Treating agent orchestration like a factory: the 7 jobs of an orchestration layer
blaizedsouza · x · 2026-11-17
A thread on agent engineering architecture: define the 'factory foundations' in code (Terraform-style), then let the orchestration layer execute it. The layer must: read the code at a version or branch; spin up sandboxes on demand with low latency; route requests to agents with the right model and harness; prompt agents per the code; persist artifacts and traces; loop in humans to hand off or steer work; and integrate with Sentry, GitHub, Jira, Slack and other systems. The core claim: agent orchestration should be declarative and version-controlled, like infrastructure-as-code.
More from coding & agent
- UCLA and AWS paper: AI agents find math shortcuts but thinking time often cancels the gains — rohanpaul_ai · 2026-11-17
- Pixel art + Gaussian splats game experiments: all code, no assets, but roadblocks ahead — algo_diver · 2026-11-17
- Fable 5.2 praised as first model that ignores unimportant feedback and tells you to ship — neuroecology · 2026-11-17
- OpenAI's jxnlco on building taste: 'If you want good taste, you have to eat' — jxnlco · 2026-11-17
- I ran 3,360 experiments to see if AI agents recommend the right products — santiviquez · 2026-11-17
- Together AI and Ollama explain how prefix caching keeps long agent sessions fast — togethercompute · 2026-11-17