Enterprise AI is practical
AI adoption in large organisations has moved from pilots to budgeted deployments, but wins are concentrated in practical areas like coding, customer support and enterprise search rather than glamour use‑cases. Firms are rewarding tools that slot into existing workflows and show measurable ROI, and enterprise contracts now account for a growing share of leading AI vendors’ revenue. That concentration matters because vendor selection should prioritise reliability, integration and economics over headline model accuracy. (a16z.news) (cnbc.com)
The surprise in corporate artificial intelligence is that the money is not flooding into robot coworkers or fully autonomous agents. It is going into code assistants, customer support tools, and enterprise search systems that answer questions from a company’s own documents. (a16z.com) Andreessen Horowitz said coding is the standout use case by revenue momentum, with customer support and search next behind it. The firm’s dataset came from leading enterprise artificial intelligence startups, public data, and anonymized information from thousands of conversations with startups and large companies. (a16z.com) That is a big change from 2023 and 2024, when many large companies were still running small pilots to see whether these tools worked at all. In 2026, the question is less “can we try this” and more “which tool fits the software we already use and can finance approve the budget.” (a16z.com) (forbes.com) The pattern is practical because these three jobs are easy to measure. A coding assistant can be judged by developer output, a support bot by ticket volume and resolution time, and an enterprise search tool by how quickly employees find the right contract, policy, or sales note. (a16z.com) The vendor race is also splitting by task instead of producing one universal winner. Andreessen Horowitz found OpenAI strongest in general chatbots, knowledge management, and customer support, while Anthropic led in software development and data analysis. (a16z.com) That split tells you what buyers are rewarding. A legal team searching internal files cares about document access controls and audit trails, while an engineering team cares about how well a model handles code, version control, and bug fixing inside existing development tools. (a16z.com) The revenue mix at the biggest vendor now reflects that shift. OpenAI chief revenue officer Denise Dresser told CNBC on April 8, 2026, that enterprise accounts for 40 percent of OpenAI’s revenue and is on track to match consumer revenue by the end of 2026. (cnbc.com) Once enterprise contracts become that large, the buying criteria change. A company signing a multimillion-dollar software contract usually cares less about a benchmark lead on a leaderboard and more about uptime, security reviews, procurement terms, and whether the tool plugs into identity systems, document stores, and customer databases. (cnbc.com) (a16z.com) That is why the glamorous vision of one model replacing whole departments keeps losing budget fights to narrower tools with clearer economics. A support assistant that cuts call-center workload by a measurable percentage is easier to defend in a board meeting than a broad promise to “transform work” sometime next year. (a16z.com) The companies moving fastest are not necessarily buying the flashiest demos. They are buying the systems that behave like boring enterprise software: reliable on Monday morning, connected to the software stack they already own, and cheap enough that the savings show up before the renewal date. (a16z.com) (cnbc.com)
