GenAI arrived as pure technology push. Nobody in any company asked for it. Four years on, organisations believe they have converted it into demand pull, that they now know what the technology does and are aiming it at their own needs. Most of them are wrong, and the way they are wrong is quite expensive.
Start with the distinction:
- Technology push is when a scientific advance arrives first and goes looking for a problem.
- Demand pull is when a felt need comes first and pulls a solution into being.
Jacob Schmookler made the case for the demand side in 1966, arguing from patent data that market need largely determines where inventive effort actually goes.
GenAI is the cleanest case of push in a generation. It did not emerge because corporations articulated a need. It emerged because a research breakthrough made it possible, then landed on every desk at once, looking for something to do.
Technological Paradigms (Giovanni Dosi, 1982)
Then something happens that innovation researchers have described well. Giovanni Dosi called it a technological paradigm, in a 1982 paper that is still one of the core concepts in the field. A breakthrough establishes a paradigm, and everything after it moves along a trajectory inside that paradigm.
What follows feels like demand, because the problems now look obvious and the solutions look available. Kline and Rosenberg had already replaced the old linear picture of innovation by 1986 with a model built on feedback loops running in both directions, where innovation is a push and pull process at the same time rather than one or the other.
People cannot articulate a need for something they cannot picture. Once they can picture it, the demand appears, and it feels like it was always there.
So the conversion you think is happening is real. It is just not finished, and it is not what you think it is.
Two Layers Moving at Different Speeds
Two fundamental things get missed:
The first is that the frontier is still pure push. Nobody in your company asked for reasoning models or autonomous agents. The labs built them and shipped them, and every time they do, the application layer inside your organisation gets reset. Your roadmap is not being invalidated by bad planning. It is being invalidated by a push layer that has not slowed down.
The second is that the evidence says most companies have not converted at all. If they had, you would see targeted applications embedded in real workflows returning measurable numbers. Instead you see enormous adoption and almost no transformation. A widely reported MIT industry study last year put the share of pilots with no measurable bottom-line impact at around 95 percent.
High adoption with low transformation is the signature of still being in push mode.
Push Wearing Pull's Clothing
Here is the part that is not very popular. "We understand what GenAI can do, so we are pulling it toward our needs" is not demand pull. It is push wearing pull's clothing.
Real demand pull starts from a pain that exists whether or not the solution does. If your need was generated by the technology being available, the technology is still driving and you have simply stopped noticing.
Dosi's uncomfortable implication is that a paradigm focuses attention in particular directions and blinds you to the alternatives it excludes. That is powerful and it is also a cage. You will find every problem your tool can address and stay blind to the ones it cannot.
The Uncomfortable Counterfactual Test
The test is one single question, and it is unpleasant to run in a room full of people who have already committed budget:
"Would this pain be on our list if GenAI did not exist?"
If the honest answer is no, you are still being pushed. If the pain would not have existed without the tool, you have not built a use case. You have built a justification.
Where Genuine Demand Signal Lives (Eric von Hippel)
Where genuine demand signal does live is a question Eric von Hippel spent a career on. His lead user work (1986, 1988) argued that the richest source of innovation is often not your own R&D, but the users who hit a need months or years before the market does and have already improvised a crude workaround.
Those workarounds are the demand signal, and they existed before the technology arrived.
The Executive Mandate
Run both directions honestly:
- Push is not finished: Stay close to the frontier and keep a few bets aimed at capability nobody has asked for.
- Pull is not automatic: Go and find the pains that were already costing you money before the LLM boom, and check them against the counterfactual before funding anything.
Technology Push has not been lost. It has learned to disguise itself as Demand Pull, which is harder to spot and more expensive to get wrong.
Ps: Try the counterfactual question in your next steering committee and watch the temperature drop. If the pain would not have existed without the tool, you have not built a use case. You have built a justification.
- Schmookler, J. (1966), Invention and Economic Growth, Harvard University Press.
- Dosi, G. (1982), "Technological paradigms and technological trajectories," Research Policy 11(3), 147-162.
- Kline, S.J. and Rosenberg, N. (1986), "An overview of innovation," in The Positive Sum Strategy, National Academy Press, 275-305.
- Von Hippel, E. (1986), "Lead Users: A Source of Novel Product Concepts," Management Science 32(7), 791-805; (1988), The Sources of Innovation, Oxford University Press.
Distinguish real demand from technology hype before committing budget.
brainTerms.ai uses multi-agent cross-examination to audit problem-solution fit, validate market demand signals, and enforce institutional rigor on every AI initiative.