WEBVTT

00:00:00.900 --> 00:00:04.372
In a few minutes, Claude Code will
be working in your project folder,

00:00:04.503 --> 00:00:06.488
and you'll know how
to keep it in check.

00:00:09.850 --> 00:00:15.763
AI for Research Efficiency.
Episode 2.1: Install Claude Code.

00:00:17.000 --> 00:00:22.776
Last time: a terminal, a project folder,
Git and Python, and an account.

00:00:22.980 --> 00:00:28.975
In this episode: install Claude Code,
sign in, run it in your project folder,

00:00:29.166 --> 00:00:31.691
and check each step
before you speed up.

00:00:34.100 --> 00:00:37.242
Step one: in PowerShell,
paste this line:

00:00:37.692 --> 00:00:45.716
irm https://claude.ai/install.ps1 | iex

00:00:49.116 --> 00:00:52.041
On a Mac, use the second line,
in Terminal.

00:00:52.432 --> 00:00:54.515
Both lines are on the companion page.

00:00:54.682 --> 00:00:56.842
You don't need administrator rights.

00:01:00.500 --> 00:01:03.945
Step two: close the terminal,
open a new one,

00:01:04.156 --> 00:01:06.950
and type claude --version.

00:01:07.541 --> 00:01:09.223
You should see a version number.

00:01:12.428 --> 00:01:15.052
If PowerShell says
claude is not recognized,

00:01:15.323 --> 00:01:18.027
open another PowerShell window,
and try again.

00:01:22.100 --> 00:01:24.965
Step three:
go into your project folder with

00:01:25.076 --> 00:01:28.722
cd agent-work, and type claude.

00:01:30.902 --> 00:01:34.671
Always start it in a project folder,
never in your home folder.

00:01:36.500 --> 00:01:40.648
Step four: the first time,
it asks you to pick a text style,

00:01:40.939 --> 00:01:46.931
then how to sign in: a Claude plan, or an
Anthropic Console account for an API key.

00:01:48.282 --> 00:01:53.386
Your browser opens. Sign in there,
and the terminal says: login successful.

00:01:55.700 --> 00:01:59.510
Step five: it reminds you
that Claude can make mistakes,

00:01:59.661 --> 00:02:02.568
and that you're responsible
for reviewing what it does.

00:02:03.356 --> 00:02:05.544
Then it asks whether
you trust this folder.

00:02:05.916 --> 00:02:08.625
It's your own project folder,
so: yes.

00:02:12.500 --> 00:02:16.102
Step six:
Claude Code now starts in auto mode,

00:02:16.232 --> 00:02:18.904
where a second model
approves most of its actions.

00:02:19.434 --> 00:02:23.596
While you learn, press Shift
and Tab once, for manual mode.

00:02:23.760 --> 00:02:27.550
In manual mode, it stops
and asks you before it edits a file,

00:02:27.681 --> 00:02:30.106
runs a command,
or reaches the network.

00:02:34.100 --> 00:02:37.832
Step seven: ask it something.
What's in this folder?

00:02:40.635 --> 00:02:44.057
It reads the notes and the data,
and answers in plain words.

00:02:44.447 --> 00:02:47.119
To open the Word drafts,
it needs to run a converter:

00:02:47.389 --> 00:02:50.110
that's a command,
so it asks you first.

00:02:55.700 --> 00:02:58.246
Step eight:
when you trust it with a task,

00:02:58.397 --> 00:03:02.056
press Shift and Tab until the
status line says auto mode.

00:03:02.447 --> 00:03:05.234
It works without stopping,
and much faster.

00:03:05.434 --> 00:03:09.400
Anthropic's own advice:
auto mode reduces permission prompts,

00:03:09.550 --> 00:03:11.473
but it doesn't guarantee safety.

00:03:14.900 --> 00:03:20.221
Pause now: install Claude Code, open
a new terminal, and check the version.

00:03:24.500 --> 00:03:29.932
So: one line to install, a new terminal,
your project folder, sign in,

00:03:30.244 --> 00:03:35.114
trust the folder, manual mode first,
then auto mode when you trust it.

00:03:36.300 --> 00:03:38.166
Next: a worked example.

00:03:38.598 --> 00:03:42.971
Agents read a whole literature folder
in parallel, and check every claim.