hiring and project planning. Treat these costs as a recurring overhead similar to insurance or benefits.
• Shift Focus to Context Management (12-18 Months): Invest in your internal agentic harness, which is how you organize, retrieve, and feed data to models. This is where your long term defensibility lies, far more than in the specific model provider you choose.