Agent Harnesses: Operating and Governing Enterprise AI Agents

Editorial placeholder -- not yet a full draft.

This article will adapt the internal Workbench Handbook article on agent harnesses (docs/workbench-handbook-agent-harnesses.md) for a public audience.

Planned scope:

• Distinguish model, agent, workflow, and agent harness as separate concepts.
• Explain the core harness responsibilities: identity and access, prompt/configuration versioning, context and memory selection, model routing, and tool guardrails.
• Emphasise that authorisation must be enforced when data is retrieved or a tool acts, not merely described in a prompt.
• Adapt the framing for an external, vendor-neutral audience and avoid exposing internal implementation details or naming internal system components directly.

An administrator or curator should turn this brief into a full draft before submitting it for review.