9horsemenStart building
Essays

Why Enterprise AI Fails: The Case for Small, Single-Purpose Tools

Published on 18 Sept 2026 by Shri Krishnan

An essay from 9horsemen's founder. For practical help running a business, see the blog.

Listen to this episode of Why Your Software Sucks on Spotify.

A desk worker's hands resting near a computer monitor surrounded by disorganised spreadsheets and legal documents, illustrating the frustration of manual data entry despite modern enterprise software.

TL;DR:

  • The Core Problem: Massive enterprise AI suites look great to the C-suite but fail operators like paralegals and HR admins because they hallucinate or lack deep specificity for granular daily tasks.
  • The Structural Reality: Big Tech SaaS relies on massive enterprise contracts, forcing them to build generalised bloatware rather than solving hyper-specific, localised administrative bottlenecks.
  • The Solution: Hyperspecialisation beats generalised intelligence. Small, single-purpose tools do the exact jobs that big systems handle badly.

Why do enterprise AI projects fail? Usually because the company buys one giant platform that impresses the boardroom but can’t handle the specific, messy work of the people who use it. Here’s why that keeps happening, and why small, single-purpose tools do better.

Why do enterprise AI projects fail?

Most are sold as one platform that will run everything, chosen by executives in a demo, then handed to staff whose daily work it doesn’t fit.

Every business publication today is selling a “God in a box”. The enterprise tech industry has completely abandoned localised problem-solving in favour of pushing massive, unified AI platforms that promise to run your entire operations department from a single pane of glass.

The pitch to the C-suite is meticulously engineered. A sales engineering team flies in, sets up in the boardroom, and demonstrates how an omniscient digital brain will optimise workforce allocation, draft ironclad contracts, and predict the financial quarter with zero human intervention. The interface is stunning. The CTO signs a multi-million dollar, three-year lock-in contract. The executive team celebrates the successful “digital transformation.”

Then you walk down to the ground floor and look at the actual screens.

What does an AI failure look like at work?

Often it’s a confident wrong answer on a job that matters, like an AI legal suite inventing a zoning rule for a multimillion-dollar lease.

Consider the reality of a senior paralegal at a mid-sized commercial real estate firm. The partners just handed down a mandate to integrate the new, all-in-one AI legal suite into daily operations.

On a Tuesday afternoon, she needs to cross-reference municipal zoning codes against a new commercial lease for a $40 million property development. The massive AI suite attempts the task and confidently hallucinates a non-existent municipal setback regulation.

When the stakes are a nine-figure property portfolio, a generalised AI guessing at municipal code is worse than useless—it is a catastrophic liability. The paralegal immediately realises the platform is fundamentally disconnected from the rigid, granular truth of local government databases.

She abandons the expensive dashboard. She quietly shifts her workflow back to manual processing, or she secretly expenses a tiny, hyper-specialised web app that does exactly one thing: flawlessly scrapes, indexes, and cross-references county zoning databases for specific property lines. It doesn’t write emails. It doesn’t summarise strategy meetings. It just executes the exact, specialised task the multi-million dollar software failed to comprehend, turning a four-hour bottleneck into a twelve-second automated query.

This exact scenario is playing out across every department. HR managers are bypassing “AI copilots” to manually format state compliance spreadsheets. Finance associates are ignoring predictive spend analytics because they desperately need a script that simply forces 50 different vendor invoice formats into QuickBooks without breaking the general ledger.

Why don’t software vendors fix these problems?

Because their business depends on large contracts signed by executives, not on fixing one team’s daily workflow.

Why do legacy software companies refuse to fix these granular problems? It is a structural failure of incentives, driven entirely by the venture capital model.

Venture-backed enterprise SaaS companies are engineered to close Chief Information Officers, not to help the desk worker whose Tuesday is currently on fire. The unit economics of Big Tech require massive annual contract values (ACVs) to justify their staggering customer acquisition costs. You cannot justify a seven-figure enterprise licence by pitching a tool that simply fixes a localised zoning code workflow or a payroll extraction glitch.

To justify the price tag, Big Tech is forced by their own economic models to build generalised bloatware. They must constantly add features, modules, and generalised AI wrappers to maintain the illusion of absolute comprehensiveness. They try to be everything to everyone, and as a result, they fail at the exact, highly specific friction points that actually choke your daily operations. The software is built for the buyer’s ego, not the operator’s reality.

What works better than one big AI platform?

Small tools that each do one job well, built around how a team actually works.

Real operational progress is never about generalised enterprise suites trying to be everything at once. It is about hyper-specialisation. The future of competitive work isn’t a bloated, seven-figure dashboard that requires six months of vendor onboarding; it is a fleet of custom-built, single-purpose software instruments that quietly eliminate manual bottlenecks and hand hours of productive time back to your team.

You don’t need an enterprise monolith to fix a localised operational failure. You need a scalpel.

That’s how we think about software at 9horsemen: small, focused tools beat one giant platform nobody wants to open.

Our first product, Gallop, is one of those focused tools, for local service businesses. Describe your business and it builds your website, online booking and business app in a few minutes. If Gallop is missing something your business needs, tell us at hello@9horsemen.com. We’d rather build it into Gallop for everyone than quote you for custom work.

If you’re building software of your own, I write more about it in the other essays and talk about it on the podcast Why Your Software Sucks.