AI Engineers' Favourite AI Engineer

Mutagent, an AI engineer that reads traces, diagnoses failures, and ships fixes on a loop, launched today in research preview as a free tool. The company reports that running it on an AI recruiting company's agents cut a $12,490 monthly bill by a third in the first month, with one missing formatting instruction costing $1,193 per month alone.

We're your AI Engineer's favorite AI Engineer Mutagent is an AI engineer that reads your traces, finds the failure, and ships the fix, on a loop. It is live today in research preview, free to try. We’re your AI Engineer’s favorite AI Engineer Every team is shipping AI agents. The real challenge is keeping them reliable. We lived in that problem for 3 years. Now we are changing it. The agent factory is coming alive. Agents fail in ways you cannot predict Your observability tools show you what broke. They do not tell you how to fix it, and they do not fix it for you. So an engineer opens the traces, reads through them by hand, finds the bad call, tweaks a prompt, and ships. Then the next failure shows up and the whole thing starts over. The way teams handle this today is to stitch together a tracing tool, an eval framework, and a running pile of prompt tweaks, then repeat that work every single time something breaks. It is slow, it lives in one person’s head, and it never really ends. What if the loop ran itself We kept asking a different question. What if the whole loop ran itself, the way a senior engineer would. Read the traces. Derive the evals. Diagnose the failure. Ship the fix. Then do it again. What we built Mutagent is an AI engineer that installs into your coding agent. Any trace source goes in, whether that is Langfuse , LangSmith , OpenTelemetry , or local traces. Any agent or skill comes out, whether you are on LangGraph , Mastra , the Claude Agent SDK , or your own stack. It finds the guardrails your agent is missing and applies the fix on your approval, or automatically if you opt in. What it found in production We ran it on our own agents first. Then we ran it on an AI recruiting company. | Finding | Value | |---|---| | Monthly agent bill at the start | $12,490 | | Share of that spend that was unnecessary | about 60% | | Cost of one missing formatting instruction | $1,193 per month | | Reduction in the first month | a third of the bill | | Model changes required | 0 | One missing formatting instruction was costing $1,193 a month on its own. Nobody was going to find that by reading dashboards. Try it Mutagent is free to try, in research preview. Start with your Claude Code and Codex agents, or with a sample repo, and get a real fix in about 5 minutes. npm install -g @mutagent/cli Or book a demo https://calendar.app.google/PyWYia8XmQ8AHmb59 and we will walk your agents through it with you. Stop debugging. Start evolving.