---
title: "Agent Work Needs Runtime Boundaries"
category: "claims"
status: "seeded"
confidence: "high"
sourceLabels: ["Harness synthesis", "Official schedule", "Question layer", "Topic synthesis"]
---

# Agent Work Needs Runtime Boundaries

## Claim
World's Fair 2026 evidence supports the claim that useful agentic work needs explicit runtime boundaries around tools, credentials, filesystem access, network reach, and review paths.

## Why It Is Supported
The security and sandboxing threads repeatedly connect agent usefulness to containment, provenance, and least-privilege execution rather than only to model quality.

## Source Evidence
- [[agent-security]] - Topic synthesis
- [[ai-sandboxes]] - Topic synthesis
- [[sandboxed-agent-execution]] - Harness synthesis
- [[what-security-boundaries-should-agents-have]] - Question layer
- [[2026-06-29-steve-yegge-agentic-security-permissions-provenance-and-the-agent-supply-chain]] - Official schedule
- [[2026-06-30-robert-brennan-sandboxes-aren-t-optional-runtime-isolation-patterns-for-coding-agents-at-scale]] - Official schedule
- [[2026-06-30-samuel-colvin-your-agent-needs-a-sandbox-not-a-desert]] - Official schedule

## Confidence
high

## Evidence Boundary
This is a synthesis claim backed by linked local pages. It is not a direct quote from a single speaker, and implementation details should be checked against the cited source layer before reuse.