How to Build an Internal AI Agent That Evolves Itself
Summary
Two founders hit $2M ARR with an AI agent that processes 100+ emails daily and self-extends by automatically building its own tools—a replicable architecture using Claude, read-only database access, and an editable instructions file that evolves through feedback.
Key Takeaways
- Self-extending agents close the capability gap by delegating tool creation to a coding sub-agent; Answer This's agent went from a skeleton to 45+ self-authored CLIs without manual engineering.
- Give agents three types of memory: factual (codebase + database), behavioral (instructions.md file edited per turn), and procedural (auto-created tools). This structure enables non-technical founders to iterate on agent behavior via Slack messages.
- Provide read-only access to your full codebase and database as a baseline; the agent can autonomously answer business questions and understand subscription logic without hardcoding business rules.
- Use Claude Code CLI wrapped in Python as the harness with a task queue fed by Slack, email, and other channels; this lets agents inspect files, run commands, and use external CLIs natively.
- Load an editable instructions.md file on every agent turn; non-technical team members can correct agent behavior directly via Slack, and the agent updates its own instructions for permanent improvement.
Related topics
Transcript Excerpt
Hello everyone. My name is Ayush. I'm the founder of Answer This. We build AI agents for evidence-based scientific workflows, and today I'm going to be sharing about how we're using AI agents internally and how you can replicate our setup. So, we've been able to do over $2 million in ARR largely being two full-time employees, which is myself and my co-founder. Do have two or three contractors for things like design and outbound, but a large reason for why we've been able to do this is because we have an internal AI ops agent that handles a lot of the work that would normally consume founder time. So, this AI agent is processing more than 100 emails a day for us, has closed over 400 customer support tickets. It handles CRM updates after meetings. It collects user feedback across channels, h…