Almost everything written about AI for a general audience stands at one of two extremes. At one pole, technological optimism: Mollick (Co-Intelligence, 2024) invites you to “always seat AI at the table”; Hoffman and Beato (Superagency, 2025) read it as an amplifier of human agency. At the other, the critique of hype: Bender and Hanna (The AI Con, 2025) dismantle Big Tech’s rhetoric; Narayanan and Kapoor (AI Snake Oil, 2024) teach how to tell what AI can do from what it cannot. It is a noisy debate, and it is worth situating, because my thesis need not take sides in it.
Why the bridge is indifferent to that fight
The reason is simple: in both cases, value is decided by the sociotechnical reading, not by the model’s power. If AI is as powerful as the optimism says, then governing it is more urgent —a tool that amplifies everything also amplifies error, bias and disorder; it is the multiplier with a sign—. And if it is as much smoke as the anti-hype says, then the diagnosis that separates where AI pays from where it only destroys value is more necessary, so you do not buy the promise. Whichever of the two worlds is true, the work is the same: read the organization before touching the technology. The bridge wins in both scenarios because it does not bet on the power of AI; it bets on the quality of the human reading that steers it.
The position that best rhymes with my work
Between those two poles there is a third, and it is the one closest to what I hold: reading AI as normal technology. It is the thesis of Narayanan and Kapoor’s essay “AI as Normal Technology” (2025): AI diffuses and is adopted the way electricity or software were —gradually, mediated by institutions, practices and organizations—, not as a singularity falling from the sky and changing everything at once. “Normal” does not mean “unimportant”: it means its impact plays out in the slow layer, that of how organizations absorb it, regulate it and fit it into their real work.
That rhymes with two ideas I have been holding. One is adoption without teleology: adopting AI is not an end in itself nor an inevitable destiny; it is a decision justified case by case. The other is the descriptive, not normative, use of diffusion theories: they describe how something spreads, they do not say that spreading it fast is good. Normality is not lukewarmness. It is, precisely, the condition for institutional work —understanding, governing, translating— to be what decides the outcome, rather than euphoria or panic.
What the reader takes away
If you are on the enthusiasm side, keep this: the tool’s power does not exempt you from the boring work of reading your organization; it makes it more urgent. If you are on the skeptic side, keep this: that much of the hype is smoke does not mean there is no real value; it means you have to know where to look for it. And if you are on no side —which is where most people running a company or agency are— the good news is you do not need to resolve the debate to start. It is enough to treat AI as what it is: one more technology, powerful and limited, whose value depends on how well you fit it to the human fabric you already have.
The framework is developed in The sociotechnical bridge and in My approach. Related essays: You cannot delegate to an agent what you cannot explain · Digital transformation is not a technical problem.