Too Many Technology Partners, Not Enough Projects
There is a strange imbalance in the software services market right now. Hundreds of firms, from global integrators down to two-person studios, are positioning themselves as AI-ready technology partners. Meanwhile the pipeline of actual projects feels thinner than it has in years. Analyst firm HFS Research notes that the world’s ten largest IT services companies have lost more than 600 billion dollars in market value since 2021, not because their revenue collapsed, but because investors re-rated the whole sector on the belief that AI-native companies are the future and services firms are yesterday’s story.
At the small end of the market, the same pressure shows up differently. Every agency and freelancer has rebranded around AI. Supply of willing partners has never been higher. Demand has moved the other way.
Why buyers stopped calling
The reason is simple and, in part, legitimate. AI-assisted development makes building look easy. A product manager with no formal engineering background can scaffold a working demo in an afternoon. Tasks that once justified a contractor genuinely no longer do. So companies conclude the obvious thing: why pay a partner when we can do this in house?
Any honest consultancy should concede the first half of that argument. The skill barrier really has dropped. Writing code is no longer the scarce resource it was even three years ago, and pretending otherwise is a losing sales strategy.
The numbers say something else is going on
If skill were the only constraint, project success rates should be climbing as the tools improve. They are not. RAND’s research puts AI project failure above 80 percent, roughly twice the rate of conventional IT projects. MIT’s Project NANDA found 95 percent of generative AI pilots produced no measurable return to the income statement. S&P Global recorded companies abandoning most of their AI initiatives jumping from 17 percent to 42 percent in a single year.
The tools got dramatically better during the exact period those numbers got worse. That tells you the constraint was never typing speed.
Skill is not the problem. Context and experience are.
What separates a demo from a system people rely on is two things AI does not supply.
Context is knowing your own business: why the invoice approval has three sign-offs, which customer master records are quietly wrong, what the warehouse actually does when the system says no stock. In-house teams have this in abundance, and it is a real advantage.
Experience is knowing what breaks: what happens at month three when the integration partner changes an API, what a security review will find, why the migration script that worked on test data dies on ten years of production records. This lives in people who have shipped and maintained systems for other companies, repeatedly. One study of failed AI and data projects at the University of Queensland concluded the two biggest causes had nothing to do with the AI itself: no organisational need was established, and no data processes existed to support it.
What this means for both sides
If you are an SME buyer, in-house the work that sits closest to your context and carries low blast radius. Partner where scar tissue matters: data migration, integrations, security, anything touching money or compliance. The cheapest project is rarely the one you started yourself and paid someone to rescue.
If you are a partner, stop selling manpower. The market for typing code is shrinking and it deserves to. The market for accountability, for making software survive contact with production, is about to grow, because the 80 percent failure rate is deferred demand wearing a disguise.
If you are weighing a build right now and want a second opinion before committing either way, our assessment is designed for exactly that conversation.