The name is the method. An open loop is a prompt in and text out with nobody checking. A closed loop feeds its own output back through a gate — verified, reviewed, recorded — before it is allowed to count. Every system Closed Loop ships is built that way, and every rule in it came from a failure that was documented rather than hidden.
How the work is done
State in plain files, so anyone — human or agent — can read the truth. Verification by evidence rather than assertion. Separation of duties enforced by capability, not by prompts. The articles on this site are those methods written up from the inside: the architecture, the gates, and the scars.
The engineer behind it
Closed Loop is run by Damian, a senior software engineer in Sheffield with many years of experience shipping production software in demanding, regulated industries. Alongside client work he builds and runs his own AI systems: autonomous development loops on Claude Code, multi-agent audit pipelines for Web3 security work, and a personal-assistant harness that manages a real life — calendars, finances, health, memory — through plain files and scheduled agents.
- AI engineering — agentic loops, multi-agent orchestration, Claude Code skills/subagents, RAG pipelines, evaluation.
- Production delivery — many years delivering full products on short cycles; remote contracts across regulated industries and fintech.
- Web3 security — CosmWasm/Golang audit work with a custom multi-agent auditing system; paid bounty findings.
- Systems from scratch — once designed and built an award-winning five-barrel microbrewery from raw materials, control electronics included. Whole-system thinking is the throughline.
Want to work together? See services, or emailhello@closed-loop.dev.