Inbox triage that routes, drafts, and waits for your OK
An agent sorts incoming email into a few clear buckets, drafts replies for the routine ones, and leaves anything sensitive for a human.
Why it works
Triage is a textbook fit for what Anthropic calls the routing pattern: classify an input, then send it down a specialised path. The same guide recommends starting with the simplest workflow that solves the problem before reaching for a fully autonomous agent.
OpenAI's practical guide to building agents also lists workflows that involve complex decisions or messy unstructured data, like email, as good early candidates.
How to set it up
Define 4–6 buckets (for example: customer request, invoice, sales lead, newsletter, needs-owner). Give the agent read access plus permission to create drafts only, so nothing sends on its own.
Keep a human approval step for anything involving money, contracts or complaints. Check a sample of its sorting every day for the first two weeks, then weekly.
Watch out for
Over-broad permissions. Start read-only, and never give "send" rights in the pilot. Ready-made triage prompts exist in the open-source botdirectory.ai catalog; treat them as starting points and test them before you rely on them.
Sources
- Anthropic: Building effective agents
- OpenAI: A practical guide to building agents (PDF)
- botdirectory.ai (MIT-licensed prompt catalog)
Figures from vendors are labeled as vendor-reported. We have no commercial relationship with any company linked here.