Picture this. You’ve just finished watching an Instagram video. Some founder, creator or AI guru is showing how they built an app in twenty minutes. No engineers. No code. Just prompts. It looks easy, almost absurdly easy. You get inspired, grab your coffee, open your laptop, take the first sip and for a brief moment it feels like the barriers that used to exist between an idea and execution have almost disappeared.

That moment is becoming increasingly common, and it says something important about where we are with AI. The cost of experimentation has collapsed. What used to require technical teams, planning cycles and meaningful capital can now begin with curiosity, a browser, and a subscription. The distance between thought and action has never been shorter.

Kate Minogue kept pulling the conversation back to this point during our recent EVOLVE session. The easier it becomes to build, automate or generate, the easier it becomes for companies to confuse activity with progress. That confusion is not new. Business has always been vulnerable to it. What AI changes is the speed. Teams can now launch workflows, spin up agents, test products and generate outputs at a pace that makes governance feel slow by comparison.

That is where the tension begins. Waste no longer looks like inactivity. Increasingly, it looks like motion.

For most of business history, waste had natural limits. Hiring someone took time. Buying software required procurement. Building systems demanded budget, specialists and approvals. Even bad decisions moved slowly because friction acted as a filter. You could still waste money, but the process of wasting it was heavier, more visible, and easier to interrupt.

AI removes much of that friction.

Kate made the point clearly: token costs, model usage and agentic workflows can scale quietly and compound fast, often without the same operational scrutiny companies would apply elsewhere. What makes this different is not only the technology itself, but how easily it spreads. One employee experiments, another team copies it, another department adopts a similar workflow, and before long the organization has multiple layers of AI activity with no shared visibility into cost, duplication or output.

This is where the economics start to shift. Compute is cheaper than headcount in some cases. That sounds efficient. But efficiency without coordination can become its own cost center. A company may save time in one workflow while quietly creating five others that overlap, repeat or solve problems that did not need solving in the first place.

That is where AI starts resembling earlier infrastructure cycles. Cloud did this. SaaS did this. The difference now is speed. AI compresses the time between curiosity and cost. That changes how quickly waste can accumulate before anyone notices.

Part of the problem is that most organizations still do not treat AI as infrastructure. They treat it as tooling. That distinction changes how costs are perceived. Software budgets are usually centralized. Headcount is tracked carefully. Infrastructure spend tends to have governance attached to it. AI often enters through a different route. It starts at the edges. Individual teams test models. Employees buy subscriptions. Product managers experiment with workflows. Marketing automates content. Customer service tests agents. None of it feels large in isolation.

Kate described this as one of the more dangerous patterns emerging right now: fragmented experimentation creating real cost without centralized oversight.

That fragmentation creates two problems at the same time. The first is financial. Costs spread across departments before they are understood as one system. The second is operational. Multiple teams often end up solving similar problems in parallel, building overlapping workflows or introducing unnecessary complexity into the stack.

This is not very different from what happened during the SaaS boom. Companies accumulated tools faster than they accumulated process. The result was software sprawl, duplicated functionality, and budgets that expanded without proportional productivity gains. AI may be repeating the same pattern, but with a steeper curve.

That changes the role of leadership. The challenge is no longer deciding whether AI should be used. In many organizations, that decision has already been made informally. The harder question is who owns the system once usage becomes widespread..

Kate introduced an idea that feels increasingly inevitable: AI Ops. The term sounds new, but the logic is familiar. Every major technology wave eventually creates its own management layer. Cloud created CloudOps. Software at scale created DevOps. Data growth created entire governance functions. AI appears to be moving in the same direction.

That progression matters because the problem is no longer access. Most organizations already have access. The problem is operational literacy. Who understands which model should be used for which task? Who decides when an expensive model is justified and when a simpler one is enough? Who monitors whether an agent running autonomously is actually improving output or simply generating more internal noise?

These are not technical questions alone. They are management questions disguised as technical ones.

That changes how companies should think about adoption. The market has spent the last two years treating AI as a productivity layer. Kate’s argument pushes it closer to infrastructure. Infrastructure requires discipline. It requires controls. It requires someone accountable for efficiency, risk, security and redundancy.

Without that, organizations drift into a dangerous middle ground. Enough AI to create complexity. Not enough governance to create clarity.

And that is usually where costs begin to outpace value.

The governance problem does not stop inside the organization. It extends outward into the models companies choose to build on. That part of the discussion became particularly sharp when Kate introduced the idea of AI interdependency. The more businesses embed third-party models into customer products, internal systems or commercial operations, the more they inherit the risks attached to those providers.

This is not only a pricing issue. It is a resilience issue.

A model provider can change access, pricing, speed or policy with little warning. Regulatory shifts can alter what is permissible across regions. Geopolitical tensions can reshape availability entirely. Davor pointed to the growing concern around European dependency on American and Chinese models, and the operational uncertainty that creates for companies building critical systems on top of them.

That introduces a strategic layer many companies are still underestimating. Buying intelligence from outside may feel efficient at first, much like cloud felt efficient in its early expansion. Over time, however, dependency changes bargaining power. The more embedded the system becomes, the harder it is to replace.

That creates a second governance question. It is one thing to manage how AI is used inside your business. It is another to manage what happens when the intelligence layer your business depends on sits entirely outside your control.

Kate kept bringing the discussion back to a principle that sounds simple and is often ignored: start with the business problem. That sounds obvious until you look at how many AI initiatives begin the other way around. A new model appears, a new tool launches, a team gets excited, and only afterwards does the company start searching for where it might fit.

That sequence creates waste because it reverses the logic of capital allocation. In most areas of business, leaders begin by identifying where value sits. Revenue leakage. Cost concentration. Process inefficiency. Customer friction. AI often escapes that discipline because the technology itself feels like the opportunity.

Kate’s point was sharper than that. Leaders need to define success before deployment, not after. That changes the conversation from experimentation to accountability. If a company cannot explain what success looks like, how it will be measured, and what cost threshold makes the project viable, then what looks like innovation may simply be expensive exploration.

That is harder than it sounds because AI creates a cultural pressure to move. No executive wants to appear slow. No board wants to feel behind. No team wants to be the one ignoring a technology that may reshape the market.

That pressure creates urgency.

Urgency often weakens discipline.

And that may be where the largest costs are currently being created.

Every major technology cycle begins with excess. The early internet produced thousands of companies that disappeared. Cloud expanded software budgets faster than most finance teams could track. Mobile created an app economy where most products never found a durable business model. AI may simply be moving through the same pattern at greater speed.

Kate made an observation that stayed with me. She compared this phase of AI to the early years of YouTube, when much of the activity looked chaotic, experimental and difficult to measure. That framing is useful because it removes some of the pressure to treat every current experiment as a signal of long-term value.

Most of what is being built today will likely disappear.

That is not failure. That is usually how infrastructure cycles mature.

The companies that benefit most are rarely the ones moving fastest at the beginning. More often, they are the ones that build enough understanding to know where speed matters and where restraint matters more.

AI may be following the same path.

The cost of intelligence is falling quickly.

The cost of misusing it may still be underestimated.

And that gap may become one of the defining management questions of this cycle.

EVOLVE is a private community of senior commerce leaders, founders, operators, investors, advisors, and board members across nearly 50 markets.

We bring together VCs, PEs, family offices, board-level leaders, ambitious founders, and senior operators across commerce markets through ongoing structured conversations.

The community exists to accelerate access to people, perspectives, and opportunities that would be difficult to create alone.

Every week, members engage with experienced practitioners building, investing in, and scaling companies across commerce, technology, retail, marketplaces, payments, and adjacent industries.

The room stays small by design.

The conversations stay practical for the same reason.

That is EVOLVE.

If you’d like to learn more how to partner with, please email us at
[email protected] or [email protected]