Coding is enterprise's biggest AI use

Big companies spent years talking about artificial intelligence as a chatbot for customer service, and the first place the money really piled up was software engineering. Andreessen Horowitz said code generation is running at about $3 billion in annualized enterprise spend, far ahead of legal work at about $500 million and support at about $400 million. (a16z.com) That gap is so wide that Andreessen Horowitz called coding “an order-of-magnitude outlier” even compared with the next biggest use cases. In the same April 8, 2026 report, the firm said coding, support, and search make up the lion’s share of enterprise artificial intelligence adoption. (a16z.com) The simple reason is that code is already written in text, stored in files, checked by tests, and reviewed against a clear output. A model can change ten files, run the test suite, and show a human exactly what moved, which is much easier than asking it to make a legal judgment that could trigger a lawsuit. (anthropic.com) (openai.com) The tools are now built for that workflow instead of just spitting out snippets. Anthropic says Claude Code can read a codebase, make changes across files, run tests, and deliver committed code, while OpenAI says Codex is designed to handle planning, refactors, reviews, and releases across editors, terminals, and cloud environments. (anthropic.com) (openai.com) That is why the buyers are not just individual programmers with a credit card anymore. Cursor says its enterprise product is built for codebases with millions of lines and hundreds of thousands of files, and its October 2025 launch post said the product was already used by tens of thousands of enterprises including Salesforce, NVIDIA, and PwC. (cursor.com 1) (cursor.com 2) The same pattern shows up in broader enterprise adoption. Andreessen Horowitz said 29% of the Fortune 500 and about 19% of the Global 2000 are already live, paying customers of a leading artificial intelligence startup, and it counted only deployments that moved from pilot to signed contract to real production use. (a16z.com) That is unusually fast for big-company software buying. Andreessen Horowitz wrote that startups usually had to win smaller customers first and wait years for a Fortune 500 contract, but artificial intelligence compressed that cycle after ChatGPT launched in November 2022 and pushed boards and chief executives to test tools much earlier. (a16z.com) The spending is also getting more permanent. In its June 10, 2025 survey of 100 chief information officers across 15 industries, Andreessen Horowitz said artificial intelligence budgets had moved out of one-time innovation funds and into recurring information technology and business-unit budgets, with leaders expecting average spending growth of about 75% over the next year. (a16z.com) Once coding became the first category with a clear return, the model companies followed the money. OpenAI now markets Codex as an engineering agent, Anthropic markets Claude Opus 4.6 and Claude Code for professional software engineering, and Cursor sells security controls, audit logs, and admin settings that look like standard enterprise software plumbing. (openai.com) (anthropic.com) (cursor.com) So the shape of enterprise artificial intelligence is getting easier to see. The first breakout business is not a robot that runs the whole company; it is a very expensive assistant sitting next to engineering teams, touching the one kind of knowledge work that already has version control, test harnesses, and a numeric definition of done. (a16z.com) (openai.com)