Full Tutorial: Build Your Personal OS with Claude Code in 50 Min | Teresa Torres

Peter Yang

44.9K views • December 21, 2025

Teresa is a well-known product coach and author who runs her entire life and business using two Claude Code terminals and a note taking app. It was genuinely mind blowing to watch her use Claude Code to manage her to do list, delegate her research, and write content.

Teresa and I talked about:

(00:00) Why Teresa uses Claude Code to run her life
(03:00) Live demo: The "today" command that builds Teresa's daily to do list
(07:00) Working with Claude Code to brainstorm tasks and ideas
(14:00) Live demo: Writing a blog post together using plan mode
(21:00) How Claude does SEO keyword research for blog posts
(26:00) Why Teresa still writes every word herself (but 10x faster)
(30:00) Pro tips to manage context so Claude doesn't "get dumber"
(36:00) The 3-layer context system: global, project, and reference files
(44:00) 3 tips to get started with Claude Code without feeling overwhelmed

Thanks to our sponsors:
Optimizely: The agent orchestration platform for marketing https://www.optimizely.com/ai/

Get the takeaways: https://creatoreconomy.so/p/build-your-personal-os-with-claude-code-teresa-torres

Where to find Teresa:
LInkedIn: https://www.linkedin.com/in/teresatorres/
Website: https://www.producttalk.org/

📌 Subscribe to this channel – more interviews coming soon!

Login or create an account to generate AI summaries

🔑 Login 👤 Sign Up

0:00 At the beginning of my day, I literally

0:02 just write today and we can see what it

0:04 does. It generates my to-do list for

0:06 today. It's going to go to Trello and

0:08 see if anybody on my team has added any

0:11 new Trello cards and then it basically

0:13 runs this Python script and then you can

0:15 see it's going to read my Today MD and

0:17 it's going to update it. And you can see

0:18 it popped up here. So, this is a link to

0:20 my research digest, my blog post. It's

0:22 9,000 words and I wrote it in 1 and 1/2

0:24 days. There is no way I would have done

0:27 this myself. That is insane to me.

0:29 here's how I'm thinking about it. What

0:30 do you think? Oh, I don't really like

0:31 that. What if we try this other way? And

0:33 then I do that throughout the whole

0:34 construction of the blog post. You can

0:36 just be like, "Enter me about my

0:37 business." And then it will ask you a

0:38 bunch of questions and then it will

0:40 write a context file for you. This file

0:42 we're looking at, this profile, I never

0:43 wrote a word of it. Every time I have

0:45 Claude add a context file, I say, "What

0:47 index needs to be [music] updated?" And

0:49 it just figures it out.

0:50 >> Let's say I'm just overwhelmed by all

0:51 this stuff. What are your three tips to

0:52 get started? Whenever you find yourself

0:54 [music]

0:54 explaining context to Claude, stop and

0:57 think about, am I ever going to have to

0:58 explain this context [music] to Claude

1:00 again?

1:01 >> Okay, welcome everyone. Uh my guest

1:03 today is Terresa Torres. Uh Teresa is a

1:05 legend in the PM field. She's author of

1:08 continuous discovery habits, but today

1:10 we're going to talk about how Teresa

1:11 uses cloud code for everything from

1:14 writing to editing to task management to

1:17 coding. uh really excited to get her to

1:19 give us a tour of her cloud code

1:21 projects and to show us how it's

1:22 actually done. So, welcome Teresa.

1:24 >> Thanks for having me. I'm excited to do

1:26 this.

1:27 >> Yeah. Um so, you know, I've also used

1:29 cloud code quite quite a bit, but I I'm

1:31 super interested in seeing your uh

1:32 highle workflow. So, maybe you can give

1:34 us a quick tour first.

1:36 >> Yeah, let me share my screen. Okay, so

1:38 what we're looking at is really simple.

1:40 I have Obsidian here on the left and

1:42 then and if people aren't familiar with

1:43 Obsidian, it's just a note-taking tool

1:45 that's based on Markdown. And then I

1:47 have two terminal windows on the right.

1:49 And the big thing that I use Claude Code

1:52 for like I started using it coding just

1:54 like everybody else most that's most

1:56 people's entry into it. Um but I what I

1:59 really loved about Claude Code with

2:01 coding is it's almost like you're pair

2:03 programming with Claude. And I was like

2:05 wow how do I pair with Claude on

2:07 everything that I do? And so I started

2:10 doing this for writing and for strategy

2:12 and for literally all of the work that I

2:14 do. This is my setup. This is my work

2:17 obsidian vault. So I have an obsidian

2:19 vault called work. In it there's a bunch

2:21 of subdirectories. One of them is LLM

2:24 context which I'll get into that. And

2:26 then my notes which is just any notes

2:28 that I have. I have a research project

2:29 that we can talk through. I have my

2:31 tasks system. Worthy reads are like

2:34 articles that I've saved that other

2:35 people wrote that I really like. And

2:37 then my writing directory. And then you

2:39 can see I have a cloud MD file here. Um,

2:42 and we can kind of go through all of

2:43 this, but I'm going to start with tasks.

2:45 So, the way my task system works is I

2:48 have folders for different things. So, I

2:50 have a folder for bugs. I have a folder

2:52 for ideas. I have a folder for tasks.

2:55 And every day, I launch CL Claude inside

2:58 my tasks folder. So, you can see this

3:00 top terminal window. I'm in tasks. I'm

3:03 just going to launch Claude. And at the

3:05 beginning of my day, I literally just

3:07 write today. And we can see what it

3:09 does. The first thing it's going to do

3:11 is it's going to go to Trello and see if

3:14 anybody on my team has added any new

3:17 Trello cards to my board. You can see,

3:19 good news, there's no new Trello cards.

3:21 I've already run this today, but we'll

3:22 just see what it looks like. And then it

3:24 basically runs this Python script where

3:26 it generates my to-do list for today.

3:28 So, that's what we're looking at here on

3:30 in Obsidian. It says Monday, November

3:32 10th. I have one overdue item. I have a

3:34 number of items that are due today. And

3:36 then I have my in progress ideas that

3:38 I'm working on. And uh this script just

3:42 generated this file. Every task is a

3:44 markdown file. So you can see in my

3:47 tasks folder, I just have a whole bunch

3:48 of tasks. Basically my this task that's

3:51 overdue was due on Friday. You can see

3:53 it's just I have to call the grooming

3:55 place for my dog and schedule an

3:57 appointment for her and I didn't do that

3:58 yet. Um and then all my other tasks like

4:01 I have calls with a couple people today.

4:04 Um, I'm writing a Claude code safety

4:06 article. I've got to add some images to

4:08 it. All this stuff just shows up on my

4:10 to-do list every day based on what I put

4:12 in my tasks.

4:14 >> Got it. So, the other tasks like the

4:15 calls and stuff, is it coming from your

4:16 calendar or is it coming from just your

4:18 notes?

4:18 >> It's tasks that I've created. And I'll

4:21 show you. We're going to create a new

4:22 task together. I'll show you what that

4:23 looks like.

4:24 >> Got it.

4:25 >> Um, so that's the first thing is when I

4:27 run the today command, I do it every

4:28 morning. It's basically looking through

4:30 my tasks folder and looking for anything

4:33 that's due today or anything that's

4:34 overdue. And then it creates this today

4:37 file for me and it's just telling me

4:39 what do I need to do today.

4:41 >> And then I distinguish between tasks

4:43 that have a due date and ideas that you

4:46 know they're like more like projects

4:48 that don't have a specific due date, but

4:50 they're things that I have going that

4:52 are ongoing. So you can see like I have

4:55 a podcast just now possible that's like

4:57 always in my in progress because it's

4:59 reminding me to like make sure I

5:01 schedule guests and um you can see here

5:04 I have a single episode. This is my

5:07 project for like getting that episode

5:09 ready to be released. Um and then I have

5:11 a task for social media, but this is

5:15 really just my like remember to schedule

5:18 guests kind of thing that's always in

5:20 progress. Um, and so what this does is

5:24 it gives me a lot of flexibility. If I

5:25 just have a random idea, I can just say

5:27 it a cla new idea and it will tell it

5:30 what it is. It will create a markdown

5:31 file. It'll put it in my ideas folder.

5:34 >> Um, if it's a task, I can give it a due

5:37 date and then it will show up on my on

5:39 this like to-do list. This is just

5:42 called my today MD on the day that it's

5:44 due.

5:45 >> Should we try making uh maybe like uh

5:48 record pockets of Peter or something

5:49 like some sort of task?

5:50 >> Yeah. So, we'll do a new task in just a

5:52 second here. The other thing that's

5:53 happening when I run this today command,

5:56 it is I've built this research system.

5:58 So, I'm trying to keep up to date on

6:00 like academic research that's related to

6:03 my um work. And so, I built this system

6:06 that searches a preprint server and then

6:08 Google Scholar every day and it gives me

6:11 a research report of what's relevant to

6:14 my work and that gets added to this

6:17 today list. So you can see right here

6:18 it's not here yet and it's because we

6:20 haven't done that part. So it's running

6:22 this slash command that's part of my

6:24 research project and now it's asking me

6:26 oh it wants to go see if there's a

6:28 research queue. So we're going to say

6:29 yes to that.

6:30 >> And then you'll see it's going to update

6:32 this today list with any research for me

6:34 to review. The Q's empty.

6:38 It's creating my research today MD.

6:42 >> Got it. So this is just like uh

6:44 searching some websites to find new

6:46 research, right? Yeah. Yep. And then you

6:48 can see it's going to read my today MD

6:50 and it's going to update it and you can

6:51 see it popped up here. So this is a link

6:53 to my [snorts] research digest. So that

6:56 that's my today command and you can see

6:57 it gives me a summary like this is what

6:59 I just did. I created your today MD with

7:02 an overdue task 10 in tasks for in

7:05 progress your research whatever. And

7:07 then you can see it's also creating a

7:09 this week view and a next week view.

7:11 >> Got it. Okay. Yeah.

7:12 >> Okay. So let's say I want to do a new

7:13 task. We're actually going to write a

7:14 blog post together. So, I'm going to do

7:17 new task. Um,

7:19 write plan auto except Claude

7:25 blog post. Do today draft

7:31 outline with Claude. And it basically

7:34 this this tasks folder is set up with a

7:36 cloud MD so that Claude knows what to do

7:39 with this. It knows to create a task.

7:41 You can see it's using Obsidian front

7:44 matter and I've defined this whole

7:46 system for Claude so it knows to do

7:47 this. It's defining as a task. It's

7:49 putting the due date as today. It's

7:51 tagging it for me. It's give basically

7:53 giving me kind of a checklist. I It's

7:56 actually not supposed to do this because

7:57 it's kind of a Chromy checklist. Um it's

7:59 supposed to just take my notes, but

8:00 pretty good. We're going to go with

8:02 that. And then let's see if sometimes

8:04 Claude is smart enough to put something

8:05 due today on my today MD. It's not this

8:08 time. So I'm going to say add to today

8:10 MD.

8:11 >> Got it. Okay.

8:12 >> And then what that's going to do is it's

8:13 just going to add that task. You can see

8:14 it's here. So now I have my write plan

8:17 autoac accept mode. What I love about

8:20 this is I work out of the terminal most

8:22 of the day and I keep this tasks

8:24 terminal open all the time. So if like I

8:26 think of something that I have to do or

8:28 I have a random idea, I literally can

8:30 just be like new idea blah blah blah

8:32 blah and it's done. And then I can just

8:34 go back to my work in my other window.

8:37 And it's it's [snorts] a lot faster than

8:39 like opening a web browser, going to

8:41 Trello, creating a new card, setting a

8:43 due date, right? It's just it's super

8:45 fast and that's what I love about it.

8:46 >> Got it. This episode is brought to you

8:48 by Optimize. The problem in marketing

8:51 usually isn't a lack of ideas, it's a

8:53 lack of time. If your to-do list keeps

8:55 getting away from you, then take a

8:56 [music] look at Optimize the Oppo, an AI

8:58 agent platform built visibly for

9:00 marketing. With OPPO, you can use AI

9:02 agents for SEO and geo recommendations,

9:04 AB testing, website analysis, and much

9:06 more. OPPO knows your brand inside and

9:08 out and plugs into your existing tools

9:10 and data systems so that you can save

9:12 [music] time on laborintensive feature

9:14 testing, reporting, and more. See what

9:16 it can take off your plate at

9:17 optimiz.com/ai.

9:20 Now, back to our episode.

9:21 >> The other thing I'll share is it allows

9:23 me to do things like, uh, what are all

9:26 my marketing ideas?

9:29 and it's just going to go search my

9:30 ideas folder for anything related to

9:32 marketing. And so like whenever I'm

9:34 working on a new project, I can just ask

9:36 Claude like what are my tasks related to

9:39 this or what am I what are my idea open

9:41 ideas that are related to this and it

9:43 just goes out and finds them and like

9:45 Trello has search but it's not the best

9:48 frankly.

9:49 >> Okay.

9:50 >> Um so I like this a lot too.

9:52 >> So you have like ideas and memories and

9:53 like random re research snippets that

9:55 you pull.

9:56 >> Yeah. So you can see

9:58 >> like in my tax folder, I have bugs that

10:00 have been filed. I have ideas. Import is

10:03 for if I'm importing a bunch of stuff

10:05 from Trello. I want to review it to make

10:07 sure it imported correctly. My inbox is

10:09 how I get stuff from my phone to my to

10:12 this whole system. Memories is like

10:14 things. It's like little snippets that I

10:16 want to remember. So we can kind of look

10:18 at what's in here. I think this is safe.

10:20 Like I have a affiliate link to the AI

10:22 evals course that I always forget what

10:23 it is. So, it's literally just here's my

10:25 affiliate link. This is a task. I don't

10:27 know actually know why that's in

10:28 memories.

10:30 >> Real real real quick. What do you mean

10:31 by inbox? Like you send an email to

10:33 yourself over the phone or or how does

10:35 it

10:35 >> Yeah. Uh so this this is a little bit of

10:38 a hack and I'm actually moving away from

10:40 it. So I mentioned I moved a bunch of

10:41 stuff and it broke some things. But um I

10:44 was using Dropbox to sync my vaults and

10:48 I'm now moving to Obsidian Sync to sync

10:50 my vaults. So, the Obsidian Sync, I can

10:52 use the iOS Obsidian app and I can just

10:54 access all of this stuff from my phone.

10:56 But when I was using Dropbox, I had to

10:59 set up this is like a iCloud linked

11:02 folder

11:03 >> so that I could add to it from my phone

11:04 and then it would automatically show up

11:06 here and then I could pull it into my

11:08 task system. That was kind of a pain in

11:10 the butt. So, I moved from Dropbox to

11:12 Obsidian Sync to make that easier.

11:13 >> Got it. Okay. So, basically Obsidian and

11:15 Cloud Code run your life right now.

11:17 Yeah, that makes sense. Yeah, pretty

11:18 much cloud code and obsidian read my

11:20 life.

11:20 >> All right, so let's go back to the

11:22 writing the blog post task. So like um

11:24 >> yeah,

11:25 >> how's it going to help us write this

11:26 blog post? Yeah.

11:27 >> So I could do this. Let's do this. We're

11:30 going to get rid of this dumb checklist

11:33 that Claude made because I don't like

11:34 it.

11:36 And we're just going to do like Okay,

11:38 I'm going to talk through I'm just going

11:39 to like talk out loud while I plan out

11:40 my blog post. So for context, I'm in the

11:43 middle of writing this huge cloud code

11:44 series for people that are

11:46 non-technical. And it started with like

11:48 what is cloud code? Why is it different

11:50 from cloud in the browser? And then the

11:52 second post in the series was about like

11:54 how to give claim memory, which if you

11:56 want to get into that, all my context

11:58 files are also here in Obsidian. And

12:00 then the third blog post that's coming

12:02 out this Wednesday is about safety. So

12:04 how to use cloud code safely. And then

12:06 this one we're going to start together

12:07 is the fourth article. And this is where

12:09 I'm going to get into like doing

12:11 projects with Claude. And so I want to

12:14 write a blog post where I talk about

12:15 like plan mode, auto accept mode. And

12:19 then I always want to use examples in my

12:22 blog post that are related to product

12:25 managers. And so like in this blog post

12:28 I might think like let's talk through

12:30 plan mode and auto accept mode if you

12:32 were planning a new feature. So I'm

12:34 starting to think about like what's my

12:36 example going to be in the blog post?

12:38 And I have an AI product called the

12:40 interview coach. This is a product where

12:43 in my class where I teach how to conduct

12:45 effective customer interviews. We do

12:47 practice interviews with each other in

12:48 class and then the students can submit

12:50 their transcripts and get like detailed

12:52 feedback on how good of a job they did.

12:54 And the thing with this product is it

12:56 was built for my storybased interview

12:58 coach or storybased interview class. And

13:01 so the coach gives feedback on

13:03 story-based interviews. And my new

13:06 feature idea is can I extend it to work

13:10 with non-storybased interviews. So this

13:13 is like a feature idea I have. And so in

13:15 the blog post there's a little meta. I'm

13:17 going to use this feature idea to show

13:20 how I would use plan mode and auto

13:23 accept mode. Right. So okay I'm gonna

13:26 this is all I have. This is literally

13:28 the state of this blog post. This is

13:29 real. I have not written a single thing.

13:31 And now I'm going to go over to this

13:33 terminal window on the bottom and I'm

13:34 going to launch Claude. And this is in

13:36 the context of my writing vault. And so

13:39 my writing vault has a claude MD that

13:41 talks about how I like to write and how

13:42 I like to write with Claude. So it has

13:44 different context than this tasks

13:47 window. So like in my writing window, I

13:50 can't type new task. It doesn't know

13:51 anything about my task system. I have to

13:53 do that in my tasks window.

13:55 >> Got it?

13:56 >> So we're going to do this. I don't know

13:58 why it thinks it's a new folder, but

13:59 we're going to ignore that for a minute.

14:01 And then I'm going to say I want help

14:03 creating an outline for a new blog post.

14:06 I do this in plan mode, which is

14:09 actually what the blog post is about.

14:11 You can find my early thoughts

14:15 in Oh, see I would not, this is a little

14:18 weird. I would not do this in the task

14:20 card. What I would do is I would come to

14:22 writing. I would go to my claude code

14:25 series and I would do a new file.

14:28 >> Got it.

14:29 >> Let's do plan mode.

14:33 And that did not go where I wanted it.

14:34 So, we're going to move it in plan mode.

14:37 And I actually did not tell Cloud enough

14:39 information because it's not it's going

14:41 to look for plan mode in the current

14:42 directory, but the plan mode is actually

14:45 in claude code. So, we'll see what it

14:47 comes back with. We'll see if it finds

14:48 it.

14:49 >> Okay. So, basically, you have a CL.

14:51 Okay. So ju just real quick, you have a

14:53 claw.md for your writing vault that has

14:57 a bunch of writing styles and like uh

14:59 you know that kind of stuff, right?

15:00 >> Yep.

15:00 >> Is that got it?

15:02 >> Yeah. So if you look at here on the

15:04 left, like tasks is a pro that you can

15:07 think about tasks is like a claude

15:08 project. I launched this instance of

15:11 claude in the context of tasks

15:13 >> and then I have a writing folder and I

15:15 launched this instance of claude in the

15:17 context of the writing folder. So the

15:20 writing folder has different rules than

15:22 the tasks folder. They each have their

15:24 own cloud MDs. So the cloud MD for tasks

15:27 explains how the task system works. It

15:29 explains how tagging works. It explains

15:30 how front matter in Obsidian works. So

15:32 that when I just type new task, it knows

15:34 exactly what to do with that.

15:36 >> Whereas my writing folder doesn't have

15:38 any of that context. It just knows it.

15:40 Like you can see here, I told it I want

15:42 to write a blog post. It found plan

15:44 mode. It read it. Now it's asking me to

15:46 read my style guide. And that's because

15:48 my cloud MD in my writing folder says

15:50 before we do anything together, start by

15:52 reading my writing style guide.

15:54 >> Got it. Okay.

15:55 >> So, I'm going to go ahead and say yes,

15:56 it can do that.

15:57 >> Do you ever just ask it to like uh skip

15:59 all the permissions because it's kind of

16:00 annoying to

16:00 >> Yeah. You know what I have? So, I just

16:03 moved everything. Everything used to be

16:05 in Dropbox and I had all the permissions

16:08 set up where my tasks could read and

16:10 edit everything inside tasks and writing

16:12 could read and everything inside

16:14 writing. But because I just moved

16:16 everything, those permissions aren't I

16:18 literally moved everything yesterday.

16:20 Uh, and so I haven't copied over those

16:22 permissions yet. I'm learning like when

16:24 you move things in claude, it kind of

16:26 breaks things a little bit. Uh, so it's

16:28 asking me which features do I want to

16:29 focus on? I actually want to do both.

16:32 Um, what's your intended audience for

16:35 this post? Product people new to Claude

16:37 should already know that. That's a

16:38 little disappointing.

16:39 >> Wait, so so like uh is asking you

16:40 questions because you prompted to ask

16:42 your questions? ask me questions because

16:44 it read this plan mode document and it's

16:47 now like trying to get more information

16:49 from me to be helpful. Normally I have a

16:52 little bit more here, right? This isn't

16:54 really enough for Claude to work with

16:56 yet, but let's let's see what Claude

16:58 does.

17:00 >> Yeah, see it's trying to do too much

17:01 right out of the gate, but let's see

17:03 what it does now. I understand the

17:05 direction. Let me create an outline.

17:12 F. It's already got a headline.

17:17 Introduce plan mode and auto accept

17:19 mode. What it is, when to use it, what

17:22 you plan, what it is, the dangers.

17:26 [laughter]

17:27 >> That's great. Yeah,

17:29 >> this is kind of nuts, right? Like it

17:30 kind of went crazy. Um, so I'm going to

17:35 tell it went crazy.

17:36 >> Uh, it's kind of like, uh, when I use

17:37 cloud code to like write a product spec,

17:39 I always wanted to put put the spec in

17:41 MD file so I can delete a bunch of stuff

17:43 because it it tends to do too much, you

17:45 know.

17:46 >> Yeah. So, usually, to be honest, usually

17:48 I have a little bit more. So, let's

17:50 let's actually play with this draft a

17:52 little bit. Um, I would probably give

17:54 Claude something like goals help people

17:58 new to Claude code do bigger projects.

18:04 Claude code uh use

18:07 planning a new feature as a product

18:10 manager as an example throughout. And

18:14 then we could do I can use my personal

18:17 experience with

18:20 expanding the interview coach. I'm going

18:22 to ignore typos for now because Cloud

18:24 won't care.

18:25 >> Um, and then, um, I'm going to do

18:28 something like less about the features,

18:32 these features,

18:35 more about how to get

18:38 Claude

18:40 to do good work. And let's see if this

18:42 helps.

18:46 >> Got it.

18:47 >> Let's see what it comes back with.

18:49 >> So, okay. So after this outline, you're

18:51 going to ask her to do online research

18:53 or what's the I can do next?

18:56 >> Yeah. So you can see already it came

18:58 back and it's like, "Oh, okay. I made

18:59 too many assumptions."

19:01 >> Yeah.

19:01 >> Um and it's like, "Okay, now like now is

19:04 this like the right direction?" So now I

19:06 might now I what I really want to do is

19:08 ask Claude like you asked about

19:10 research. We got to come out of plan

19:12 mode for this. Has anyone else written a

19:15 blog post about this? and I'll go and

19:18 I'll go search and see what other blog

19:19 posts are out there. So, I'll often do

19:22 this. I'll often like before I sit down

19:23 to write, I want to know what else has

19:25 been out there. If somebody else has

19:27 written this article for my audience,

19:29 I'm not going to write it. I'm just

19:31 going to like write a different article

19:32 and just link to that article. Um, I

19:34 want to make sure that I'm writing stuff

19:36 that's unique.

19:37 >> And then now sometimes someone may have

19:39 written this blog post like in the

19:40 context of coding and that's fine. Most

19:42 of my audience isn't going to read that.

19:44 And so, but I can read that article and

19:46 learn from their experience and then

19:48 translate it to product management. Um,

19:50 so I always use Claude Code to like do

19:53 searches for me. The other thing I use

19:55 it for while I'm writing is like I'll

19:57 make a claim and I'll be like, "Is that

19:59 claim true?" And then I'll ask Claude

20:01 like, "Can you do some academic research

20:02 for me and see if there's evidence

20:04 behind this? Like, is this really a true

20:06 claim or is it just something that I

20:07 randomly believe?" So, let's see what

20:09 Claude came back with. So, a lot of

20:11 people have written about this. Of

20:12 course, Anthropic has written about it.

20:14 They have great documentation. Most are

20:16 written for developers, engineers. Yeah.

20:18 Perfect.

20:19 >> I think people wrote like technical

20:20 stuff about plano but not about this use

20:22 case.

20:22 >> Yeah. So then I might ask like okay um

20:26 how do people search for this kind of

20:29 content?

20:31 What keywords from an SEO standpoint

20:35 should I keep in mind? I almost never do

20:37 this to start. I do this at the end, but

20:39 we can do it to see how it does. Oh,

20:41 that's very smart. I I I never asked

20:43 this question for my blog post, so

20:45 that's that's that's important. Yeah.

20:47 >> And again, it just goes off and does its

20:49 thing, and I let it do its thing, and

20:50 the meanwhile, I'll be like over here

20:51 noodling. I'm like,

20:53 >> how am I going to how am I going to

20:54 structure this blog post? And like this

20:57 is really stubbed out. There's not

20:58 really a blog post here, right? So

21:01 >> I might start to think about an outline

21:02 and be like okay well the introduction

21:06 needs to be strong hook around value of

21:11 doing project work with claude code.

21:15 Um then introduce

21:18 like claude modes

21:20 plan auto accept

21:23 um reference

21:25 safety article

21:27 um walkth through detailed example

21:32 interview coach. So let's see what it

21:34 came back with. So I found it's really

21:37 good at keyword research

21:39 >> like really good at keyword research. I

21:42 tend to do this after I've written the

21:43 article because I really want to write

21:44 my articles for humans, but then I might

21:47 tweak some words in my subheaders or my

21:49 titles to to like target higher volume

21:52 keywords.

21:53 >> What what what is it actually doing? Is

21:54 it just using Google searches here or is

21:56 it actually doing some uh fancy stuff to

21:58 >> Yeah, you can see it's just doing

22:00 searches and looking for what's ranking

22:01 well.

22:02 >> Oh, is that what it's doing?

22:03 >> Yeah.

22:03 >> Oh, interesting. Okay.

22:04 >> Yeah.

22:05 >> Okay.

22:05 >> So, it's sort of my SEO researcher.

22:07 >> Interesting. And then I could be like,

22:09 I've started to stub out an outline in

22:14 plan.

22:16 Can you suggest

22:18 alternative structures? There's not

22:21 enough meat here. It's probably not

22:22 going to do a very good job here, but

22:24 >> um we'll see what it does. And so I do

22:27 this. This is literally how I write. I'm

22:29 like doing all my thinking really rough.

22:31 I'm going back and forth with Claude.

22:34 I'll have questions like what do I do?

22:36 like how does this work in Claude? Like

22:38 I think I know how auto accept mode

22:40 works, right? But then I'll be writing a

22:41 blog post about it. I'll be like, is

22:42 that really right? And I literally will

22:45 just pop over here and be like, Claude,

22:46 is that are you allowed to do this in

22:47 auto accept mode? Um, and so that's

22:50 really nice, too. And I know this is a

22:52 little meta. I'm writing a blog post

22:53 about Claude code, but Claude can

22:55 actually answer questions about

22:56 anything, right? So, I could be writing

22:58 a blog post about how product teams do

23:01 customer interview analysis and I can

23:03 still ask Claude, in fact, I wrote a

23:04 blog post about how about customer

23:06 interview analysis with AI. And when I

23:09 was writing that article, I was

23:10 constantly asking Claude to look at

23:12 research for me, like academic research,

23:14 like what do we know about what is lost

23:17 when we rely on AI synthesis? What do we

23:19 know? And a lot of that came from just

23:21 asking Claude to do the research for me.

23:23 >> Got it. So it's very much two screen uh

23:26 kind of process. Yeah, they're

23:28 definitely using as a thought partner

23:29 and like a researcher along the way.

23:31 >> And so you can see here now Claude came

23:33 back with like here's some other ways to

23:34 structure it. So we could do example

23:36 first structure,

23:38 uh problem solution structure,

23:42 journey structure, like your personal

23:43 story. And you can see for each it's

23:45 it's kind of like giving me a rough

23:48 outline. So, what I like about this is

23:50 like when I write, I usually have a

23:52 pretty fixed idea of what the structure

23:54 is in my head. And Claude helps me

23:57 explore alternatives. So, it's like I'm

23:59 comparing and contrasting in the context

24:01 of writing, which before working with

24:03 Claude, I would have literally never

24:05 done that. And then the other thing that

24:07 it really helps me with, and maybe I'll

24:09 pull up my web browser to show this. Um,

24:11 let me think about actually I think I

24:13 can show it right here. Um the article

24:16 that I'm publishing this coming

24:17 Wednesday is on safety. So it's how do I

24:20 teach beginners cloud code beginners how

24:22 to run an LLM safely on their computer.

24:26 And you can see I'll show you every it's

24:29 organized by tier. So like what's the

24:30 risk of letting it read files and then

24:33 at the end of every section I include

24:35 what cla is doing and how you know what

24:38 it's doing. So like there's these tables

24:39 of like this just indicates if it's

24:41 reading. There's like cuz like I'm

24:43 assuming most of my readers don't know

24:44 Unix commands, right? So

24:46 >> yeah, probably not using terminal. Yeah.

24:48 >> And there's no way I would have written

24:50 this article. I'll show you at the

24:52 bottom. Look at all these. Look at what

24:53 we cover. Like it's insane. This blog

24:56 post is enor. You can see it's like

24:57 8,800 words. And I created this like

25:00 quick reference commands table. This is

25:02 like how Claude reads. This is how it

25:04 searches beyond your current directory.

25:06 This is how it searches the web. This is

25:08 how it writes files. This is how it

25:09 executes code. And like this is an

25:13 awesome reference now for people that

25:14 have never used Claude code. There is no

25:17 way I would have done this myself.

25:19 Claudia.

25:20 >> So you're Yeah.

25:21 >> I started to generate a list like you

25:24 can see in my outline like I started to

25:27 generate a list of like commands as I

25:31 used Claude like oh this stuff should go

25:33 in my document but as I worked with

25:35 Claude it taught me like oh I also used

25:38 these other commands and it so this blog

25:41 post is way more thorough than it would

25:43 have been if I had just done it on my

25:45 own. co-authoring with Claude along

25:47 along the way like section by section.

25:49 Yeah.

25:49 >> Yeah. And then the other thing that I do

25:51 is like once I get to a complete

25:53 outline, I actually start writing and I

25:55 still do all of my own writing. I want

25:57 it to be in my voice. I personally like

25:59 to write and I have I feel like I have a

26:01 very specific cadence to my writing that

26:03 like Claude just doesn't really get. It

26:05 gets close but not really. And so what

26:07 I'll do is once I have a really detailed

26:09 outline, I will write a section. And

26:12 then as soon as I'm done with the

26:13 section, I just pop over here and I'll

26:15 say like, "Claude, I wrote the intro.

26:17 Give me feedback."

26:19 >> And then Claude can see the same file I

26:21 can see, right? So I don't have to cut

26:23 and paste anything. And then Claude will

26:25 give me feedback and then it always

26:26 finds typos and it'll just be like, "Do

26:27 you want me to fix the typos?" And I can

26:29 just type yes.

26:30 >> Why don't we try it? Can you get it to

26:31 fix the typos right here? Like does it

26:33 edit your file?

26:34 >> Yeah.

26:34 >> Yeah. Can you fix the typos in my plan

26:39 mode?

26:41 Got it. Yeah.

26:42 >> And it's just going to go and do it. Um,

26:44 so you can see here it's showing me a

26:46 diff of what it's fixing. Um, I don't

26:48 want it just fixing random typos in my

26:50 blog post as I write. So, one thing I've

26:53 taught Claude to do is when we review a

26:55 section, first it tells me what's

26:57 working well. Then it tells me what

26:58 could be better or what could be clear,

27:00 it does a technical review. Am I saying

27:02 anything that's technically wrong? And

27:04 then the last thing it does is it lists

27:06 the typos that it found. and it asks me

27:08 if I want to fix those typos. And that's

27:10 just sort of a safety check. Sometimes

27:12 like I'm using a word that it thinks is

27:13 a typo, but it's not. And so it allows

27:15 me to quickly look down like this is

27:17 kind of an example. This is one it told

27:18 me after the fact. But when we're doing

27:20 our section by section review, it'll be

27:22 like here's what I want to change. And

27:24 then I can just say yes, change it all.

27:26 Or I can be like yes, change it all, but

27:28 don't change this one.

27:29 >> Got it. Got it. Got it.

27:29 >> So I kind of still keep like safety

27:32 guardrails on claude before it just

27:34 starts editing my blog post.

27:35 >> Interesting. It's interesting that you

27:36 still Okay, so you still manually write

27:38 the actual blog post from a detailed

27:40 outline. You still write it your

27:42 >> Yeah, I've experimented with letting

27:44 Claude write stuff.

27:46 >> Um I don't like the voice and Claude

27:49 knows a lot about my writing. Like I

27:50 have my entire product talk archives in

27:53 here. I have taken the time to develop a

27:56 really good writing style guide. For me,

27:57 good writing has a cadence that like you

28:00 can hear. And that's the part of my

28:03 writing like I just can't get Claude to

28:05 get 100% right. And then I also like to

28:08 write to figure out what I think. And so

28:10 like I don't want to outsource all of

28:11 that to Claude because I want to force

28:13 myself to do the thinking myself.

28:15 >> Got it. Got it.

28:16 >> I'm using Claude almost like a sparring

28:18 partner. So like here's how I'm thinking

28:20 about it. What do you think? Oh, I don't

28:22 really like that. What if we try this

28:23 other way? Um and then I do that

28:25 throughout the whole construction of the

28:27 blog post.

28:27 >> Yeah. It's like having a editor and a

28:30 researcher and like you know just kind

28:32 they're sitting there with you, right?

28:33 >> But like full-time, right? It's not like

28:35 I did everything and then you edit it.

28:37 We're doing it very collaboratively.

28:39 >> And then the other piece that I really

28:40 like is that the way that I used to

28:42 write, I used to write like a 2 to

28:43 3,000word blog post in like three or

28:46 four days. And a reason why it would

28:48 take so long is I would write a section

28:50 and writing is exhausting, right? And so

28:53 at the end of writing a section, I'd be

28:54 like, I'm going to go check my email.

28:55 and then I'd go get distracted for an

28:57 hour and then I come back and write

28:58 another section. But when I'm writing

29:00 with Claude, as soon as we finish the

29:02 review of a section, Claude's always

29:04 like, "Are you ready for phase two?" And

29:06 I'm like, "Yeah, [clears throat] okay. I

29:07 guess I am ready for phase two." So just

29:10 having Claude like poke me, I maintain

29:12 momentum. And so I now a lot of my blog

29:15 posts are a lot more in-depth. They're a

29:17 lot more detailed. They're getting

29:18 longer. Um like you can see this blog

29:21 post is pretty darn long. But you'll see

29:24 there's there is literally no fluff.

29:26 Like I challenge people. This article is

29:28 coming out on November 12th. Um, go read

29:31 it. You can tell me if you think there's

29:33 any fluff in it. Like to me, this is

29:36 just a very detailed guide for how to

29:39 work with Claude safely. And I wrote all

29:42 of this myself. It's 9,000 words and I

29:45 wrote it in one and a half days.

29:46 >> Wow.

29:47 >> Like that is insane to me.

29:49 >> Okay, let let me ask you like a few

29:50 questions about this process. Yeah.

29:52 >> So, so you have one window open for

29:54 writing, do you just uh do you ever

29:56 clear the context or you just keep keep

29:57 going and and you let

29:59 >> That's a really good That's a really

30:01 good question. When I'm working on an

30:03 outline, like when we're developing, so

30:05 like in this writing exercise, we're

30:07 developing an outline, I try to keep all

30:10 of it in the context of the same

30:11 conversation.

30:12 >> Okay?

30:12 >> There will be times like I'll keep an

30:14 eye on whether or not it's going to want

30:15 to compact the conversation. I always

30:17 try to keep Claude from compacting. Like

30:20 if it looks like that's getting close,

30:21 what I'll do is I'll be like, "Claude,

30:23 we're going to run out of context

30:24 window. Let's write a summary of where

30:27 we are that you can read in when I clear

30:30 the context window." And the reason why

30:32 I do that is I want oversight on how

30:34 Claude is compacting the conversation.

30:36 Whereas, if you use the compact, like if

30:38 you just let Claude use its own compact

30:39 tool when you run out, I find that it

30:41 loses a lot of context and detail that I

30:44 don't want it to lose. So, one thing

30:46 that I've done, I don't always do this

30:47 with writing, but I always like when I'm

30:49 working on a project with Claude, in

30:51 fact, that needs to go in this blog

30:53 post. So, I'm glad we're talking through

30:55 this. Uh, I have Claude create this

30:58 document called process notes. And as

31:01 we're working, if it looks like we're

31:03 getting towards the end of the compact,

31:04 the context window, I'll stop and have

31:07 Claude update process notes. And process

31:10 notes is just a text file that's like,

31:12 here's what we did in every session. so

31:15 that I have a history of what we've done

31:16 in the decisions that we've made. And

31:18 then sometimes Claude like just loses

31:21 stuff in the context window. I can be

31:23 always be like, can you just search

31:24 process notes? Like I feel like we

31:26 already made this decision. Um and so

31:28 this is a habit I've gotten into

31:30 whenever I work with Claude and anything

31:31 meaty.

31:32 >> We're constantly co-creating process

31:34 notes together. And I'm getting to the

31:36 point where like I have a a process for

31:39 my process notes where I'm gonna write a

31:42 sub agent that's my document that Claude

31:45 will just call to write those process

31:46 notes when we're running out of context

31:48 window.

31:49 >> Okay, got it. [laughter]

31:50 You've gone to like uh this is like

31:52 inception. You have like

31:53 >> Yeah.

31:54 >> processes for I mean this is the hard

31:56 part, right? Like when the context

31:58 window fills up, bad things happen. Like

32:00 I think you waste a lot of work. Claude

32:03 loses details.

32:04 >> Yeah. He gets dumber. it just gets

32:05 dumber. And so I feel like it's our job,

32:08 at least for now, to like

32:10 >> maintain this process so that when

32:13 Claude resets and its memory gets wiped,

32:16 it can pick up where you left off and

32:17 you'll be fine. Anthropic has tried to

32:19 build that in with that compact

32:21 conversation feature, but it's it's not

32:24 very good. So I like to I like to manage

32:26 it myself a little bit more.

32:28 >> Well, I think that's the thing with this

32:30 uh this like that that's the thing with

32:32 clock code. It just like it's it's

32:35 really kind of made for power users to

32:36 be honest like who really want to

32:37 personalize every everything like it's

32:39 not super it's not super intuitive to

32:41 use if you're a complete beginner but

32:42 you have to like go into the rabbit hole

32:45 >> and it gets more and more powerful.

32:47 >> You know what I like about it though for

32:48 product people is that like using cloud

32:52 code right now and especially pushing

32:53 the boundaries with cloud code is living

32:55 on the edge of what's possible today.

32:57 And I feel like if we're going to build

32:59 with AI, like as product people, if

33:01 we're going to build AI into our

33:02 products, like we should be living on

33:04 that edge. And so like even just the

33:06 stuff we talked about with context

33:08 windows, I wouldn't have learned any of

33:10 that if I hadn't been in cloud code all

33:12 day every day. And so now when I go work

33:14 on like my AI product, when I'm working

33:16 on my interview coach, I have like a

33:18 depth of experience of understanding how

33:20 to manage a context window that there's

33:22 no way I would have developed otherwise.

33:24 And so like yeah, like in the long run,

33:26 is this how we're going to work with

33:27 LLMs? Probably not. Like there'll be

33:29 tooling around it and the labs will get

33:31 better at managing the context for us.

33:34 But like as product people, we can't

33:35 wait for that. We're building AI into

33:37 our products today. So I feel like we

33:39 have to be living on that edge so we can

33:41 build our products on the edge as well.

33:43 >> Yeah. You got to build on the edge. You

33:44 got to learn the latest info.

33:46 >> Yeah.

33:46 >> So that's kind of uh I I just want to

33:48 kind of talk about the context window

33:49 thing a little bit more. So you know so

33:52 so context is like everything for how

33:54 effective these LMS work right. So I

33:56 just want to talk about the three the

33:58 three layers of context that you talked

34:00 about in your previous blog post.

34:01 >> Yeah.

34:02 >> And uh you don't have to maybe share

34:04 your cloud MD file but maybe like talk

34:05 about maybe start with the highest layer

34:07 and talk about what you put in there and

34:08 then and then kind of like you know

34:10 >> Yeah. So I actually have that article

34:12 right here. Let's see. Let's see what

34:15 state it's in. So this article is life.

34:17 You can find it at producttalk.org. And

34:19 in this article, I just talk about like

34:21 the opening hook is this story about how

34:23 I just got sick of trying to get Claude

34:26 and chat GPT to like be reliably good.

34:29 And I realized that like one of the

34:30 challenges is that in order for Claude

34:33 to be good, it needs to know all about

34:35 me and my business. It needs to know

34:36 that I'm a product discovery coach. It

34:38 needs to know that I write at Product

34:40 Talk. It needs to know that my audience

34:41 is crossunctional product teams. It

34:43 needs to know that I have a course

34:45 business and what all those products

34:46 are. It even needs to know things like

34:48 who I work with, right? In order for me

34:51 to say like not only in my task window,

34:54 I can't I can add tasks to my to-do

34:56 list, but I can also say create a Trello

34:59 card on Wena's board and Wena is my

35:02 admin, right? And so I can be like

35:04 create a TR or I can even better I can

35:07 create a task file in my system in

35:09 Markdown and then say create a Trello

35:11 card on Wenna's board using this task.

35:14 So it's like in my system I'm creating a

35:16 task for Willena and then Claude goes

35:18 and pushes it to her Trello board. But

35:20 okay, that only works if Claude has the

35:22 context like who's Wena and what's her

35:24 Trello board and like what's a task,

35:27 right?

35:27 >> Yeah.

35:28 >> So in order to get this like I wrote

35:30 some pretty lazy stuff new task blah

35:32 blah blah blah like in order for that to

35:34 work Claude has to have this view that

35:37 we're looking at doesn't have images but

35:39 I just explained this idea that every

35:40 conversation starts from scratch. We

35:42 have to create memory. Memory has three

35:45 layers. You asked about the three

35:46 layers. The first one is my global

35:48 preferences.

35:50 >> Let me actually pop over to the web

35:51 because I can show some of my CloudMD

35:54 files pretty safely.

35:55 >> Okay.

35:56 >> So, this is my global cloud MD. The

35:58 start of it. Um, I can zoom in a little

36:01 bit.

36:01 >> Perfect. Yeah.

36:03 >> Um, this top section I write about in

36:05 the blog post is just a little thing I

36:07 learned is that like I don't write my

36:10 Claude MDs anymore. Whenever I'm working

36:12 with Claude, when we're done working, I

36:14 say, "Hey, Claude, what did you learn

36:15 about working with me? Like, what should

36:17 we add to the Claude MD so this goes

36:19 smoother next time."

36:21 >> And that's been awesome because I don't

36:23 have to maintain this file anymore.

36:24 Claude does. But you can see like I've

36:26 just defined some personal preferences.

36:28 Like I always want to plan before Claude

36:30 does anything. I never let Cla Claude

36:32 just do stuff. Um, and I do plan at

36:35 multiple levels. You could saw that with

36:36 the blog post. Like I start with just

36:38 really rough thoughts. We work our way

36:40 to an outline. Once I have an outline,

36:42 we start writing. So there's just this

36:45 this file is across all my projects,

36:47 coding, writing, tasks, everything. I

36:50 just want Claude to know like this is

36:52 what I like. And it's not very long.

36:54 This file doesn't go much further behind

36:56 be beyond this. It's pretty short. I

36:58 just have a little section on what how I

37:00 prefer to get feedback from Claude. And

37:02 it's really simple. And this lives, you

37:04 can see here, it lives in my local user

37:08 directory. So on my Mac it's like tilda

37:11 ttorus and thencloud cloudmd.

37:14 >> Yeah. Do you put like theres as a

37:15 product coach and like you know here's

37:17 my business and all that.

37:18 >> All I don't put any of that here because

37:21 this file is going to get loaded in the

37:23 context window every single time you use

37:25 claud.

37:26 >> Got it.

37:27 >> So you only want to put stuff in here

37:29 >> that you always want it to follow. And

37:32 so like if I'm using cloud to like

37:33 brainstorm Christmas gifts for my

37:35 husband, Claude does not need to know

37:36 about product talk.

37:37 [laughter and snorts] So I don't want

37:39 that here, right? But no matter what,

37:42 >> you want to keep this short. Yeah.

37:43 >> Yeah. This is literally global. So you

37:45 want to keep it short. It's going to go

37:46 in every single context window. And like

37:49 I ask Claude all the time, can my dog

37:50 eat this food? Like Claude doesn't need

37:52 to know about product talk for those

37:53 stupid queries.

37:54 >> Right. Okay. Got it.

37:56 >> Like what's really important is you have

37:58 to keep the context window as clean as

38:01 possible.

38:02 >> Yeah.

38:02 >> Right. And so this literally is just my

38:05 global rules. But you'll see later if we

38:08 just scroll down, I have project

38:09 specific instructions. So this is

38:11 actually my writing cla MD. And in my

38:14 writing cla MD, I have like it says at

38:17 the start of each section, read my

38:18 writing style guide. This is actually

38:21 old. I moved things out of Dropbox. But

38:23 if you remember, if I go back to the

38:25 terminal, which is hard to do with this

38:28 sharing thing here, um when I ran um

38:32 sorry, here when I said, "Let's work on

38:33 a blog post together," the very first

38:35 thing Claw did was it went and read my

38:37 writing style guide.

38:39 >> That's right. Yeah.

38:40 >> And it's somewhere in here. Yeah. And

38:42 that's because here it says at the start

38:44 of every session, read the writing style

38:46 guide.

38:47 >> Okay.

38:48 >> And then this is where I get into my

38:50 rules about writing. So Claude is a

38:52 thought partner, not a writer. You're

38:54 acting as an editor. You can do some

38:56 research and development. And this

38:58 document is also not very long. It's

39:01 really just telling Claude, "This is how

39:02 we write together."

39:04 >> Okay.

39:04 >> One thing that is in my global claude

39:07 MD, which I don't, let's see if I have a

39:10 screenshot of it somewhere, is [snorts]

39:11 this idea of context. So I have my I

39:15 have my context files defined. So I've

39:18 created little markdown files. And the

39:20 key is they're little. So I have a lot

39:23 of them because I want to I want to mix

39:27 and match them so that I can tell Claude

39:30 which files matter for the task at hand

39:32 and it doesn't have to read all of my

39:34 context every time. So I have a business

39:37 profile that just is like this is what

39:39 my business does. Here's where you can

39:41 find my product descriptions. Here's

39:43 where you can find my marketing

39:44 channels. Here's where you can learn

39:45 about my team. I can show you that one.

39:47 But it basically is telling Claude where

39:49 to find more details in all these files.

39:52 And so I think this blog post has a

39:54 screenshot of that. Yeah. So this is my

39:56 global cloudmd. We already looked at

39:58 this top part. You can see at the bottom

40:01 I'm telling it here's where you can find

40:03 reference context files. And I basically

40:05 tell it specifically

40:08 only use these files if they're relevant

40:10 to what I'm asking you about.

40:12 >> So if I'm asking you about something

40:13 related to my business, go look at my

40:15 business profile. And if I'm asking you

40:17 something personal, go look at my

40:18 personal profile. And then you

40:20 >> Okay, this makes sense. So you don't

40:21 have to like load a lot of into the

40:23 default context. Yeah. So you don't

40:24 crowd the context.

40:25 >> Exactly. So like if I ask it, I have a

40:28 product file all about my storybased

40:29 customer interview course. So if I ask

40:31 it like, hey, let's work on the landing

40:34 page for my storybased customer

40:35 interview course. It already knows where

40:37 to find details about this course. It

40:39 already knows details about my company.

40:41 It already knows details about my target

40:42 audience. I don't have to tell it any of

40:44 that stuff. And it's because I've given

40:46 it an index of context files so it can

40:49 just go pull in the relevant stuff.

40:51 >> Wow. Okay. But don't you feel like uh

40:53 there's going to be like too many files

40:55 to manage or like I guess you you just

40:56 make cla update your context files,

40:58 right? Is that [laughter]

41:00 >> So I do in this article I talk about

41:03 this. By the way, this article has a

41:05 really fun use case of like how to use

41:07 cloud to do a competitive analysis. So

41:09 you can see here I'm having Claude like

41:11 >> this isn't for me. I used 11 Labs as my

41:13 as if that was my company. Um, and it's

41:16 generating like uh let's see if I can

41:19 find it. I don't know if I'm going to

41:21 find it. But here's the goal. Like the

41:23 key is to give Claude just enough to go

41:25 find what it needs when it needs it. And

41:28 part of this article is about like,

41:31 okay, well, how do I maintain all this

41:32 stuff? Like I have a ton of files now.

41:34 How do I keep it current? And so what I

41:37 do is I have this section about like,

41:40 okay, how do I keep it up to date? The

41:41 way we're going to keep it up to date is

41:43 at the end of every session with Claude,

41:45 I literally say, "What did you learn

41:47 about me that we should add to a context

41:49 file?" And then we have a conversation

41:51 about where to add it. Should it go in a

41:53 cloud MD? Should it go in one of these

41:55 LM context files? Um, like what's the

41:58 right place? And then I'm working with

42:00 Claude to make sure like Claude wants to

42:02 jam everything into your CloudMD. You

42:04 don't want that. That's getting loaded

42:06 in every conversation. So I work with

42:08 Claude to be like, "Hey, it seems like

42:10 that should go in my marketing channels

42:11 file. Why don't we add that there?"

42:13 >> Got it.

42:14 >> So I think about it as like information

42:16 about me and my business goes in a

42:18 context file. Whereas my working

42:20 preferences like this is how me and

42:22 Claude work together goes in a cloud MD.

42:24 >> So basically, yeah, this is a really

42:26 important tip. Like after every

42:28 conversation, you ask Claude, how can

42:29 you make the context better? And you

42:31 have all your contexts in the LM context

42:33 folder with a bunch of sub subfolders

42:36 and files.

42:37 >> Yeah.

42:37 >> Right. Yeah. Wait, so how much are you

42:39 playing this uh cloud employees? They

42:42 sound pretty useful. How much are you

42:43 paying? [laughter] 200 bucks a month.

42:45 >> Uh yeah, I'm on the $100 a month plan.

42:49 >> Okay.

42:49 >> So you can see I'll show you here in in

42:52 Obsidian. You can see all my context

42:54 files. So like for my business, I have a

42:58 target audience file, a marketing

43:00 profile, a list of my differentiators,

43:02 my company overview, my business model.

43:04 >> I didn't like sit down and create this

43:06 all one day. That would have been

43:08 tedious and horrible.

43:09 >> I basically anytime I felt like I needed

43:12 to describe something to Claude, instead

43:15 of just doing it that one time and then

43:17 having to repeat it later, I'd be like,

43:19 "Oh, Claude, in order to do this task, I

43:21 feel like you need to know about my

43:22 differentiators. Maybe interview me."

43:24 And I literally claude will interview

43:26 you can just be like interview me about

43:28 my business and then it will ask you a

43:30 bunch of questions and then it will

43:31 write a context file for you based on

43:33 what you learned.

43:34 >> Got it. Okay.

43:35 >> Um so that's that's a lot of how I did

43:37 most of this stuff.

43:38 >> Okay. Yeah. As long as you just sat down

43:40 one day and made all these folders,

43:41 right? It's not

43:42 >> No, I add to it like gradually over

43:44 time.

43:44 >> So what about like uh Okay, last

43:46 question. Let's say I'm just overwhelmed

43:48 by all this stuff. like this is like

43:51 [laughter]

43:51 >> like I want to learn stuff but like it

43:53 sounds like you really personalize it

43:55 for your life.

43:56 >> Yeah.

43:56 >> Like what are your three tips to get

43:57 started? Let's say I already use cloud

43:59 code for like co coding but I want to

44:01 use it to manage my life.

44:02 >> The first thing is with context you

44:05 don't have to get here overnight. Like

44:08 >> I think the simplest rule is whenever

44:10 you find yourself explaining context to

44:12 Claude, stop and think about am I ever

44:15 going to have to explain this context to

44:17 Claude again? And honestly, the answer

44:18 is probably yes. So instead of

44:21 explaining it to Claude, change tasks

44:23 just for a minute and just be like,

44:24 "Claude, I need to explain some context

44:25 to you. Let's capture it in a context

44:27 file."

44:28 >> Got it. Yep, that makes sense.

44:30 >> You probably need to think through like

44:32 at a minimum, do you work with Claude at

44:35 work and in personal stuff? Then maybe

44:36 set up a work folder and a personal

44:38 folder. That's it, right? And then I

44:41 think from there, you can just do it

44:44 iteratively over time. And that's what I

44:46 did. Like I literally never sat down and

44:48 just created all these files. I just And

44:50 some of them are just stubs, right?

44:52 They're not that complete. Um but I know

44:55 I'll add to them with time. And then

44:57 this file we're looking at, this

44:58 profile, I never wrote a word of it.

45:00 Every time I have Claude add a context

45:03 file, I say, "What index needs to be

45:05 updated?" And it just figures it out.

45:06 >> Got it. Okay. So maybe like the process

45:09 here is like think about your weekly

45:11 calendar. What's like the most like

45:12 what's taking up most of your time? And

45:14 then and then uh try to you know you

45:16 have Claus as your employee try to get

45:17 it to you know give a bunch of

45:19 instructions backgrounds context have it

45:20 save a bunch of files and then just work

45:22 with it to start doing stuff and then at

45:24 the end of every conversation ask it to

45:26 like add more stuff to context or like

45:29 improve his prompt that that's kind of

45:30 [laughter]

45:31 that's kind of what you do right

45:32 >> yeah okay I think there's two steps to

45:35 this first is I have worked with an

45:38 admin for like 10 years so I'm really

45:40 good at like looking at things that I

45:43 can delegate And for anybody who's ever

45:45 delegated something to another person,

45:47 what makes that work well is to have

45:49 like a good like standard operating

45:53 procedure for the task you're giving

45:54 them, right? It's not you can't just say

45:57 to someone another human like, "Hey, go

45:59 do this thing and have them expect them

46:02 to be able to do it the way that you

46:03 would do it." And so if you want them to

46:04 do it the way that you would do it, you

46:06 have to provide for them like a process

46:09 like here's how I do it. Here's the

46:11 outcome that I want. And so like I do

46:13 this with my admin but I record videos.

46:15 I record a video of how I do things. She

46:18 then looks at the video. She creates

46:20 like a Trello checklist for herself and

46:22 then that becomes like our standard

46:23 operating procedure. So we've been doing

46:25 this for years. And so I'm already good

46:27 at this muscle of like what can I

46:29 delegate? So when I started doing this

46:31 with LLMs, I thought about it as like

46:33 okay well if Claude is just a person on

46:35 my team, how would I delegate to Claude?

46:37 And so I started by literally looking at

46:40 my own Trello board and like what do I

46:41 need to work on today? And then it for

46:44 every task I got in the habit of like

46:46 how can Claude help with this.

46:47 >> Got it. Got it.

46:49 >> And then there's pieces for some tasks

46:51 like I don't want to automate them. I

46:53 enjoy doing them. Like I enjoy writing.

46:54 I don't want Claude to write for me. So

46:56 like I don't want to automate that. So I

46:58 think about like how can Claude augment

47:00 this. What do I do while writing that

47:02 Claude could accelerate? well, research

47:04 and web searches um and like coming up

47:06 with alternative analogies. And then

47:09 there's other tasks like send this

47:12 stupid receipt to my finance system. I

47:14 don't need to ever do that. I could

47:16 automate that. So then I look at like

47:18 what do I have to do to give Claude

47:19 context to be able to automate this

47:21 completely. Um so

47:23 >> okay,

47:24 >> I didn't even start with the things that

47:25 take the most time. I literally just

47:27 forced myself all day, every day, every

47:30 time I do a new task to think about like

47:33 how can Claude help.

47:34 >> Got it. All right, man. That that that

47:36 that's a great tip to end on. I I think.

47:39 [laughter]

47:39 >> All right.

47:39 >> Yeah. Just hire hire $100 employee and

47:42 then uh get it to help on different

47:44 things. Yeah.

47:44 >> Yeah. And I mean, it's funny that I even

47:47 thought twice about upgrading from $20

47:49 to $100. I started on the $20 plan. I

47:52 only had to bump to the $100 plan when I

47:55 was doing both a lot of coding in one

47:58 week and a lot of writing in one week.

48:00 So, it's like I was literally pushing

48:03 hard on Claude for a long time. So, you

48:06 can get pretty far on the $20 plan.

48:08 >> Yeah. Yeah. Yeah. Yeah. And hopefully

48:10 Anthropic gets his limits together, you

48:12 know.

48:12 >> Yeah. [laughter]

48:13 >> So, yeah.

48:14 >> Okay. So, to find your detailed cloud

48:15 code guides, just go to producttalk.org,

48:18 right? Is that the right site? Yeah,

48:20 it's all we can take a look right here.

48:22 It's productt talk.org and it's all like

48:24 all my recent articles. You can see what

48:27 is claude. Stop repeating yourself.

48:30 Probably by the time this episode goes

48:31 out that safety article will be live.

48:34 >> Awesome. I'll I'll definitely read it.

48:36 Yeah, [laughter]

48:37 >> there's a ton here. I'm going to be

48:38 writing a lot more. Like a lot of my

48:40 goal for the next several weeks is just

48:43 to help make Claude code accessible for

48:46 people for nontechnical people and to

48:48 really start to show the power of like

48:50 pair working with Claude. Um so if

48:53 listeners were like excited about this

48:55 episode definitely check out this

48:56 series. It's been a fun to write and I'm

48:59 I'm even going to be hosting Claude code

49:01 office hours. So if you want to even

49:02 come and get help uh I'm going to be

49:05 doing that every month.

49:06 >> Awesome. Awesome. You should should be

49:07 there. You should be Anthropic exposing.

49:09 Yeah, [laughter] that sounds great.

49:11 Yeah. Yeah.

49:11 >> I know. Someone asked me if I was being

49:13 sponsored by Anthropic, and I'm not. Uh

49:16 if I was, I would disclose that. I

49:17 really am just a big fan.

49:19 >> Yeah, I'm a really big fan, too. I'm a

49:20 really big fan of like Capors and the

49:22 team and what they're building and um

49:24 yeah, it's funny because they're they're

49:26 really kind of trying to build towards

49:27 the coding use case, but like you know,

49:29 there's so much more you can do to build

49:30 a really good agent.

49:32 >> Yeah.

49:32 >> Cool. All right, Teresa, thank thanks so

49:34 much for your time.

49:35 >> Thanks for having me. This has been fun.

You have ... regenerations remaining for this video.

Length

Short (100 words) Medium (300 words) Long (500+ words)

Tone

Casual Professional Conversational Persuasive

Focus

Overview Key Points Actionable Takeaways

Language Style

Simple Standard Detailed

Additional Notes (Optional)

500 characters remaining

Save these preferences for future summaries

Length

Short (~300 words - quick blog post or Substack snippet) Medium (~600 words - standard blog post) Long (~1,000+ words - in-depth article or feature)

Tone

Casual (friendly, informal) Professional (formal, business-like) Conversational (chatty, engaging) Persuasive (call-to-action focused)

Keyword Emphasis and SEO Optimization

These keywords will be used to optimize the article for search engines

Auto-generate meta title (50-60 characters) Auto-generate meta description (150-160 characters)

Content Focus

Full Overview (use entire transcript) Actionable Takeaways (focus on practical advice or steps)

Additional Notes (Optional)

500 characters remaining

Save these preferences for future blog posts