The AI Question That Actually Matters

Let's get one thing out of the way. AI is real. It is useful. It is not going anywhere.

You can watch it draft, summarize, classify, and write code at a pace that would have looked like science fiction a few years ago. Teams move faster with it. Researchers cover more ground. Support tickets get triaged before a human ever opens them. Anyone telling you the technology is fake or a passing fad is selling their own kind of hype.

So this is not a "kill your AI budget" essay. The tools work. That part is settled.

But here is the harder truth: a technology can be genuinely useful and still be adopted in ways that are unhealthy. Those two things live comfortably side by side. And right now, a lot of companies are using AI in ways that look increasingly bubble-like — not because the models are bad, but because the behavior around them has come loose from reality.

Useful Tool, Unhealthy Habits

Watch how many organizations talk about AI and you notice something. The conversation is rarely about outcomes. It is about adoption. How many seats. How many tokens. How "AI-first" the roadmap looks in the next all-hands.

That is the tell. When the story shifts from "here is what this improved" to "here is how much of this we are doing," the connection to value has started to fray. The activity becomes the point. The spending becomes the proof. And nobody in the room can quite explain what actually got better for the customer.

This is not fraud. It is not even stupidity. It is drift — a series of reasonable-sounding decisions that quietly detach a company from the thing that made the technology worth buying in the first place.

What "Detached" Actually Means

Detached is a specific word, so let me make it plain.

Adoption is detached when it becomes disconnected from the conditions that actually make good outcomes possible. Not the enthusiasm. Not the press release. The real, unglamorous conditions underneath.

I find it useful to check that against four of them — Health, Direction, Operations, and Morale:

• Health: Is the business measurably better — revenue, quality, retention, cost — or just busier? Detached adoption confuses activity with health.

• Direction: Can leaders explain where this is taking the product and the customer? Or is "we're using AI" the entire strategy? Detached adoption confuses a tool with a plan.

• Operations: Does the workflow actually work end to end, including the human oversight AI still needs? Or did we automate a step and hope? Detached adoption confuses shipping AI with solving problems.

• Morale: Do the people who do the work believe in this, or are they quietly cleaning up after it? Detached adoption confuses a headcount cut with a productivity gain.

When those four connections hold, you have grounded adoption. When they snap and the spending keeps flowing anyway, you have detachment. The metrics still move. The invoices still clear. But the line from effort to outcome goes fuzzy — and a fuzzy line is where expensive delusions grow.

Two recent stories show the pattern from opposite directions.

Ford: Cutting the Judgment, Then Buying It Back

Ford leaned into AI and let go of experienced engineers, betting automated systems could carry the load. Then the quality problems arrived — the kind of issues that seasoned engineers catch on instinct, the edge cases and "that doesn't feel right" moments that come from years on the job.

So Ford reversed course and started rehiring hundreds of the very people it had pushed out. Its own VP of vehicle hardware engineering put it plainly: AI is a fantastic tool, but it is only as good as the information you train it on.

Read that through the four lenses and the detachment is obvious. Operations broke because human judgment was treated as a removable cost rather than part of the system that made quality possible. Health took the hit through defects. And the reversal itself — the walk-back, the rehiring, the lost time — cost far more than keeping the expertise ever would have.

That is not an anti-AI story. It is a story about cutting the wrong thing because the trend said you could.

Uber: Spending Hard, Struggling to Explain Why

Uber ran the other version of the same problem. The company burned through its entire 2026 AI budget in four months, partly by gamifying adoption with an internal leaderboard ranking teams by tool usage.

Then leadership looked up and asked the uncomfortable question. Uber's COO said it was very hard to draw a line between the rising AI spend and more useful features for customers. His words: "That link is not there yet." The company has since capped employee AI tool spending at $1,500 per month, per tool.

Look at the four lenses again. Usage became the proxy metric — a leaderboard measuring motion, not results. Direction blurred, because even the executives could not connect the spend to the outcome. This is detachment in its purest budget form: the money moves, the dashboards fill up, and the causal chain to anything a customer would notice goes missing.

The healthy move, to Uber's credit, was to stop and ask. The detached move was needing to spend that much before anyone did.

The Signs, So You Can Spot Them Early

Put the two together and the shape is clear. Usage, signaling, and spend are outrunning grounded value logic. You do not need a crash to diagnose that. You just need to watch for the symptoms:

• Layoffs justified by AI before replacement quality is proven

• AI budgets growing faster than any evidence of outcomes

• Token use or tool use treated as a success metric in itself

• Executives who cannot connect spend to customer value

• Human oversight quietly reintroduced after bold automation claims

• Workflow disruption reframed as "innovation"

• "AI-first" rhetoric standing in for an actual business case

Notice that none of these are technology failures. Every one is a behavior. That is the whole point. The models are fine. The judgment around them is what wanders off.

We Have Watched This Movie Before

Here is the part that should give everyone pause. A genuinely useful underlying technology can get culturally distorted by hype, speculation, bad incentives, and detached narratives — until the distortion becomes the thing people remember.

Blockchain is the cautionary tale. Strip away the noise and there were real infrastructure ideas in there: verifiable records, distributed trust, systems that did not need a single referee. Some of that had legitimate value.

But crypto swallowed the conversation. Speculation, scams, and pure spectacle piled up until the whole category became hard to discuss seriously. The useful substrate got buried under bad behavior. Blockchain did not fail because the tech was worthless. It got a black eye because the culture around it optimized for belief over value.

AI is a far more capable technology, which is exactly why the risk matters more, not less. The better the tool, the easier it is to perform faith in it instead of proving results — and the more expensive that performance gets before anyone notices.

The Line That Keeps You Honest

None of this requires you to declare a bubble. It just requires one honest question, asked out loud, on a regular basis.

The moment a company cannot explain how AI spend improves actual customer, employee, or operational health, it has started to drift.

That is the whole test. Not "are we using enough AI." Not "do we look modern." But "can we trace this to something real for the people we serve, the people who work here, or the way the business runs."

If you can draw that line, keep going — you are using a powerful tool well. If you cannot, no amount of usage, spend, or "AI-first" language will save you. You will just be funding a very expensive belief.

The most important AI question right now is not whether the models are impressive. They are. It is whether the organizations adopting them are staying attached to reality.