Tech • AI • Robotics • Game
Everyone Is Buying AI, but the Money Gets Stuck in the Same Place
7/10TL;DR
AI spending is shifting away from generic tools and toward specialists who can move pilots into production inside a specific industry, where execution, governance and last-mile integration now command the highest fees.
KEY POINTS
A market rich in budgets but stuck in pilots
Companies have bought models, subscriptions and executive titles, yet many still fail to operationalize AI. An IBM study found 76% of large organizations had an AI lead last spring, up from 26% a year earlier, but 53% of executives said projects remained in pilot or experimentation rather than production. The central commercial problem is no longer access to models but making them work reliably inside existing firms.
Why simple AI products are losing value
Thin wrappers around foundation models are becoming harder to sell as model providers absorb features and cut costs. The Stanford AI Index reported that the price of a machine response fell by a factor of 280 in two years at constant performance. Usage-based pricing, such as GitHub Copilot billing by consumption from June 1, 2026, further squeezes resellers caught between platform costs and clients unwilling to fund anything short of real deployment.
Generic automation and SaaS face the same pressure
Hyper-personalized integration is getting cheaper than forcing clients into rigid software. That weakens the classic subscription model for broad, standardized tools and shifts value toward bespoke implementation. Even large software groups have signaled this change through deeper partnerships with model companies rather than insisting on full-stack control.
The old content-volume model is breaking down
The decline in the marginal cost of producing generic text, code and chatbot interactions is eroding business models built on scale and traffic. The lesson is that abundance drives commodity prices toward zero. In the AI era, clients pay less for standard information and more for precise execution in a narrow context, backed by domain expertise and accountability.
External partners are more likely to get projects live
A MIT report published in July observed roughly 50 organizations and found that AI projects led with an external partner reached production in about 67% of cases, versus 33% for efforts kept fully in-house. That gap suggests the biggest obstacle is often not the model itself but organizational execution, decision-making and credibility across departments.
The people getting paid are deployment specialists
The critical role is increasingly the deployment engineer or transition lead: the person who sits with teams, handles exceptions, opens real files and keeps systems standing after launch. Harvey, a legal software company, has described placing a former practicing lawyer into deployments so business rules reflect what law firm partners will actually accept. Clients are buying the transition to production, not just the tool.
The winning offer is vertical, scoped and time-bound
The most credible commercial approach starts with one industry the provider already understands, such as law, transport, accounting, construction or agriculture. From there, the offer should be narrow: a diagnostic, then a fixed production phase with a written go-live date and a third party able to validate the rollout. Boardrooms may discuss transformation, but they sign contracts around scope, timing and accountability.
Pricing reflects interim leadership more than coding
Transition-management style work is already commanding meaningful rates. Human Partner reports that experienced executives on AI missions bill roughly €1,000 to €2,000 per day, firm margin included. At 120 billable days a year, that implies €120,000 in billings even at the low end, though actual take-home pay is lower and depends on winning and retaining clients.
Case studies show where returns are being found
In a spring white paper by Bpifrance Conseil and SI Partners, Aldes, a French ventilation manufacturer, developed an AI solution for small quotations over five months. The build cost was €40,000, annual operating cost €60,000, and estimated first-year gain €800,000. Selectour also put a virtual agent into production in late November 2025, underscoring the importance of choosing the right implementation partner.
Who wins and who gets commoditized
A PwC analysis of 1 billion job offers found demand for AI-related roles growing twice as fast, with wages up 42%. Upwork said AI consulting and strategy were growing about 50% annually even as its total marketplace volume fell 4%. The premium is flowing to people who exercise judgment, manage risk and solve the last mile, while routine execution is increasingly automated.
CONCLUSION
The strongest demand in AI is not for another chatbot or generic software layer, but for trusted operators who can carry a project from pilot to production in a specific business context. The money is going to last-mile execution, and firms are paying a premium for judgment they can put into production.
Ask a question