Rippling splits AI diagnosis from deterministic authorization in its IT Helpdesk agent
andreisavu · x · 2026-10-10
Rippling's engineering blog explains the core design of its Helpdesk IT agent: separate AI interpretation from authorization. The agent diagnoses ambiguous requests like "my laptop is slow" by inspecting device age, memory patterns, and running apps, but fulfillment actions (approving and shipping a replacement) run through deterministic, IT-defined policies rather than model decisions—avoiding dangerous writes with no durable policy to point to. The post contrasts this with traditional workflow tools' rigid forms and with end-to-end agents' lack of accountability, and covers learning from recurring device issues.
More from coding & agent
- Every launches Slack-based shared company agent, passing token costs through with no markup — every · 2026-10-10
- Security researcher: open-weight models beat closed ones for blackbox bug bounty workflows — rez0__ · 2026-10-10
- OpenClaw's Cognitive Engine: Postgres-backed memory and RAG across 500+ gateways — heyneighbor · 2026-10-10
- Celesto MCP ships: share one cloud computer across Claude Code, Codex and more — aniketmaurya · 2026-10-10
- Agent Arena swaps preference voting for causal tracing to benchmark agents — arena · 2026-10-10
- Matt Pocock's skills repo out-stars React; animated video breaks down his workflow — mattpocockuk · 2026-10-10