Projects / 2026

Autonomous AI-agent development pipeline

An unattended nightly software-development loop driven by AI agents, with guardrail hooks, error tracking, and daily digests for human review.

  • AI agents
  • Guardrails
  • Human-in-the-loop
  • Observability

A nightly development loop that lets a lean team ship every night with a human in the loop.

  • Guardrail hooks limit what the agents are allowed to do.
  • PostHog error tracking surfaces failures as they happen.
  • Telegram digests give a human a daily summary to review.

Running it in production has given me hands-on experience with the oversight and reliability problems of autonomous systems, which is the area I want to research full time.