Should you work in product or tech, you’ve most likely sat by means of this assembly. Somebody forwards a vendor demo or a viral publish, and abruptly the roadmap wants an agent, an MCP app, or a harness. We see the identical scene in shopper conferences on a regular basis, often beginning with “we want an agent for this.” Typically, after decomposing the request from first ideas, it seems the shopper wants one thing utterly totally different: a customized predictive mannequin, a greater use of the LLMs they have already got, or no AI in any respect. Somebody has to say “not so quick”, with out sounding defensive.
That’s not straightforward, as a result of the AI dialog is loud and convincing. This text offers you three strikes to remain grounded:
-
Perceive how the AI worth chain works, and the place you realistically sit in it.
-
Construct a basis of data that permits you to construction and reuse what you find out about AI.
-
Devour data deliberately, retaining your personal perspective.
Collectively, these practices make you a stronger sparring associate when the subsequent hype wave hits your group. On the finish, you’ll discover six widespread vendor claims and the query that deflates every one.
1. Perceive your home within the AI worth chain
Earlier than you may decide an AI declare, you could know who’s making it and why. This part maps the AI ecosystem, reveals the place you seemingly sit in it, and explains how that place may work in opposition to you.
Who sells what to whom?
The AI worth chain may be modeled in 5 major layers, from the chips on the high to the businesses that put AI to work on the backside:
On the high, returns are near banked: the silicon is offered and paid for on supply. As you progress down, worth will get much less sure. The cloud suppliers and mannequin labs are betting on future demand, so their actual returns are tougher to pin down. Plenty of that “demand” is definitely the identical cash circulating contained in the ecosystem: chipmakers fund the labs, labs decide to the clouds, and the clouds purchase chips. Actual end-user demand is barely determined on the finish of the chain.
The chances are stacked in opposition to AI shoppers
In case you are studying my work, chances are high you sit at that receiving finish, as an enterprise AI adopter or utility developer consuming fashions, instruments, and platforms from the layers above. You might be a part of the end-user demand, and the entire ecosystem is working exhausting to maximise it. That places you in a weak place:
-
Everyone seems to be promoting at you. AI firms have perfected the artwork of selling. Their message is that AI is reasonable, straightforward, works out of the field, and can remodel your life and enterprise. I wish to name this the “accessibility phantasm.” The extra unsure the precise product, the heavier the advertising behind it.
-
You understand much less. Distributors know the bounds of their merchandise, however you usually uncover them solely when you find yourself already fighting the final mile: the damaging stretch between a demo and a system that delivers worth to actual customers.
-
Your payoff comes later. Distributors receives a commission while you purchase; you receives a commission solely when the system works and you may show it. Getting from uncooked substances like fashions, APIs, and agent frameworks to measurable worth takes a mature mixture of conviction, technical talent, and enterprise data.
The underside line: AI’s final mile continues to be largely undone. In McKinsey’s 2025 survey, greater than 80% of firms utilizing generative AI stated they’d not but seen a transparent affect on their total income. RAND studies a failure price above 80% for AI tasks, though this covers extra than simply tasks that by no means attain manufacturing. And even a deployed system doesn’t assure worth. Many firms don’t have any dependable technique to measure whether or not AI really improves enterprise outcomes, so the loop stays open (cf. Dataiku’s 2026 CIO survey).
Creating common sense about AI and studying to use it in your organization’s context is your major protection in opposition to falling for the hype.
2. Construct a basis of data
Common sense comes from figuring out the fundamentals nicely sufficient to see by means of the noise.
Peeling off the emotional layer
Most AI content material mixes details with feelings that had been added on goal: pleasure, urgency, worry of lacking out. Emotion works even on skilled folks as a result of it exploits three blind spots:
-
Demos over workflows. A demo reveals the very best of 1 run. Manufacturing means the identical job a thousand occasions, edge circumstances included.
-
Benchmarks over your knowledge. Benchmarks measure fashions on clear, usually public datasets, not in your messy inner ones.
-
Survivorship in case research. Vendor case research function the tasks that labored, not those that had been quietly shelved.
To peel off this subjectivity, you could perceive the fundamentals of how AI works, and particularly its limitations and dangers. With out that basis, you’re mentally constructing a home of playing cards. Every card is a headline, a demo, or a vendor declare, propped up by the others. The construction can develop impressively tall, however one sharp query can convey it down.
Structuring your AI data
A stable basis grows extra slowly, however the whole lot you study later has a delegated place to relaxation. In my opinion, it has two important elements:
-
The maths, first-hand or second-hand. AI is rooted in linear algebra, likelihood principle, and calculus. That’s the way you get to know its intrinsic limitations — like the truth that at this time’s language fashions frequently fail by design as a result of they estimate possibilities. Studying the mathematics takes years; if that isn’t real looking, borrow it from just a few specialists whom you belief.
-
Programs pondering. To uncover the worth of AI in your particular enterprise context, you could perceive how they join and work together. Good beginning factors are Donella Meadows’ Pondering in Programs or Shane Parrish’s The Nice Psychological Fashions; for AI, our AI Technique Playbook maps among the psychological fashions that we use throughout our work with purchasers.
With this basis, claims begin to sound totally different. On a home of playing cards, a vendor promising “zero hallucinations” sounds nice. On a stable basis, it seems like a query for the subsequent name: zero, measured how, and on which knowledge?
3. Devour data deliberately
How do you discover credible sources and actual perception in an awesome sea of AI content material? On this part, I share the psychological habits that assist me acknowledge content material that can really train me one thing new.
Perceive who advantages
Behind most sources sits somebody who advantages while you comply with their name to motion, and their incentives are seemingly totally different from yours. To maintain your personal perspective, it helps to invert the everyday move of content material creation:
-
Info: a statistic, a benchmark, a survey consequence.
-
Story: examples, buyer quotes, and feelings wrapped across the details.
-
Motion: the step the story strikes you towards, like reserving a demo or shopping for a platform.
Learn backwards from the motion, and it turns into clear which details had been chosen and why. The Dataiku survey I cited above is an effective instance. The information is helpful, nevertheless it was commissioned by an organization that sells agent administration software program, and the story of CIOs shedding management nudges readers straight towards that product. That doesn’t make it improper: use the numbers, however low cost the conclusion.
Take note of language
How a chunk is written usually tells you greater than what it claims. Be careful for these pink flags:
-
AI slop. Saying one thing genuinely new about AI is tough. Individuals who make that mental effort are likely to put their ideas in their very own phrases, edit closely, and disclose once they used AI. Polished, generic language indicators recycled content material.
-
Anthropomorphic framing. Enterprise distributors more and more body AI merchandise as “digital staff” that be a part of your group. This invitations us to assume in headcount quite than outcomes, which makes ROI guarantees really feel intuitive earlier than they will really be measured.
-
Inflated feelings. Headlines a few looming job apocalypse or AI-induced threats to humanity are not often simply journalism; usually, there’s a advertising machine behind them. Concern cuts each methods: if a know-how is highly effective sufficient to finish the world, certainly additionally it is highly effective sufficient to rework your small business (see Lee Vinsel’s Notes on Criticism and Expertise Hype). If you break these claims all the way down to first ideas, they not often maintain up as acknowledged.
-
Heavy jargon. Jargon usually clothes up an current idea as one thing new. Gartner, for instance, describes “agent washing” because the rebranding of current merchandise, resembling AI assistants, robotic course of automation (RPA), and chatbots, with out substantial agentic capabilities. Many use circumstances positioned as agentic at this time don’t really require agentic implementations.
Additionally, take a look at how a product reaches you. When the worth is apparent, an organization can afford to let the product converse for itself. Cursor’s founders did no outbound gross sales till late 2025: the product was helpful from day one, and customers did the advertising. When the worth is unsure, the advertising will get louder as an alternative. Builder.ai promised to make software program creation “as straightforward as ordering pizza” and marketed its AI assistant Natasha as a breakthrough. In 2025, the corporate filed for chapter amid monetary scandals and accusations of AI washing (Wikipedia).
Placing it collectively: my filter for brand new AI ideas
As a lot as I really like exploring new AI improvements, operating two firms leaves me restricted time for experimentation. Right here is the filter I take advantage of each time a brand new AI concept or idea hits the headlines:
-
Triage. I ask two questions. Is it genuinely new, or a rebrand of one thing that already exists? And is it strategically related for our work?
-
If each solutions are sure, go deep. I learn the first sources and take a look at it out myself, ideally on actual knowledge.
-
If not, park it. I examine again as soon as third-party knowledge and opinions from specialists I belief turn out to be accessible.
-
Revisit with proof. If the info reveals promise, the thought goes again to step 2.
Internally, we additionally develop and use the AI Radar as a quantitative overview of the AI panorama. numbers and development curves is one other nice technique to make your choices extra goal.
Conclusion: Pushing again with out being “in opposition to AI”
You don’t have to win a heated argument about whether or not we’re in a bubble. Reasonably, you want a call logic that holds up both means. Subsequent time an AI declare reaches you, decode it first:
Use questions like these to interrupt an AI concept all the way down to what may be verified. Over time, you’ll study to uncover gaps and acknowledge these concepts which are possible and may ship true worth in your small business.
Which hype claims are you pushing again on proper now? Depart a remark, and I’ll decide them aside in a future article!
Observe: All photos are by the creator.
