How to See Which AI Coding Tools Your Developers Use
Editor assistants, terminal agents, and desktop AI apps are now part of how software gets written. Here is how to see which ones your team uses, which AI services they reach, and what that usage costs.
The short version
How do I see which AI coding tools my developers use?
Start with each vendor's admin analytics, which cover one tool each. For a view across tools, look at the machines: coding assistants and CLI tools call AI services straight from the laptop. Behest Radar shows those connections, free and observe-only, with no network proxy in its default mode.
Start with the vendors' own admin analytics
If your company pays for a team or business plan of a coding assistant, its admin dashboard is the best source for that tool. Plans for tools such as GitHub Copilot, Cursor, and Claude Code typically show seat assignments and active users, and often usage metrics for the product. Use them: they are accurate for what they cover, and they come with the plan you already pay for.
Pair each dashboard with your license list. Seats with no activity are spend you can reclaim, and heavy users are the people to ask what is working.
Why one dashboard is not the whole picture
Vendor dashboards only see licensed accounts on their own product. Most engineering teams use several tools, and some of that usage never touches a company account.
Developers use more than one tool
An editor assistant for completions, a terminal agent for larger changes, a chat app for questions. Each vendor dashboard covers its own product, so no single one shows the mix.
Personal accounts and API keys stay invisible
A developer trying a new agent on a personal account, or pointing a CLI tool at their own API key, never appears in the company's admin console.
CLI agents skip the usual checkpoints
Command-line tools call AI providers directly. They rarely ask for single sign-on, never open a browser, and keep working on a laptop that is off the corporate network.
Look at the machine itself
The one place every coding tool shows up is the developer's machine. There are two useful views of it. An installed-software inventory, from your device management tool or an open-source agent like osquery, tells you which tools are present. A connection view tells you which tools are actually reaching AI services, and which services they reach.
Network logs can help while laptops are on the corporate network, but many developers work remotely or on split-tunnel VPNs, where those logs go quiet. The shadow AI detection guide compares all six places to look.
One option: Behest Radar on the developer's machine
Behest Radar is a free, observe-only app for the endpoint. Here is what it does and does not do for coding tools.
- In its default passive mode it recognizes 20+ coding agents and AI apps by name, such as Claude Code, Codex CLI, Gemini CLI, Cursor, Aider, and Windsurf, with no network proxy, no certificate, and no changes to your tools. An agent it does not recognize still appears under its process name.
- It also sees local models run with Ollama, LM Studio, or llama.cpp, and names the model.
- Default counts are per connection, not per request, and it does not change, slow, or stop any traffic.
- With the optional certificate turned on, it reads the model name and token counts for CLIs and SDKs from AI responses and discards the rest. Certificate-pinned desktop apps, such as the ChatGPT desktop app and Claude desktop's main process, show connections only.
- With your permission, it reads the tools' own local records: token counts for Claude Code, Codex CLI, and Claude Cowork; list-price cost for Claude models. For a developer on a subscription, that cost is a list-price equivalent, not the bill.
- A standalone install sends nothing and has no telemetry. Prompts are not stored or sent by default; if your organization connects Radar to Behest Control, only the full sharing level adds Git branch names, session titles, and redacted prompt excerpts.
Install it and start it with administrator access, then use your coding tools as normal.
From one machine to the team, then to control
Machines an organization connects to Behest Control appear in one device inventory, so engineering leads see which AI services each machine reaches across tools, not one dashboard at a time. Each machine sends usage data at the sharing level chosen in Behest Control. Before you connect developers' machines, tell them what Radar collects and at which sharing level, and check the privacy and employee-monitoring rules where they work with HR, legal, or your privacy team.
Seeing usage is the first step. To control it, route sanctioned AI calls through Behest, the control plane for enterprise AI, where AI Token FinOps, its cost-control layer, attributes spend to teams and projects and enforces budgets before the provider invoice arrives. See how the pieces fit in the shadow AI discovery overview.
Frequently asked questions
- Can I see which AI coding assistants my developers use without a proxy?
- Yes. Vendor admin dashboards need no proxy, and Behest Radar's default mode recognizes 20+ coding agents and AI apps by name without a network proxy, a certificate, or changes to the tools. Model and token counts for other CLIs and SDKs need Radar's optional certificate; Claude Code, Codex CLI, and Claude Cowork's exact counts come from their own records, with permission.
- Does Behest Radar read my developers' code or prompts?
- Not by default. Radar sees which app reached which AI service and how much data moved, not the content. With the optional certificate it reads model names and token counts and keeps no content. A standalone install sends nothing. If your organization connects Radar to Behest Control, only the full sharing level adds Git branch names, session titles, and redacted prompt excerpts.
- Which coding tools does Behest Radar count tokens for?
- With your permission, Radar reads exact token counts from the local records of Claude Code, Codex CLI, and Claude Cowork, and estimates list-price cost for Claude models. The optional certificate gives model and tokens for other CLIs and SDKs. Codex usage shows tokens but no cost; other tools show model and tokens only with the optional certificate, and no cost.
- How do I see AI coding tool usage across the whole team?
- Combine the admin dashboards of the tools you license with an endpoint view. Machines an organization connects to Behest Control appear in one device inventory, so you can see which AI services each machine reaches, across tools. Tell developers what you collect, and respect employee-monitoring rules in your jurisdiction.
See your own coding tools first
Behest Radar is a free download. Try it on your own machine, then talk to us about a team-wide view in Behest Control.