AgentCost now works with coding agents, not just application-level LLM calls

One thing we've been working on with AgentCost is making observability useful closer to where developers actually work.

We've expanded our coding-agent integrations to cover Claude Code, Codex, and Antigravity, with a focus on understanding usage at the session level rather than looking only at an aggregate API bill.

We're also working on the control side of the problem: configured budgets and guardrails, so developers can do more than inspect usage after the fact.

The interesting challenge is making this useful across different coding-agent environments without forcing developers to change how they work.

If you're building with coding agents, I'm curious: what's hardest for you to understand or control today: usage attribution, unexpected execution behaviour, or setting limits that actually hold?

I'd be happy to discuss the technical details and trade-offs we've encountered while building this.