DEV Community

Rick Leffke
Rick Leffke

Posted on

Are You Paying for AI Coding Seats Nobody Uses?

If your company writes software, there's a good chance your developers now use an AI coding agent. That's usually a good thing. These tools draft code, explain unfamiliar parts of a codebase, and handle the boring edits. But they also add a line to your monthly costs that tends to grow quietly, and few owners can say exactly what they're paying for.

Full disclosure: I built Agent Ready Rankings (https://rankings.leffke.com), a site that scores AI agent tools out of 10. The scores below are overall scores from my latest data run, not a same-day refresh. Browsing the site is free; the full score breakdown is part of Pro, $10/month.

Two ways the bill grows

Most coding agents charge in one of two ways.

Per seat. Each person gets a license for a monthly fee. It's predictable, but seats are easy to forget. Someone changes roles, a contractor finishes, a person leaves, and the seat keeps renewing. Nobody notices because the amount per seat is small.

Per use. You pay for what the tool does, often based on how much work the underlying AI model handles. There are no idle seats, but a busy month, or one person running long agent sessions, can make the bill jump without warning.

Then there are free, open-source tools. The tool costs nothing, but it still has to call an AI model, and that model usually bills by use. "Free" here means you're paying the model provider directly instead of the tool vendor.

Someone has to own the list

The practical fix is boring: someone owns the list of who has access, and checks it against who's actually using it. Some vendors are making that easier. On November 10, 2026, GitHub announced that enterprise GitHub Apps can now call the Copilot seat management APIs to control which users and teams are assigned Copilot seats. In plain terms, a company can tie seat assignment to its own systems, so people who leave can lose their seat automatically instead of whenever someone remembers.

The coding agents I track (overall scores)

Cursor: 8.75 out of 10. An AI-first code editor with an agent mode and multi-file edits. Free tier with paid plans. Latest tracked update November 6, 2026.

Claude Code: 8.4 out of 10. Anthropic's coding agent for the terminal. Billed by use. Latest tracked release v2.1.322 on November 9, 2026.

Windsurf: 8.05 out of 10. An agentic code editor. Free tier with paid plans. Latest tracked release v3.12.43 on November 9, 2026.

GitHub Copilot: 7.4 out of 10. The widely adopted copilot built into VS Code and GitHub. Paid. Latest tracked update November 10, 2026, the seat management change above.

Aider: 7.3 out of 10. Open-source pair programmer in the terminal with strong git integration. Free; you pay for the model it uses. Last tracked release February 11, 2026.

Continue: 6.7 out of 10. Open-source assistant for VS Code and JetBrains where you bring your own models. Free; model costs are yours.

A short owner checklist

Ask whoever runs your engineering team:

  1. Which AI coding tools are we paying for, and is each one per seat or per use?
  2. How many seats do we have, and how many were used last month?
  3. When someone leaves or changes roles, who removes their access, and how fast?
  4. For usage-based tools, is there a monthly cap or an alert before the bill jumps?
  5. For free, open-source tools, which AI model are they calling, and on whose account?

None of this needs you to understand the code. It's the same discipline you'd apply to any other subscription, just for a category that's growing fast.

About the rankings

Agent Ready Rankings scores AI agent tools out of 10 and is free to browse. The coding agents category is here: https://rankings.leffke.com/categories/coding-agents. I built it, so treat the scores as one informed opinion, and tell me where you think they're off.

Top comments (0)