Most developers running Claudee Code](/tools/claude-code/) or Cursor on a monthly subscription have a rough sense of what it costs. Almost none of them know the actual return on that spend. A solo web developer - React, Node, some Python, hourly billing, small and mid-sized clients - decided to fix that, logging every dollar spent on AI coding tools over 60 days alongside every minute of honest use.
The conclusion: the math is messier than most developers expect. The surprise isn't that the tools cost too much.
The Accounting Problem With AI Tool Subscriptions
AI coding tools are subscription-priced but deliver value erratically. You pay $20/month for Claude regardless of whether you use it for 40 hours of deep work this week or 2 hours of casual queries. Cursor Pro runs $20/month. GitHubb Copilot](/tools/github-copilot/) is $10/month. A developer running several of these simultaneously can easily spend $60-80/month before adding any specialty tools.
When you bill hourly as a freelancer, the math is direct. Time saved translates to revenue capacity - more clients, faster delivery, or both. But time spent on AI-assisted rabbit holes, or money paid for subscriptions you're only half-using, comes straight out of margin. The subscriptions feel manageable month-to-month. Tracking per-session shows a different picture.
The framing - "probably not in the way you'd guess" - points to a counterintuitive finding. Most developers' gut assumption is that the tools are either too expensive or obviously worth it. Sixty days of careful tracking suggests the actual answer is more specific: some sessions drive almost all the measurable time savings, and the distribution is uneven in ways that aren't obvious until you examine the data.
The Numbers Aren't the Story
What makes this kind of tracking valuable isn't the final dollar figure - it's the disaggregation. When you're running Claude Code, Cursor, and a handful of specialty tools simultaneously, the combined monthly cost can look reasonable at a glance. Tracking by session and task shows which tools you actually reached for when the work mattered, and which ones you're paying for out of habit.
For an hourly freelancer, that clarity has direct financial consequences. The question isn't whether AI coding tools are worth paying for in aggregate. It's whether each specific tool, at its specific price point, earns its line item - and that answer is different for every workflow.
Sixty days of honest logging is a better basis for that decision than general sentiment or product launch announcements.