Why do general AI tools like ChatGPT spread faster than AI built just for one industry's workflows?

This explores why general-purpose AI tools spread faster than AI built for a single industry. One caveat first: no note in the collection directly compares horizontal and vertical products. What the corpus offers is two halves of an answer. One half explains why general tools get adopted with almost no friction. The other explains where industry-specific value stalls. Together they suggest the gap is less about the products than about who has to say yes.

The adoption numbers for general tools are striking. Generative AI reached 54.6% of US adults within three years of ChatGPT's release. The PC reached 19.7% at the same point and the internet 30.1% Is generative AI adoption now outpacing the personal computer?. The key is that one person can adopt a horizontal tool alone. Developers show this most clearly: 80% use AI tools even though only 29% trust the output Why do developers keep using AI tools they don't trust?. Individual uptake doesn't need trust, budget approval or process redesign, only a login. Prices help too. The cost of a given level of AI performance falls about 66% per quarter right after a capability debuts Do AI performance prices fall faster when capabilities debut?. That falling cost makes general models cheap enough to try before anyone has written a business case.

Industry-specific solutions hit a different wall. They only pay off when they reshape how a whole organization works, and that is slow. One formal model finds that AI improves individual tasks but makes decisions vary more across people. Capturing the real gains therefore means redesigning how interdependent decisions are coordinated Does AI adoption require rethinking how organizations coordinate?. Benedict Evans makes a related point: making tools easier to build doesn't help if workers don't see their own tasks as automatable, or if adoption needs sign-off across several departments and budget cycles Does easier tool-building actually solve enterprise adoption problems?. A vertical product is effectively a bet on that redesign, so it moves at the speed of committees.

A less obvious reason is that the evidence for vertical value is weaker than it looks. An analysis of 960 real occupational workflows found that agents win contest-style benchmarks but fail at long, multi-step professional tasks. This is partly because the field has been measuring contests rather than real work Why do agent benchmarks not predict real economic value?. A horizontal tool can win on impressive demos. A vertical one has to prove itself inside a messy, specific workflow, and today's benchmarks don't capture that. Gains also aren't spread evenly: firms more exposed to AI replace contract labor faster and more cheaply than other firms, which suggests in-house capability compounds Do firms substitute labor for AI at different rates?.

The surprising implication is that 'scaling faster' and 'creating more value' may be different things. In the BCG experiment, consultants who were required to use GPT-4 narrowed the skill gap. When use was optional, that equalizing effect could reverse Does AI help lower-skilled workers catch up to top performers?. Executives are also unsure whether agentic AI is a tool or a coworker, and that uncertainty complicates any organization-wide rollout Does agentic AI fit the tool-or-coworker framework?. Horizontal tools win the race to adoption because individuals can choose them alone. The larger gains, the ones that show up in organizational redesign and labor substitution, still depend on the slower vertical, organization-level work.