Getting started & examples

Real examples from a Swiss workday

Inbox triage and replies

Before: 40 emails, two hours of sorting, drafting and filing. After: the agent drafts replies in your voice, flags what needs your eyes, and files the rest — you review in 20 minutes.

Meetings into next steps

Before: notes sit in a notebook and die. After: the agent turns notes into a summary, assigns next steps and sets calendar reminders — no copy-paste.

Client follow-ups

Before: "I should follow up with that prospect" is forgotten by Friday. After: the agent tracks touchpoints, prepares the follow-up and reminds you at the right moment.

Background & research

Before: an hour of tabs before a call. After: the agent gathers context, builds a one-page brief and drops it in your channel.

A sensible starting path

1. Pick one task you do weekly that's repetitive and low-risk (e.g. summarising a shared inbox).
2. Choose an agent — OpenClaw if you live in chat apps, Hermes if you want the learning loop.
3. Install it on your machine or a small server, and run onboarding.
4. Connect one channel and the minimum tools the task needs.
5. Set approvals so anything sensitive waits for you.
6. Run it for a week, review the logs, then expand one step at a time.

What you'll need

• A machine to run it on — a laptop or small office computer is fine; a server works too.
• A model provider or API key (Claude, GPT, DeepSeek…) — or a local model if you prefer.
• The channels you use daily — WhatsApp, Telegram, Signal or email.