AI tooling: boom, then hiccups
- Enterprise AI coding tools are scaling rapidly even as autonomous agent projects run into reliability problems. - Cursor is reportedly in talks to raise $2 billion after hitting about $2 billion ARR with large enterprise clients. - That contrast shows strong demand for AI dev tools but also operational immaturity in agent systems, risking wasted compute and chaotic workflows ( ).
Cursor’s parent company is in talks to raise about $2 billion at a valuation above $50 billion, even as engineers warn that many AI agents still break in expensive ways. (bloomberg.com) Bloomberg reported on April 17 that Anysphere, the startup behind Cursor, is discussing a new round led in part by existing backer Andreessen Horowitz, with Nvidia planning to participate and Thrive Capital also in talks. (bloomberg.com) Cursor crossed $2 billion in annualized revenue by early March, according to TechCrunch’s report on Bloomberg’s sourcing, and large corporate buyers now account for about 60% of its revenue. (techcrunch.com) An AI coding tool is simpler than a fully autonomous agent: it suggests, edits, and explains code inside a developer’s workflow, while the human still decides what ships. That model has been easier to sell to enterprises than software that is supposed to plan, act, and supervise itself across many systems. (techcrunch.com, cnbc.com) At two Silicon Valley events this week, CNBC reported, executives and engineers described AI agent systems as “rickety,” costly to run, and hard to manage at scale. Google software engineer Deep Shah said large multi-agent systems hit inference-cost problems first, before teams even solve the rest of the operational complexity. (cnbc.com) Kevin McGrath, chief executive of startup Meibel, said at one session that too many teams are routing every task through a large language model, which can burn money instead of saving it. CNBC said those discussions followed the rise of OpenClaw, a harness for building and coordinating fleets of digital assistants. (cnbc.com) The split has widened over the past few months. Cursor was valued at $29.3 billion in its November 2025 funding round, TechCrunch reported, and the new talks would mark another sharp step up in investor pricing if the deal closes. (techcrunch.com, bloomberg.com) Competition inside AI coding is also getting tighter. TechCrunch said some individual developers and smaller startups have moved from Cursor to Anthropic’s Claude Code on price, while OpenAI’s Codex, Replit, Cognition, and Lovable are also chasing the same market. (techcrunch.com) Nvidia chief executive Jensen Huang told CNBC in March that AI agents are “definitely the next ChatGPT,” but the engineers speaking in San Jose this week described a market still working through monitoring, cost control, and reliability. (cnbc.com) For now, the money is flowing fastest to tools that help developers work with AI one step at a time, while the systems meant to run on their own are still being debugged in public. (bloomberg.com, cnbc.com)
