Enterprise Stipends Hit $2,000/Month as AI Coding Tools Sprawl Past One Per Developer
Brex now reimburses engineers $2,000/month for AI coding tools, Uber tops out at $1,500. The new question isn't which assistant to pick — it's how to manage a stack of four.
Two years ago, the AI coding tools debate was binary: Copilot or nothing. As of June 2026, the conversation has flipped. Brex is reimbursing engineers up to $2,000/month for AI coding tools, with Uber not far behind at $1,500 — and as Simon Willison points out on X, those caps may be per-tool, meaning an engineer running Cursor, Copilot, Claude Code, and Codex side-by-side could be expensing $6,000/month in subscriptions. The story isn't a single winner. It's that the category has fragmented into specialist tools, and budgets are catching up.
The sprawl is happening because each tool has carved out a distinct lane. Cursor, the IDE-first assistant, has grown to over half a billion in annualised revenue according to VaaSBlock's 2026 landscape review, built on inline edits and codebase-aware completion. Claude Code, by contrast, lives in the terminal and Claude Desktop — Pasquale Pillitteri's June 2026 guide describes it as the tool that "writes code, runs tests, manages Git repositories," pitched at autonomous task execution rather than autocomplete. Copilot remains the default for GitHub-native teams. Windsurf occupies the agentic-IDE niche. They're complementary, not substitutes, which is why finance teams are writing four-figure monthly stipends instead of picking one vendor.
The self-hosted track is also maturing fast enough to be a real procurement question. DeepSeek shipped V4 on April 24, 2026, with V4-Pro at 1.6T total / 49B active parameters and V4-Flash at 284B total / 13B active, both MIT-licensed with 1M context windows, per the Pinggy roundup of open-source coding LLMs. For regulated industries that can't send code to a third party, the open-weights gap has closed enough that a self-hosted V4-Flash on internal infrastructure is a credible alternative to a $200/seat Cursor subscription.
For builders, the implication is that AI coding spend is now a portfolio decision, not a tool decision. For engineering managers, the budget question is shifting from "can we afford it" to "how do we audit four overlapping subscriptions per engineer." For platform teams at regulated companies, the self-hosted option finally clears the bar where it's worth piloting rather than dismissing.
The bottom line: If you're still treating AI coding tools as a one-vendor purchase, your engineers are already running three of them on personal cards — get ahead of the policy now.