liveBook · Manning

Isolation

As explained in Designing AI Agents

Isolation is a technique for containing the effects of a failure or operation so they do not propagate across system boundaries.

Isolation limits the blast radius of a failure or operation by restricting what it can affect. For agents, separate execution boundaries from coordination boundaries: a Git worktree prevents workers from editing the same checkout, but does not isolate shared databases, APIs, quotas, or configuration. section 8.3.3 section 9.3.3

Evaluate isolation by the boundary it enforces and the risks it leaves outside that boundary:

  • Filesystem policies can limit writes and reads separately; restricting writes alone does not restrict reads. section 9.3.3
  • Network policies can route outbound connections through a proxy that restricts destinations or requests approval. section 9.3.3
  • A policy-checking executor is not itself a sandbox: the supplied runtime must enforce physical isolation and applicable resource limits. section 9.3.5
  • Technical boundaries limit reachable resources; they do not guarantee business correctness. Choose stronger boundaries according to the threat model and the consequences of escape. section 9.3.3
Table 9.5 Isolation mechanisms and operational trade-offs
Mechanism
Isolation boundary
Startup/footprint evidence
Operational trade-off
Typical fit
Docker/runc
Shared host kernel
Fast process-level startup
Low overhead; host kernel remains shared
Trusted workloads with host hardening
gVisor
User-space application kernel
Adds a system-call mediation path
Compatibility and I/O cost require workload testing
Untrusted containers needing user-space syscall mediation
Firecracker microVM
Dedicated guest kernel
Design goals: <125 ms startup; <5 MiB VMM [11]
Separate guest kernel with VM image and lifecycle management
Workloads requiring a dedicated guest kernel
bubblewrap/Seatbelt
OS-level process sandbox
Process-level startup
Platform-specific policy surface; not a VM boundary
Local coding agents
WebAssembly
Language/runtime sandbox
Fast module instantiation
Host capabilities must be explicitly exposed
Plugins and portable functions

Compare isolation mechanisms by boundary, startup and footprint evidence, operational trade-offs, and typical fit. section 9.3.3

How Designing AI Agents covers it

How deep it goes
Distinguishes filesystem, network, and runtime boundaries, but leaves physical enforcement to the supplied sandbox runtime.
Examples
Compares Docker/runc, gVisor, Firecracker, bubblewrap/Seatbelt, and WebAssembly, and describes parallel coding with Git worktrees.
What you can do after
Evaluate isolation choices against threat models, resource reach, compatibility, startup cost, and shared-resource conflicts.
How it connects
Connects isolation with distributed systems, parallel agents, sandboxing, and threat modeling.

Also explained in

Compare how 15 books explain isolation →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Designing AI Agents. The section links point to the parts of the book it draws on.