Blog

Engineering notes on agent orchestration, FinOps, security, and shipping with Swarm 357.

• 2026-07-20

  Multi-agent orchestration without the headcount fantasy

  Why 357 roles is an ontology, not a billing disaster.

• 2026-07-18

  Cost control for LLM agent fleets

  Layer budgets, model downgrade, and honest telemetry.

• 2026-07-15

  Portable agent memory with Memvid

  Single-file .mv2 recall across runs and layers.

• 2026-07-12

  Why your agent needs a bash policy gate

  Pattern-based defense before autonomous shell access.

• 2026-07-10

  Human-in-the-loop approvals for autonomous agents

  Durable ApprovalRecord workflow for Bash commands.

• 2026-07-08

  How to actually evaluate an agent swarm

  Separate single-agent and swarm metrics with budget caps.

• 2026-07-05

  CrewAI, AutoGen, and role ontologies

  When a 357-role catalog beats ad-hoc crews.

• 2026-07-01

  Deploying a Claude agent runtime on Railway

  API + landing split, BFF auth, and health probes.