AI Tools Usage & Reimbursement Policy
"The only way you win if you are smaller, is by being faster. Automation compounds our ability to deliver."
AI tools are leverage. This page describes both our cultural posture and our policy for using and expensing them.
Policy
AI tool subscriptions are required for engineering team members.
Team members are expected to maintain active access to one or more top-tier AI coding tools and use them as part of their day-to-day workflow. In practice, this may include tools such as:
- Claude Code Max
- Codex / ChatGPT Pro
- Cursor
- Copilot
- OpenCode
The point is not tool compliance for its own sake, nor strict standardization on a single vendor or workflow. The goal is to build fluency with the systems that materially improve speed, clarity, and output.
This is consistent with Acceleration and We Ship Experiences, Not Code: we use leverage where it helps us deliver better outcomes faster.
We expect people to actively experiment with the tools that best fit their work, including using multiple tools when that improves planning, implementation, review, or iteration. Team members should also ensure they have access to Linear via MCP through at least one agent in their workflow so project and ticket context can be pulled in without manual copy-pasting. The company will reimburse relevant plans that materially improve engineering leverage.
Team Expectations
Kitchen Communication applies here too.
If you find something effective, share it with the team. That includes:
- workflows
- prompts
- review patterns
- test patterns
- failure modes
- tool comparisons
- reusable skills or configuration
- setup or folder conventions that make good workflows portable
The goal is team acceleration, not isolated personal advantage.
Quality Expectations
Tool choice is flexible; quality standards are not.
AI-assisted work is still subject to normal expectations for correctness, maintainability, review rigor, and testing quality. Human accountability does not go away because a tool was involved.
In particular:
- AI-generated code should be reviewed with the same care as any other code
- AI-generated tests should be evaluated critically, especially when they rely too heavily on mocks or fail to exercise real behavior
- team members should share patterns that improve signal in review and testing, not just speed in generation
Reimbursement
- US employees: request reimbursement through Ramp
- Contractors: include purchases as line items on your invoice and attach receipts
- If upfront cost is a burden: raise it directly so another arrangement can be made
TL;DR
Use serious AI tools. Experiment actively. Share what works. Do not prematurely standardize on one tool, but do maintain high standards for review, testing, and code quality.