Breaking

Open-source 'loop-engineering' framework pitched for self-improving trading agents

July 8, 2026 at 05:15 EDT

AI Agents
Open Source
Finance

An open-source framework called "loop-engineering," which lets AI coding agents design their own self-improving loops, has been released and is drawing attention as a way to build trading agents. Released for free under the MIT license, the repository is available on GitHub at cobusgreyling/loop-engineering.

Jun–Jul 2026 · Open Source

"Loop Engineering" reframes agents as loops, not prompts

A new open-source framework offers a practical template for building self-improving coding and trading agents — designing the iterative loop itself, complete with a 12-step self-running quant desk that spread widely.

Steps in the self-running quant desk sequence

Production-oriented loop patterns included

L1→L3

Loop Ready score: report → assisted → unattended

The self-running quant desk loop

A 12-step sequence that repeats until strategy converges.

Strategy intent → Market data → Signals → Trading agent → Verify → Refine ↺ Rerun

Patterns by execution cadence & token cost

Column height reflects relative token cost per pattern.

Low–Med

Daily Triage

Periodic triage

Medium

PR Babysitter

PR monitoring

Med–High

CI Sweeper

CI repair

What developers value

Loop Ready score & audit tooling
Parallel safe fixes via Git worktrees
Maker/Checker sub-agent split
Model Context Protocol (MCP) support

Cautions raised

Ballooning token costs on long loops
"Unattended loops make unattended mistakes"
Verification is still on you
No trading benchmarks or long-term results yet

"Verification is still on you."

Concrete win rates and profit-and-loss track records in real trading remain limited — how much reporting accumulates from real operators will decide the framework's standing.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.