Cutting Through the Noise: Where We Actually Stand With AI

Everyone has an opinion about AI right now. Few of them are disinterested. This is the first post in what I intend to be a recurring exercise: stepping back from the daily churn of announcements to ask a simpler, harder question — what is actually true, and what is just loud?

The noise problem

We are living through one of the loudest technology cycles in living memory. AI is discussed everywhere — in boardrooms, in dinner conversations, in op-eds written by people who could not define a transformer if their job depended on it. That is not a criticism of curiosity; curiosity is healthy. The problem is that volume has detached from understanding. The people talking most confidently are often the ones with the least to lose from being wrong.

When a technology becomes a cultural phenomenon, the signal-to-noise ratio collapses. Genuine progress gets buried under recycled takes, doom forecasts, and utopian promises that all sound equally authoritative. For someone trying to make a real decision — whether to adopt a tool, restructure a team, or just figure out how worried to be — the noise is not harmless. It is exhausting, and it is expensive.

Everyone selling you AI has an agenda

Here is the part that does not get said often enough: most of the loudest voices in AI are not neutral observers. They are participants with something to sell.

The CEOs of the largest AI labs are not lying to you for fun. They are operating under enormous structural pressure. Many of these companies are burning cash at a staggering rate, sitting on valuations that only make sense if the most optimistic scenarios come true, and looking toward future funding rounds or IPOs that depend on sustaining a story. In that context, “AI will change everything, and soon” is not just a belief — it is a business requirement. The narrative is the product, at least until the revenue catches up.

This cuts in two directions. The same incentives produce breathless predictions of imminent superintelligence and carefully staged warnings about existential risk. Both grab headlines. Both position the speaker as someone who sees further than you do. And both conveniently imply that you should pay close attention to — and ideally pay for — whatever that person is building. When the optimist and the doomsayer are the same small group of people, you should ask why fear and hype so often arrive in the same press release. Critics have a name for the pattern: regulatory capture — the industry framing the risks and the rules at the same time — and they warn to “be very wary of AI technology leaders that throw up their hands and say, ‘regulate me, regulate me.’” The loud focus on speculative future doom also conveniently distracts from the unglamorous harms and limitations that already exist today.

I am not claiming everyone is acting in bad faith. Plenty of the enthusiasm is sincere. But sincerity and accuracy are different things, and an honest person with a billion-dollar incentive to believe something will, reliably, come to believe it.

ROI is not guaranteed — on either side

Strip away the storytelling and you are left with an unglamorous question that applies to any technology: does the money come back? Right now, the honest answer for AI is “not necessarily.” And that uncertainty exists on both sides of the transaction.

For the AI companies themselves, the path to durable profitability is far from settled. The numbers are sobering: OpenAI reportedly booked roughly $13 billion in revenue in 2025 against about $34 billion in costs, for a net loss north of $38 billion — and it is preparing to go public anyway. Training and serving frontier models is extraordinarily capital-intensive: the largest hyperscalers are on track to spend hundreds of billions of dollars on AI infrastructure, a build-out many compare to the classic hallmarks of a bubble. Even Nobel laureate economist Daron Acemoglu has cautioned that “these models are being hyped up, and we’re investing more than we should.” Pricing is under constant downward pressure as capable open and competing models proliferate. The moat that justifies the valuations is real in some places and assumed in many others. A great product and a sustainable business are not the same thing — plenty of companies have had the former without the latter.

For the companies trying to adopt AI, the picture is just as murky. The proof-of-concept that dazzles in a demo often stalls on contact with real data, real workflows, and real compliance constraints. The most-cited data point here is blunt: an MIT study, “The GenAI Divide: State of AI in Business 2025,” found that roughly 95% of enterprise generative-AI pilots delivered no measurable impact on the P&L. Gartner, separately, predicted that at least 30% of generative-AI projects would be abandoned after proof of concept by the end of 2025, and that over 40% of “agentic AI” projects would be cancelled by 2027. The cost is not just the API bill — it is integration, change management, oversight, and the quiet tax of maintaining systems that, as AI researcher Gary Marcus keeps pointing out, remain fundamentally unreliable and behave probabilistically.

But that 95% figure deserves exactly the scrutiny this blog tries to apply to everything else — including the claims that flatter our own skepticism. It travelled far faster than its caveats. The headline number leaned on a small set of interviews the authors themselves described as “directionally accurate” rather than audited reporting; it defined “success” narrowly as measurable P&L impact within roughly six months — a short window for the messy work of organizational change — and it largely set aside softer gains like efficiency, cycle time, or quality. Awkwardly, the report’s proposed remedy pointed toward NANDA, an MIT-affiliated agentic-AI initiative, which gives the authors a stake in the very market they are grading. Critics such as the Marketing AI Institute’s Paul Roetzer argue the study went viral precisely because it confirmed an existing mood, not because the evidence was airtight. The more careful reading is not “AI doesn’t work” but “most pilots, measured narrowly and early, don’t yet show up on the P&L” — a far less dramatic claim, and a more useful one. More balanced surveys, like UC Berkeley’s California Management Review, land in the same place: real value exists, but it is uneven, hard-won, and easy to overstate in either direction.

The measurement problem is real even where adoption is genuinely enthusiastic. By 2026, Uber had gone all-in on AI-assisted coding — roughly 95% of its engineers use AI tools every month, and about 10% of committed code is now written by autonomous agents — and the company burned through its entire annual AI budget in four months. Yet Uber’s president and COO, Andrew Macdonald, was candid that all that activity has not yet translated into anything he can put a number on: “it’s very hard to draw a line between one of those stats and ‘Okay now we’re actually producing 25% more useful consumer features.’” When one of the most aggressive corporate adopters on the planet still cannot connect heavy AI usage to shipped customer value, “the ROI is obvious” starts to sound less like accounting and more like faith.

None of this means AI is a bubble with nothing inside it. The technology is genuinely useful, and in specific, well-scoped applications it already pays for itself handsomely. The point is narrower and more important: useful and guaranteed return are not synonyms, and anyone telling you the ROI is automatic is selling, not advising.

Why the hype has a human cost

There is a real toll to all of this, and it is measurable. The constant drumbeat — your job is about to vanish, your company will be left behind, the machines are about to wake up — leaves people anxious about a future they feel powerless to influence. Surveys bear it out: a Mercer study of 12,000 workers found that 40% now fear losing their job to AI, up sharply from two years earlier, and other polls put the share of Americans who fear AI-driven job loss at roughly seven in ten. Some of that anxiety is justified; some of it is manufactured, deliberately or not, because fear and FOMO are excellent motivators for clicks and purchases.

Clarity is the antidote to manufactured anxiety. You cannot make good decisions — or sleep well — while marinating in someone else’s narrative. The goal is not to dismiss AI, nor to fear it, but to see it accurately enough that it stops being a source of dread and becomes what it actually is: a powerful, limited, fast-moving set of tools.

The philosophy: don’t ride every wave

This is the worldview I want this blog to stand for. I am not interested in riding each overhyped wave as it crests. Every few weeks there is a new “this changes everything” moment, and most of them are forgotten by the next one. The people who chase every wave end up exhausted, broke, and no closer to understanding the water.

The alternative is slower and far more valuable: cut the noise, ignore the theater, and do the work of understanding where we genuinely stand and where we are plausibly heading. That means separating the speaker from the claim, asking who benefits from a given narrative, looking for evidence instead of vibes, and being comfortable saying “we don’t know yet” when that is the truth.

Lazy in the smart sense: refuse to spend energy on hype, so you can spend it on what actually moves the needle.

What to expect here

• Grounded takes on what AI can and cannot do today — not in two years, today.
• Honest accounting of cost and ROI, for builders and adopters alike.
• A consistent habit of asking “who benefits from me believing this?”
• Practical strategy for adopting AI without betting the company on a narrative.

Conclusion

The most contrarian thing you can do in this moment is to be calm and patient. Not dismissive, not credulous — calm. AI is real and it matters, which is exactly why it deserves clear eyes rather than hot takes. The hype will keep coming. The losses, the IPO narratives, and the confident strangers will keep coming too. The job — mine here, and yours as a reader — is to keep cutting through all of it and ask the only question that ever pays off: what is actually true?

That is where we start. Welcome to LazySmart.

Sources

1. MIT NANDA, “The GenAI Divide: State of AI in Business 2025” — 95% of enterprise GenAI pilots show no measurable P&L impact (via Fortune).
2. Marketing AI Institute (Paul Roetzer), “That Viral MIT Study Claiming 95% of AI Pilots Fail? Don’t Believe the Hype.” — critique of the MIT report’s methodology, narrow success definition, and conflicts of interest.
3. Fortune, “Uber burned through its entire 2026 AI budget in four months. Now its COO is questioning whether it’s worth it.” — heavy AI-coding adoption with no clear line to consumer value.
4. Gartner, “30% of Generative AI Projects Will Be Abandoned After Proof of Concept by End of 2025.”
5. Gartner, “Over 40% of Agentic AI Projects Will Be Canceled by End of 2027.”
6. OpenAI’s 2025 financials — ~$13B revenue against ~$34B in costs and a net loss above $38B (via Yahoo Finance).
7. NPR, “Here’s why concerns about an AI bubble are bigger than ever” — hyperscaler capex and the investment-vs-returns mismatch.
8. Gary Marcus, “AI’s Reliability Crisis” — Project Syndicate.
9. TechCrunch, on regulatory capture and the existential-risk narrative.
10. Mercer Global Talent Trends / Resume Now — worker AI anxiety and public fear of AI-driven job loss.
11. UC Berkeley, California Management Review, “Cutting Through the AI Hype: The Facts Leaders Need to Know About GenAI Adoption and ROI.”