Engineering teams everywhere run on AI tools. Few organizations have settled who picks the AI model for each task, or who tracks the cost. A 2026 PwC survey of more than 4,000 chief executives shows the price of leaving that unsettled. Of those CEOs, 56% say AI has not increased revenue or cut costs over the past year.
Only 12% of them saw AI raise revenue and cut costs, a group PwC calls the vanguard. It traces their lead to stronger technical foundations, built so AI can work across the whole company instead of in scattered pockets. Companies with that kind of foundation are three times more likely to report a meaningful return.
A separate BCG report puts a number on how rare that foundation is. BCG surveyed 1,250 senior executives worldwide. Only 5% of their companies get substantial, consistent value from AI across their operations, a group BCG calls future-built. Another 35% are working toward that, and the remaining 60% see minimal revenue or cost gains and lack the basics to scale AI.
Engineering posts from a handful of tech companies point to what the small group does differently. These companies build a system around the AI model that decides which tool handles each task and records what happened. At Stripe, AI programs called Minions write code on their own, and each week Stripe accepts more than a thousand finished pieces of it. An engineer describes the job in a message and walks away, and a person reviews the code only after it passes automated tests.
Uber and DoorDash describe a related pattern company-wide. Uber sends every request to an AI model through one checkpoint, which it calls its GenAI Gateway, and tracks usage and budget there instead of team by team. DoorDash’s similar checkpoint, its Agent Gateway, checks who sent each request, limits how many come in and logs what each cost before passing it to the right tool. At both companies, no engineer connects to a model directly.
That setup lets a company send routine work to a cheaper model and save a stronger one for harder planning, while still seeing what each choice costs. A record of which model did each job, and at what cost, is the kind of foundation PwC found behind the vanguard’s lead.
