Andrew Avanessian, CEO at Haiilo.
A piece published by a16z made a claim worth sitting with. George Sivulka’s “You Just Hired a Million Bad Employees” opened with a provocation that cuts directly to the economics of enterprise AI: “For the first time in history, humans are cheaper than software.”
AI productivity gains are real, but the costs of sustained usage are real, too. While your teams celebrate gains and your board asks about returns, invoices are quietly building in the background.
Productivity illusions mask cost problems.
Pilots work. A team deploys an AI tool. A day-long process takes hours. The case study writes itself. I have seen this at Haiilo. Request for proposal responses once consumed days of specialist resources. They now take hours. I am certainly not questioning the gains, but what a pilot rarely surfaces is how cost structures scale with usage.
Traditional SaaS licensing is predictable. You pay-per-seat, and finance can model it. Token-based AI consumption is a different beast. It scales with activity. It’s what makes it so powerful, and it’s what makes it dangerous when underlying workflows are poorly designed.
The moment you embed an AI system into a high-frequency workflow, you are no longer buying a tool. You are commissioning a workforce that bills hourly, with no visibility into whether those hours are being spent productively.
Agentic loops are the AI version of ghost spend.
Sivulka explores a concept that anyone running agentic workflows should understand: the loop.
Loops occur when an AI agent calls itself repeatedly to compensate for an inadequately specified task. The agent cannot complete what was required, as the task wasn’t articulated clearly enough for it to succeed. So it retries and retries, each retry consuming tokens and each token costing money.
The problem is that most organizations have not built the internal discipline to clearly define tasks enough for AI to execute them well. Employees may be adept at natural language prompts but inexperienced in building sophisticated workflows.
The financial implications quickly become significant. If 80% of tokens are consumed unproductively, the economics of that deployment look nothing like the pilot stage models.
The counterintuitive cost comparison is worth understanding.
For years, the argument against SaaS sprawl was that businesses paid for seats nobody used. The waste was in underutilization. Token-based AI consumption inverts this. Waste is in overutilization—AI systems running at full capacity on ill-defined tasks, generating cost without generating value.
Legacy SaaS had a natural cost ceiling. Nobody accidentally spent three times the contract value by running a spreadsheet too many times. Token-based models have no such ceiling. Cost scales directly with usage, with Gartner predicting global end-user spend on AI will reach $64 billion in 2026, up from $39 billion in 2025. Usage scales directly with the number of workflows connected to the system.
An enterprise with AI embedded into core workflows, each running hundreds of times a day with agentic loops triggered by unclear prompts, can accumulate token spend that exceeds what a comparable SaaS contract would have cost in years by Q2.
This should not slow AI adoption. It should enforce rigor in how systems are designed before deployment scales.
Here’s what good AI economics actually look like.
Businesses managing AI costs well are those treating prompt discipline and task definitions as operational priorities. They build evaluation frameworks before scaling workflows, not after.
At Haiilo, this has shaped how we approach AI. We are building a knowledge graph that takes signals from across the platform, understanding context at a structural level rather than users needing to reconstruct context from scratch. The intent is to increase precision of AI interactions, reduce waste, improve outputs and make the economics of sustained usage defensible. It is a deliberate architectural and commercial choice.
The businesses that extract durable AI value are not necessarily those with the most sophisticated models or most aggressive deployment timelines. They are those that encode their own processes into AI with enough precision that the systems run cleanly, at scale, without leaking cost through poorly defined loops.
This requires deep knowledge of your own workflows, the right people building them, the discipline to articulate them clearly and the governance to measure what is actually happening when systems run.
There are three questions leadership must ask.
The organizations navigating AI most effectively ask these questions before expanding usage:
1. What do productive token interactions look like?
If teams cannot describe what good output looks like for a given workflow, they cannot identify waste. If they cannot identify waste, they cannot stop it. Define success criteria before deployment to make sure the economics hold.
2. Are we measuring token efficiency alongside task completion?
Most AI dashboards surface completion rates and time saved. Fewer track the ratio of tokens consumed to outcomes produced. That is where the cost lies. When trending in the wrong direction, the problem is almost always task definition, not model performance.
3. Who owns the costs?
In most enterprises, AI spend typically sits in technology budgets. As token costs compound, that needs to change. Finance teams must understand consumption levels. Business unit leaders need to understand the workflows driving it. The gap between who controls usage and who sees the bill is where unseen cost challenges develop.
AI economics require commercial discipline.
Humans being cheaper than software is a striking inversion of decades of assumptions. But the more operationally important insight is how good tokens, deployed well, are cheaper than humans at scale. The key word being “well.”
While AI has made it appear simple to build custom systems from scratch, when done badly, it can cost significantly more than anticipated.
As I’ve stated previously, the organizations that get AI usage right treat governance as a commercial discipline, not a technical one. Deployments are instrumented with the same rigor applied to headcount or capital allocation. Internal capabilities are devised to define tasks clearly, evaluate outputs consistently and identify inefficiency before costs compound.
The question, then, is whether leadership teams running AI at the enterprise level have built the systems to tell the difference and the governance to act on what those systems reveal.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?