Product consultant Antonia Landi has spent the past twelve months talking with leaders across industries, and she has not heard a single organization explain how it determined that a specific AI tool solved a specific, validated problem. That absence, she argues, is the early stage of tool-stack debt compounding across the product function.

Landi has been giving talks about irresponsible tooling since 2023, when the culprit was undisciplined SaaS adoption: teams accumulating subscriptions that solved slightly different problems, never integrated, quietly compounding into a stack nobody could map or maintain. Organizations eventually pared that back through slow, unglamorous work — fewer tools, better connections, more deliberate choices. Licensing costs had exploded, onboarding had become unmanageable, and the hidden tax of maintaining dozens of integrations only became urgent once something broke.

"Then AI arrived, and apparently we forgot everything we learned," she writes.

A bad kind of déjà vu

Organizations are buying AI tool licenses at a pace Landi hasn't seen in years. The enthusiasm is partly justified — some tools are genuinely transformative, and the competitive pressure to "be doing AI" is not imaginary. What's missing is everything around the purchase: how the tool integrates with the existing stack, who owns and maintains it, what success looks like, and how teams will know when it isn't working.

What she hears instead is: "We've rolled out [tool] across the product team." That, she says, is not a tooling strategy.

The debt nobody's tracking

The failure pattern is familiar. One team adopts an AI writing assistant because a PM loved it and expensed it. Another picks a competitor for its UX. Someone in research buys a third for a synthesis feature the others lack. None talk to each other. Nobody owns the admin. The licenses renew automatically. Multiply that by every function currently "experimenting with AI" and you get a fragmented, ungoverned layer on top of your operating system that nobody designed and nobody fully understands.

The real cost isn't licensing — it's complexity. Cognitive load from switching between tools never designed to work together. Data living in five places instead of one. Institutional knowledge trapped inside a tool half the team can't access. The three months you'll eventually spend untangling which tool is the source of truth for what.

The biggest cost is opportunity cost. Fragmented operating systems can't move at the speed they're capable of, and AI adoption can make you slower, not faster — precisely when fast, coherent movers are the ones winning.

AI is a solution hypothesis

Landi's core framing: any new tool, AI included, is one of many possible solutions to a given problem — a hypothesis, not a foregone conclusion. That means starting with the problem. What capability does the organization lack? Where is the measurable bottleneck? What would change concretely if this worked? Those questions are the difference between a tool that transforms how a team works and one that fades into monthly OPEX after three enthusiastic weeks.

The conditions for the framework to hold: you must also do the unglamorous post-purchase work — assigning an owner accountable for adoption, maintenance, and the six-month "is this still right?" review; integrating with how work actually gets done today, not in theory; and deciding what gets retired. The most disciplined organizations she works with treat every addition as an impulse to consolidate something else.

None of this is anti-AI. Landi uses and recommends AI tools. But remarkable capability and organizational fit are two different things, and conflating them is how organizations end up with bills and nothing to show.

Her five questions for senior product leaders before the next renewal: Have you validated the problem, not assumed it? Do you have measurable success criteria? Does the tool integrate with how work gets done today? Who owns it? What are you consolidating?

The organizations that win the AI race, Landi argues, won't be those adopting the most AI tools, but those integrating AI into a coherent operating system where insights flow between tools and humans spend their time on judgment rather than tool management. Leaders still have a window before the debt matures to insist on validated problems before purchased solutions — and to build systems that absorb new capability rather than be destabilized by it.