# Designing agent systems

> Structuring agents: orchestration and multi-agent patterns, tool and MCP server design, memory, planning loops, state and handoffs.
> 2 recommendations, newest first. Canonical: https://wpnews.pro/playbooks/agent-architecture

- Give each parallel coding agent its own isolated remote environment with simulators and SSH access, and bar runner agents from editing the tests they execute so passing results stay trustworthy.  
  — [How Shopify Orchestrates AI-Powered Mobile App Development Without Overloading Developers' Machines](https://wpnews.pro/news/how-shopify-orchestrates-ai-powered-mobile-app-development-without-overloading), source: <https://www.infoq.com/news/2026/11/shopify-remote-ai-dev/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global>, 2026-11-17
- Separate planning from control: pair a slow reasoning model with a fast on-robot policy, since Astra alone averaged 87.3 seconds per decision and succeeded on only 8.3% of pick-and-place tasks.  
  — [CoRL 2026: takeaways, launches and the papers to know](https://wpnews.pro/news/corl-2026-takeaways-launches-and-the-papers-to-know), source: <https://read.corematter.com/p/corl-2026-recap>, 2026-11-12
