Skill 06 · Automation & AI Agents
Work that runs itself.
Automation is software doing the repetitive work: on a schedule, when something happens, or when you ask in plain words. Here is every way it's done, from a ten-line script to an AI agent, shown with things that actually run in your browser.
Running right now
This page is running an automation.
A real one, in your browser: it listens to every change made across Wikipedia and its sister sites, keeps the article edits, and alerts on new pages and big changes. Start it, then change the rule while it runs.
When
anything changes on any Wikimedia site
0
If
it's an article edit on a Wikipedia in any language
0
And
it's a new page, or ≥ 3,000 bytes changed
0
Then
post an alert below
0 alerts
Press start: alerts appear here as real changes match the rule, usually within seconds. The languages of Indonesia are much quieter, so expect long waits with that filter.
Real changes from Wikimedia's public EventStreams feed, read directly by your browser. Page titles, usernames and comments are dropped on arrival; nothing is stored. Pauses itself after 90 seconds.
01 · One chore, six ways
The same morning chore, automated six ways.
Every morning: get yesterday's orders from a supplier portal, add them up per store, update the sales sheet and tell the team. The four steps keep their colour, so you can watch who does the work change.
You remember, open the portal and log in.
Swipe the drawing sideways to see all of it.
The chore, before
By hand
The same clicks every single morning, done by a person who has better things to do.
- Starts when
- you remember to
- Reaches systems through
- the screen
- Who can build it
- anyone
- Breaks when
- you're busy, ill, or make a typo
Often done with
02 · Rules or reasoning
Where rules end and judgement begins.
Most automation follows fixed rules, and that's a strength. AI earns its place only where a task needs reading, writing or judgement. Slide to see what changes.
Fixed rules
Use plain automation: it's exact, cheap to run and easy to test. Most automation lives here.
Fits here: RPA bot, Pipeline, No-code workflow
- Same input, same output
- always
- Cost per run
- next to nothing
- When it goes wrong
- loudly, with an error
- How you test it
- ordinary tests
0 good calls of 0 answered
Pick the kind of automation you'd use for each task.
01Back up the database every night at 2 am.
02When someone fills in the website form, add them to a sheet and send a welcome email.
03Type 300 invoices into an old desktop accounting app that has no API.
04Read customer emails, work out what each one wants and draft replies for a person to check.
05Every night, pull sales from five systems, clean them and rebuild the dashboard tables.
06Sort incoming support tickets by topic and urgency.
07Run the tests and publish the website every time code is pushed.
08Download a report from a supplier's website every morning. It has no API.
03 · Worth it?
Is it worth automating?
Building takes time, and so does keeping it running. Put in how often a task happens and how long it takes, and see when the automation pays for itself.
How often is it done?
How long does it take by hand?
Upkeep is the part people forget: logins expire, a website changes, an API is renamed.
By hand, per month
2.5 h
Pays for itself after
10.4 months
Saved over 5 years
76.2 h
Max worth building (5 yrs)
6.3 days
| how often ↓ · time saved → | 1 sec | 5 sec | 30 sec | 1 min | 5 min | 30 min | 1 hour | 6 hours | 1 day |
|---|---|---|---|---|---|---|---|---|---|
| 50× a day | |||||||||
| 5× a day | |||||||||
| Daily | |||||||||
| Weekly | |||||||||
| Monthly | |||||||||
| Yearly |
Your numbers, simple arithmetic: time by hand × how often, against build time plus monthly upkeep. The grid idea comes from xkcd 1205, “Is It Worth the Time?”; the values are computed here. It counts hours only: automation also removes typos, runs while you sleep and doesn't forget.
Explore
Go deeper.
Methods
Every kind of automation and its building blocks: triggers, schedules, webhooks, queues, retries, approvals, each animated.
Tools
No-code platforms, RPA, browser bots, schedulers, orchestrators, CI/CD, messaging APIs, models and agent frameworks, side by side.
Workflow lab
Build a workflow from nodes and run it on real public data, with retries, branches and a human approval step.
AI agents
How an agent works inside: the loop message by message, its parts, twelve patterns, autonomy levels, a guardrails lab and costs with live model prices.
Build an agent
The actual API calls, tool schemas and MCP, step by step, with the request and response at every turn.
Prompt → automation
Describe a task in plain words and watch an AI turn it into a workflow, then run it.