Enterprise AI has left pilots

The enterprise story around artificial intelligence changed when finance teams started treating it like normal software instead of a science project. Andreessen Horowitz said enterprise spending has moved out of innovation funds and into recurring information technology and business unit budgets, with chief information officers expecting average large language model budget growth of about 75% over the next year. (a16z.com) That shift usually happens only after a company finds a job simple enough to measure. The clearest early wins are narrow tasks with a before-and-after number attached, like software code generation, customer support deflection, and internal question-answering on company documents. (a16z.com) Coding help is the easiest example to picture. A coding assistant sits inside a developer’s editor like an autocomplete system for whole functions, and an International Data Corporation report published in May 2025 said enterprise adoption had expanded rapidly enough that security, compliance, and value measurement had become core buying criteria. (services.google.com) Customer support is the second big bucket because every ticket already has a cost, a handle time, and a resolution rate. Andreessen Horowitz found that enterprise buying now favors off-the-shelf artificial intelligence apps over custom internal builds, which makes support software attractive because companies can compare vendors the way they compare help-desk software. (a16z.com) Internal knowledge work is the third bucket because most big companies already have thousands of policies, contracts, manuals, and sales notes scattered across shared drives and chat logs. A retrieval system pulls the right document the way a librarian pulls the right folder, and companies will pay for that if it saves employees from spending 20 minutes hunting for one answer. (a16z.com) Once those tools start working, the buying process gets more formal very quickly. Andreessen Horowitz said procurement now looks more like traditional enterprise software buying, with heavier evaluation, hosting reviews, and benchmark scrutiny, because switching costs rise when an artificial intelligence tool gets wired into daily workflows. (a16z.com) The surprise is that the organizational plumbing has not caught up to the spending. A CIO Dive report published on April 8, 2026 said only 14% of global enterprises using artificial intelligence reported having a clear strategy with defined goals and outcomes, while 71% said their strategy was incomplete or still developing. (ciodive.com) That leaves many companies in a messy middle where one team buys a coding tool, another team buys a meeting tool, and a third team buys a support bot with no shared rules for data, access, or accountability. The same CIO Dive report said technology leaders feel pressure to deploy before governance systems, employee training, and responsibility plans are in place. (ciodive.com) Older business software usually behaved like a calculator, because it followed fixed rules and gave the same answer to the same input every time. Mark Baker of Altimetrik told CIO Dive that generative artificial intelligence is different because it is probabilistic, which means companies also need new ways to decide who owns mistakes, approvals, and risk when the system acts less predictably. (ciodive.com) So enterprise artificial intelligence has not stalled out. It has narrowed. Companies are buying it first where the math is easiest, and the next fight is not whether the tools work at all but whether the company can stop 20 separate purchases from turning into one expensive, leaky, ungoverned stack. (a16z.com) (ciodive.com)