Just Hype and Noise, or Real ROI?

Here is the honest problem, before any advice. The conversation about AI is so loud, and so contradictory, that a reasonable business owner has almost no way to tell what’s real. Every tool is the one that changes everything. Every week brings a breakthrough that makes last week’s breakthrough obsolete. Every piece of software you already own has quietly grown an “AI” badge. And every confident prediction about what AI will do next year is repeated until it sounds like something that already happened. It is genuinely hard to think straight in that much noise, and being confused by it is not a failing on your part. The noise is doing exactly what it’s designed to do.

So this isn’t a piece that tells you AI is overhyped, and it isn’t a piece that tells you AI will transform your business. Both of those are just more noise pointed in opposite directions. This is a piece that hands you a small, practical test for telling the hype from the things that actually return money, so you can stop paying for theatre and start buying outcomes. Let me start with where the noise comes from, because once you see the machine making it, you can’t unsee it.

The noise is manufactured, not accidental

It would be comforting to think the hype is just collective excitement spilling over. Some of it is. But a lot of it is manufactured, and it helps to follow the incentives honestly. Enormous amounts of money have been bet on AI, and money that size needs a loud story to justify itself. Software companies have learned that attaching “AI” to a product moves it, whether or not anything underneath actually changed. Demos are engineered to dazzle in the ninety seconds you watch them. And the analyst predictions that get passed around as if they were facts are, underneath, forecasts with someone’s interest riding on them.

The clearest single piece of evidence I can point to is from the research firm Gartner, which is not exactly a band of AI skeptics. They coined a term for one of the loudest trends of the moment, “agent washing,” to describe vendors rebranding ordinary chatbots, scripts, and automation as autonomous “AI agents.” By their count, of the thousands of companies claiming to sell agentic AI, only a tiny fraction are doing anything that genuinely qualifies. They went further and predicted that more than forty percent of agentic-AI projects will be cancelled within a couple of years, on escalating costs, unclear value, and weak controls. When the people whose job is to track the industry are warning that most of a category is a costume, that tells you how much of what reaches you is noise.

The four tells of hype

Once you’re watching for it, manufactured hype has a handful of recognizable tells. None of them require you to be technical. They just require you to be a little unimpressed.

The demo that never meets production. A demo works because nothing real is connected to it and nothing is trying to break it. It runs on ten tidy records, never sees a malformed input, and is never asked what happens when two things go wrong at once. The gap between that demo and a system your business can actually run on is exactly where most projects quietly die. When something looks magical in a sales meeting, the only useful question is what happens on the ten-thousandth real record, on a bad day, with messy data. Hype lives in the demo. ROI lives in production, which no demo ever visits.

The “AI-powered” sticker. A label is not a capability. Plenty of products wear the badge with nothing behind it but a marketing decision. The deflating, clarifying question is simply: what does this actually do now that it didn’t do before, and can you show me that specific thing? If the answer is a fog of adjectives, the AI is on the box, not in the product.

Agent-washing. This is the sticker’s more ambitious cousin. The word “agent” implies something that takes real action on its own, across real systems. A great deal of what’s sold under that word is a chatbot with a confident tone, or a script that was doing its job before anyone called it intelligent. Ask what it actually does without a human driving each step, and what it’s allowed to touch. The honest answers are usually much smaller than the pitch.

Prediction sold as fact. “AI will replace X by next year” is a forecast, and a forecast with a seller’s interest behind it is a sales pitch wearing a lab coat. You can’t run your business on what a vendor says the technology will do soon. You can only act on what it demonstrably does today, for your specific work. Treat the future tense as a yellow flag.

What real ROI actually looks like

If the hype is loud and vague, the signal is quiet and specific. Real return from AI almost always looks a little boring, and that’s the point. It’s a named task that got measurably faster, cheaper, or more accurate, with an actual number attached, and a human still accountable for the result. Not “AI transformed our operations.” More like “the thing that used to take a person two hours now takes fifteen minutes and a quick review, and here’s the before and after.” Specific. Measured. Checkable.

And here’s the part the noise works hardest to hide: when real value shows up, it almost never came from the model itself. It came from the unglamorous work around the model. The studies bear this out with uncomfortable clarity. MIT found that roughly ninety-five percent of enterprise AI pilots produced no measurable impact on the bottom line, and the cause wasn’t weak models, it was, in their words, flawed integration. McKinsey, looking at the companies that do get value, found that the single biggest factor was the redesign of the actual workflow so the AI sits inside the process rather than beside it. Neither of those is the shiny part. Both of them are engineering and operational discipline. The model was never the hard or valuable part, which is exactly why hype, which only ever sells you the model, so reliably fails to deliver the return. If you want the longer version of that argument, it’s the whole of why your subscription and real AI engineering aren’t the same thing.

A plain test you can run on any AI pitch

So here’s the tool to take away. The next time anyone, a vendor, a consultant, an enthusiastic employee, even me, pitches you on an AI something, run it through five plain questions. You don’t need to understand the technology to ask them, and the answers sort signal from noise quickly.

First, what exact task does this change, and what is the number? A real answer names one specific job and a measurable before-and-after. A hype answer describes a vibe. Second, what happens when it’s wrong? Every AI system is sometimes confidently wrong, so a credible pitch has a designed answer for that, and a vague one pretends it won’t happen. Third, what does it actually touch, my real data and my real systems, and how? This is where the demo-versus-production gap shows itself, and it’s also where the safety and privacy questions live, which I’ve written about separately. Fourth, who is accountable when it breaks? “The AI made a mistake” is not an answer a business can run on; a name and a process is. And fifth, the one that cuts through everything: would you tell me if AI were the wrong tool for this? Anyone genuinely on your side has a category for “don’t build it.” Anyone selling noise does not. If a pitch can’t answer those five plainly, you haven’t found ROI yet. You’ve found noise with a good microphone.

Where the honest answer is “yes,” and where it’s “not yet”

I want to be careful here, because the lazy move at this point would be to wave the whole thing off as a bubble, and that’s just cynicism, which is its own kind of noise. AI is genuinely, substantially useful in specific places right now, and dismissing all of it because most of the marketing is hot air would cost you real opportunities. Grounded systems that answer from your own data, automation of well-defined document and support work, search that actually understands a question, these are real, and they pay off when they’re built properly around a real workflow.

And in the same honest breath: plenty of the time the right answer is to wait, or to not use AI at all. Some jobs need a simple piece of software and a clear form, not a model, and a forty-dollar-a-month tool that’s right beats a twenty-thousand-dollar AI build that’s wrong every single time. The willingness to say that out loud is, I think, the cleanest possible signal that you’re being given the truth rather than being sold. Refusing to hype the technology is itself the tell that someone isn’t running the hype machine.

The antidote to noise is boring honesty

Strip all of it down and the test is almost embarrassingly simple. Ignore the adjectives and count the outcomes. Distrust the demo and ask about production. Treat “AI-powered” and “agent” and “by next year” as questions, not answers. And trust the person who will tell you when the answer is no.

That last line is also, honestly, how we work, so I’ll say it plainly once and leave it there. We judge AI by whether it pays, we put numbers where other people put adjectives, and “this is the wrong tool here” is an answer we’re glad to give, because it’s the one that earns the next ten. If you want AI looked at that way, by people more interested in your outcome than in the trend, that’s what an AI engineering studio is for, and it’s exactly what our Blueprint™ is built to deliver: a straight read on what’s worth building, what’s worth buying, and what’s just noise. In a conversation this loud, boring honesty is the rarest and most valuable signal there is.