The AI Mirage  ·  Essay  ·  July 2026

Crossing the AI Chasm

Thirty years ago, Geoffrey Moore explained why brilliant technologies stall right before the mainstream. His map explains exactly where enterprise AI is stuck today — and the mirage is what's holding it at the edge.

By Raj Lal  ·  AI Edge for Leaders  ·  Strategy  ·  12 min read

Two people watch the same AI demo. The first leans forward, sees the future, and wants it now — flaws and all — because the possibility is intoxicating. The second folds their arms and asks a colder set of questions: Will it work every time? Who else in my industry runs it in production? What happens when it's wrong? These are not two moods. They are two different buyers, separated by the most famous gap in the history of technology strategy. In 1991, Geoffrey Moore named that gap the chasm, and argued that most promising technologies die in it. Enterprise AI is standing at its edge right now — and the reason it hasn't crossed is the subject of this entire issue.

01 — The MapWhat Moore actually saw

Moore's Crossing the Chasm took a familiar bell curve — the technology adoption life cycle, running from innovators to early adopters to the early and late majority and finally laggards — and made one piercing observation. The curve looks continuous, but it isn't. Between the early adopters and the early majority there is a discontinuity so severe it deserves its own name. The people on the left of that gap and the people on the right don't just adopt at different speeds; they buy for fundamentally different reasons, and a strategy tuned to one is nearly useless for the other.

On the left sit the visionaries. They buy a technology for its potential. They'll tolerate rough edges, incomplete products, and outright failures because they're chasing a strategic leap their competitors don't yet see. A slick demo of what could be is often all they need. On the right sit the pragmatists — the early majority who represent the real mainstream market. They buy for the opposite reason: not potential, but proven, reliable productivity. They want the whole product, references from people like them, and the confidence that it simply works. The chasm is the void between "wouldn't it be amazing if" and "show me it already works for someone like me." Countless technologies have delighted visionaries, run out of them, and quietly died before a single pragmatist signed on.

What made Moore's insight durable is that it kept coming true. The personal computer, enterprise software, the internet, cloud computing, mobile — each had a visionary honeymoon followed by a treacherous gap, and each was crossed not by the company with the flashiest technology but by the one that assembled a complete, trustworthy solution for a specific mainstream buyer. The graveyard beside the chasm is full of technically superior products that never built the whole product a pragmatist needed. That history is the reason to take the framework seriously now: it is not a metaphor loosely borrowed for AI, but a pattern that has governed every major platform shift of the last forty years, and there is no reason to believe AI is exempt.

02 — The DiagnosisAI is not early. It's at the edge.

The prevailing story of the last three years is that AI adoption has been a runaway success — and by the standards of the left side of the curve, it has. Innovators and visionaries embraced it with extraordinary speed. Pilots launched in every function. Boards demanded an AI strategy. But rapid adoption by early adopters is precisely what the chasm looks like from the wrong side. It is not evidence of crossing; it's the run-up to the jump.

Look again at the data this issue keeps returning to. MIT found that roughly 95% of enterprise generative-AI pilots deliver no measurable impact on the bottom line. S&P Global reported the share of companies abandoning most of their AI initiatives leapt to 42% in a single year. Read through Moore's lens, those numbers snap into focus: they are not the sound of a failing technology. They are the exact signature of a technology stuck at the chasm — plenty of visionary pilots, almost no pragmatist production. The pilots were bought on vision. Production is bought on proof, and the proof isn't there yet.

This reframes the entire "AI disappointment" narrative of 2026. The pundits declaring the technology overhyped are making the classic error Moore warned about — mistaking the chasm for the ceiling. The early-adopter market got saturated, the easy visionary enthusiasm got spent, and the pragmatist majority hasn't moved. That feels like a plateau. It's actually a gap, and gaps can be crossed — but only with a completely different strategy than the one that won the visionaries.

03 — The BarrierWhy the mirage is a market problem

Here is the connection that ties Moore's framework to everything else in this issue. Hallucination is not merely a technical defect. It is the specific reason enterprise AI cannot satisfy the pragmatist buyer — and therefore the specific reason it cannot cross the chasm.

Recall who the pragmatist is. She does not buy potential; she buys reliable, provable productivity. Now hand her a system that is fluent, impressive, and occasionally, confidently wrong in ways no one can predict. To the visionary, that unpredictability is a charming quirk of a powerful new tool. To the pragmatist, it is disqualifying — because her entire buying criterion is "can I depend on this?" and the honest answer is "usually." A technology that works 95% of the time is a visionary's delight and a pragmatist's nightmare, because the pragmatist has to answer for the other 5% to a customer, a regulator, or a board. The confidence trap we examined earlier isn't just a UX hazard; it's the thing that makes AI fail the pragmatist's one non-negotiable test.

This is why the demos that dazzled the early market now land with a thud in front of the mainstream one. The pragmatist has seen the demo. She isn't asking whether the technology can do the task — she assumes it can. She's asking the questions a demo can never answer: How often does it fail? Who else in my industry trusts it in production? What's the whole product around it, and who supports me when it breaks? Every headline failure in this issue's Casebook — the airline bound by its bot's invention, the dealership's dollar car — is a data point the pragmatist has filed away as a reason to wait. The mirage, in other words, is doing its damage at the level of the market, one withheld pragmatist deal at a time.

04 — The BridgeThe whole product for AI

Moore's answer to the chasm was not "build a better core technology." It was the concept of the whole product: the generic product — the thing you actually ship — is only the center of what a pragmatist buys. Around it must sit everything required to deliver on the full promise: integration, support, standards, references, the complete solution to the customer's real problem. Visionaries will assemble the whole product themselves out of enthusiasm. Pragmatists refuse; they expect it delivered. The work of crossing the chasm is the work of completing the whole product for a specific market.

For AI, the generic product is the model. And the whole product — the part that turns a clever model into something a pragmatist will actually deploy — is precisely the trust infrastructure this issue has been describing all along.

Grounding gives the pragmatist answers anchored to sources she can audit. Evals give her a measured, defensible failure rate instead of a vibe. Guardrails and human oversight let her promise safe behavior under stress. Integration, security, support, and — above all — references from peers who already run it in production complete the circle. Notice that none of this is about a smarter model. The core was good enough for the visionaries years ago. What's been missing is every ring around it, and building those rings is exactly the "boring" reliability engineering that the earlier articles in this issue laid out. Moore would recognize the pattern instantly: the technology didn't stall for lack of brilliance. It stalled for lack of a whole product.

The crossing strategy follows directly. You don't cross the chasm by selling AI to everyone; you cross it by picking a beachhead — a single, narrow use case where the value is undeniable and the risk is bounded — and building the complete whole product for that one niche until pragmatists in it trust you. Their references become the bridge to the next niche, and the next, until the mainstream tips. It is the opposite of the "AI everywhere" sprawl that produced the 95% pilot graveyard. Focus, not breadth, is what carries a technology across.

05 — The Wider ChasmWhy AI's crossing is harder than the ones before it

It would be too neat to say AI is just another technology facing the usual chasm. In one crucial respect its chasm is wider than the ones software crossed before it, and leaders should understand why. Every prior platform pragmatists adopted was, at bottom, deterministic. A spreadsheet that computed a total gave the same total every time; if it worked in the demo, it worked in production. Reliability, once achieved, stayed achieved. AI breaks that contract. It is probabilistic by nature — the same input can yield a different output, and "it worked when I tested it" carries no guarantee about the next ten thousand calls. The pragmatist's instinct that a working demo implies a working product, an instinct every previous technology rewarded, is exactly the instinct AI punishes.

This has two consequences. First, the whole product for AI has to include something no earlier whole product did: machinery for managing non-determinism — the grounding, evals, guardrails, and monitoring that convert an unpredictable core into a dependable service. That's a heavier lift than adding documentation and a support line, which is part of why the crossing is taking longer than the optimists predicted. Second, AI's failures are unusually public and reputational. A pragmatist evaluating a database doesn't fear that a bug will become a viral screenshot; a pragmatist evaluating a customer-facing AI has watched exactly that happen to peers. The Casebook failures in this issue don't just represent risk — they actively raise the height of the chasm wall, because every pragmatist has filed them as a reason to wait one more quarter.

The encouraging half of this analysis is that a wider chasm rewards the disciplined even more richly. When crossing is easy, everyone crosses and advantage evaporates. When crossing is genuinely hard — when it demands trust infrastructure most competitors are too impatient to build — the few organizations that do the boring work of the whole product don't just reach the mainstream; they arrive there with a moat. The very difficulty that has stranded 95% of pilots is what will make the eventual crossers durable.

Sources: Geoffrey A. Moore, Crossing the Chasm (1991; rev. eds.) — technology adoption life cycle, the chasm, and the whole-product concept; MIT, The State of AI in Business 2025 (~95% of GenAI pilots deliver no measurable P&L impact); S&P Global Market Intelligence, 2025 (42% abandoning most AI initiatives); Moffatt v. Air Canada, 2024.
Companion essay · The AI Mirage · AI Edge for Leaders.