You Can Learn AI Chief Of Staff With Claude Code In 29 Min | Founders, PMs, Skills — Transcript
Full transcript
- 0:00Everyone's Sean. So, today I want to
- 0:01show you how we can use Claude code as
- 0:03your AI Chief of Staff. Previously, I
- 0:05made a lot of videos about how you can
- 0:06break down, you know, agent harnesses
- 0:08and memory systems in a very technical
- 0:09way. But, I realized that a lot of my
- 0:11audience are also running their own
- 0:12business, owning their own agency, or
- 0:14they're a product leader who's leading a
- 0:15technical team. So, I'd like to show you
- 0:17exactly how a startup founder or an AI
- 0:21developer would use Claude code to run
- 0:23their day-to-day as if it's your Chief
- 0:25of Staff. And, I'm not only going to
- 0:26show you a system design chart where I'm
- 0:28going to show you how you can set up a
- 0:30company brain for your Claude and how
- 0:32exactly it should be working using some
- 0:33of the skills and some of the tools that
- 0:36you should be connected to using the
- 0:37connectors or MCPs. And, eventually, how
- 0:39you will be able to trust the decisions
- 0:41they made. But, also, I will walk
- 0:43through five top use cases that I think
- 0:46are going to be really helpful for your
- 0:48day-to-day tasks. And, they include, you
- 0:50know, how you can plan for your day, for
- 0:52tomorrow especially, using some of the
- 0:54connectors. How you can triage your
- 0:56inbox if you're a client-heavy person.
- 0:58And, how you can prepare for your
- 1:00meetings for tomorrow using some of the
- 1:02tools such as, you know, these Google
- 1:03Suite, Waku Memories, and Meeting Notes.
- 1:06To how you can align the rest of the
- 1:08team with your agents by sharing some
- 1:11context together and making decisions
- 1:12together. And, last but not least,
- 1:14closing the deals by, you know, creating
- 1:16a quote to instant it to your CRM. So, I
- 1:19think this should be something that's
- 1:20going to be helpful for you if you are
- 1:22just getting started using Claude code.
- 1:24But, also, if you are someone who's
- 1:26collaborating with the rest of your team
- 1:28using agents together, this should be
- 1:30very helpful for you, too. So, without
- 1:32further ado, let's get started. Based on
- 1:34these five items I'm going to show you
- 1:36for the rest of the video, I prepared
- 1:38five possible tasks that we need, right?
- 1:41Maybe you want to set up a 7:00 a.m.
- 1:43morning brief every day for your team.
- 1:45And, maybe there's a client email that
- 1:46just arrived. Or, maybe you want to
- 1:48prepare some meeting notes or take some
- 1:49voice notes. All the way to there's a
- 1:51team checking you want to do with the
- 1:53decisions that are made by agents or by
- 1:55human. And, eventually, maybe you have
- 1:57got a request for a quote from your
- 1:59clients. Okay? And all of these things
- 2:00will firstly hit Agent harnesses, which
- 2:02Claude is our main character today. I
- 2:04list the rest of them here just for
- 2:06examples that you can use the rest of
- 2:07them, too. This is the place where
- 2:09you're going to talk to your AI and
- 2:11where the AI is going to work. And since
- 2:13you clicked into this video, I kind of
- 2:14can assume that you already have been
- 2:16using Claude. And just in case you don't
- 2:18know what it is, you can try immediately
- 2:19at claude.ai and then can ask any
- 2:22questions you want. And I know a lot of
- 2:23agency owners or business people who are
- 2:25going to use Claude chat or Claude for
- 2:27work very heavily. But today our main
- 2:29character is Claude Code, which is a
- 2:31different species. The reason why it's
- 2:33different is that everything in your
- 2:34claude.ai is going to be stored on
- 2:36Anthropic's cloud. So you actually don't
- 2:39own any of these things, okay? Claude
- 2:40Code instead is something you will need
- 2:42to download to your laptop as a desktop
- 2:45app and you will be able to connect it
- 2:46to all these tools and then all the
- 2:48sessions and all the chat history and
- 2:50all the skills are going to be local.
- 2:52And this is what we're going to try to
- 2:53show you today, okay? You can click on
- 2:55download for Mac OS if you're on a Mac
- 2:57OS or I'm sure there's a way to download
- 2:59it on Windows, too. If you just click on
- 3:01new, you can open a new chat into
- 3:02something like this. And please bear
- 3:04with me if you're already familiar with
- 3:05Claude. I guarantee that I will dive
- 3:07into more interesting use cases very
- 3:10quickly. And this is just an intro for
- 3:12the beginners. Now that we have started
- 3:13talking to your local Claude Code, the
- 3:15first thing you should think about,
- 3:16always, is how can I set up a company
- 3:18brain for the rest of the team, okay? I
- 3:21would consider the brain to have three
- 3:23main pillars. The first thing is what I
- 3:25call a handbook, which is something that
- 3:27you write as some kind of durable
- 3:29long-term memories for your agent. And
- 3:31they could be things like the offers and
- 3:33prices you offer to your clients. They
- 3:34could be the tone of how you should be
- 3:36talking to you and the SOPs of your
- 3:39company. And what is something true
- 3:41about your business? I have stored
- 3:42something like this in a Notion page
- 3:45where you can you can see that I have a
- 3:46handbook here and the handbook's
- 3:48basically telling me uh what we're
- 3:49offering at waku.one, which is a shared
- 3:51memory and a board for teams and their
- 3:53agents to collaborate together. All
- 3:55right, this is the explanation for our
- 3:56company. And you you probably have to
- 3:58have some rules, for example, uh for
- 4:00sponsorships, uh
- 4:01anything over $3,000, needs a yes from
- 4:04Sean, okay? Or any code sales pitches
- 4:06get no reply, stuff like that, okay? You
- 4:08can turn this into something more
- 4:09complicated, but it's like Obsidian,
- 4:11right? You can also store something like
- 4:12this in your notes, in your Obsidian.
- 4:14For me, I'm just storing such a thing in
- 4:16Notion, and uh, I'll show you later how
- 4:19you will be able to connect this into
- 4:21your chief of staff, so that you can get
- 4:22it up to working. The second thing is
- 4:24called a record. If you own your own
- 4:26brand, if you have your own agency, or
- 4:28you're just selling software to your
- 4:29clients, you probably have heard of this
- 4:31term called CRMs. A lot of technical
- 4:33people will sometimes prefer to use
- 4:34database by themselves, uh, which I
- 4:36would recommend starting with, uh,
- 4:38Superbase, which is a good tool, but,
- 4:40uh, if you're more technical, feel free
- 4:41to just use AWS for everything. But, if
- 4:43you're someone who is running your own
- 4:45business, running your own team, and you
- 4:47don't really have time to worry about
- 4:48this, uh, having a CRM, or a just a
- 4:50Google Sheet, could be really helpful
- 4:52for storing these records, okay? And
- 4:54this is just a place for you to store
- 4:55information about your customers, your
- 4:57deals, and your dates, okay? And in
- 5:00today's example, um, later I'm going to
- 5:02show an example using HubSpot, uh, which
- 5:05is also connectable to your AI. And the
- 5:08third thing is what we call the memory.
- 5:09For those who are familiar, you probably
- 5:11have watched my previous memory system,
- 5:13uh, videos, which are more technical.
- 5:15But, for today, in plain English, what
- 5:17memory does is something that agents has
- 5:19made decisions on, or you have made a
- 5:21decision on something, or you have done
- 5:23something, or the agents have done
- 5:24something, anything that has been done,
- 5:26instead of something that's long-term,
- 5:28like what the handbook or playbook
- 5:30would, uh, would be stored in some kind
- 5:32of drives, right? Or it's also not a
- 5:35customer information coming in from
- 5:36emails, or dropped into your CRM system.
- 5:39It's something that you want to keep
- 5:40sharing across your chats and the tools,
- 5:43so that your agents will always be fed
- 5:45with the right context, and context is
- 5:48everything, okay? You might heard that
- 5:50some AI models are getting stronger, you
- 5:52know, some agent harnesses, like Hermes,
- 5:54are getting really popular, or even uses
- 5:57really popular, like that one of the
- 5:58most downloaded apps in the US, or
- 6:01across the world in the past few weeks.
- 6:03Um, literally, they cannot do anything
- 6:04without memories. Okay? Memories could
- 6:06be chat history uh, uh, on Claude AI,
- 6:08could be chat history in your Claude
- 6:10code sessions, but you just need to
- 6:11prepare it really well and then feed it
- 6:13to your agents. Okay? So, these three
- 6:15are what I call the brain for a Claude
- 6:16code. And then we're going to start
- 6:17worrying about how does it get things
- 6:19done? If you're not familiar with this,
- 6:21there are two main things. Number one is
- 6:22skills, number two is artifacts. These
- 6:25are the things that we use every single
- 6:27day by running our own company. And
- 6:29skills is basically a recipe that you
- 6:31want to teach your machine how to do
- 6:33things. Okay? For example, how do we
- 6:35even triage the inbox? How do we prepare
- 6:37for the meeting? How do we write a quote
- 6:39for the clients? How do we do some
- 6:41weekly reviews? Skills essentially is
- 6:43something that allows you to not
- 6:45repetitively explain something to your
- 6:47AI. And the AI should be able to trigger
- 6:49a skill to proceed to work on something,
- 6:52which also in the memory terms is called
- 6:54procedural memories. Okay? And then
- 6:55artifacts is like a dashboard that you
- 6:58just saw earlier. I'll show you exactly
- 6:59how you can create this in a bit, but
- 7:01this is something that we find really,
- 7:03really helpful when we use Claude as our
- 7:05chief of staff to share information
- 7:07across the rest of the team and for team
- 7:09discussions. We don't write PowerPoints
- 7:11anymore these days. We write artifacts
- 7:13and we tell Claude to use a similar
- 7:15design system to create the artifact so
- 7:18that the rest of the team will be able
- 7:19to see at least like a similar template
- 7:20for PowerPoints. Okay? You can also come
- 7:22to the sidebar and click on customize to
- 7:24find out about skills over here. You can
- 7:27use some of the skills by other people.
- 7:28If you use QuickBooks to bookkeeping
- 7:30your small business, bank fees,
- 7:32reconciliation reports and everything,
- 7:33you can just click on plus and then
- 7:35bring it into your system. All right?
- 7:37But we don't we're not going to show
- 7:38QuickBooks today. We're just going to
- 7:39show you some of the skills that we
- 7:41built in house in a bit, but yeah, this
- 7:43is a very important thing to have. I
- 7:45would consider all of these to be the
- 7:48absolute theoretical part of building a
- 7:51context around you know, your agent.
- 7:53What actually makes your Claude your
- 7:55chief of staff is you should inject the
- 7:57actual context into your agent, which in
- 7:59this term we call it the connectors.
- 8:02Okay? The connectors are basically MCPs
- 8:04before. Somehow the AI just love
- 8:06renaming things and now they're called a
- 8:07connectors at clock code. Okay, they're
- 8:09just MCPs basically. Maybe there's
- 8:10something more on top of that. I will
- 8:12feed all these contacts from the company
- 8:15rules you hold in notion and drive and
- 8:18to the rekkers or CRMs where like
- 8:20HubSpot is one of the examples, or you
- 8:22can use Waku memory to share the
- 8:25decisions and memories across teams and
- 8:26agents
- 8:28all the way to workspace where you have
- 8:29Gmail, calendar, Slack where where the
- 8:32work actually arrives to your agents.
- 8:34And last but not least, you can use
- 8:36these tools like Granola, Calendly, or
- 8:38Linear to record meetings, book
- 8:40meetings, and set up tasks. So, all
- 8:42these tools are connectors, which you
- 8:45need, as I mentioned earlier, to inject
- 8:46context into your clock code to turn it
- 8:49into achieve a staff that actually
- 8:51understand what exactly is going on in
- 8:52your world, okay? And the connections
- 8:53again are in the same place. If you come
- 8:55to customize over here, okay? You can
- 8:57see there's a connectors tab over here,
- 8:59right? And if you scroll down a little
- 9:00bit, you can see I we're connected with
- 9:02like Google Drive, calendar, Slack,
- 9:03Notion, uh I lost CM, we're not using
- 9:06it. We're using Linear. Google Docs,
- 9:08Superbase, Figma, blah blah blah blah
- 9:09blah, right? And you can also add some
- 9:11like custom connectors if you build your
- 9:13own thing, right? In our case, we built
- 9:16Waku memories, which you can just add in
- 9:18by https api.waku.one/mcp
- 9:22and click on continue. And because this
- 9:24already exists, I already added it. If I
- 9:26just search Waku over here, you can see
- 9:28that it's already connected. That was
- 9:29connectors, okay? Now you have contacts
- 9:31from all of these things. If we start a
- 9:33new chat right now, and make sure that
- 9:35you create your own folder for storing
- 9:37everything here, right? In our case, I
- 9:39created a folder on my laptop called
- 9:40Chief of Staff. I can ask it questions
- 9:42be like, what connectors do you have
- 9:45access to now? So, you can see it told
- 9:48me that I'm connected to Gmail, Slack,
- 9:50Google Calendar, Calendly, Notion, blah
- 9:52blah blah HubSpot, Stripe, Figma, Canva,
- 9:54and for memory and board, we have Waku
- 9:56here, okay? We're ready for adding this
- 9:58context layer on top of your agent. And
- 10:00before we jump into the real examples,
- 10:02you should make sure that you will
- 10:03always be the one who will approve um
- 10:06the emails that you want to send as a
- 10:07draft. None of these emails or none of
- 10:09these like information injections should
- 10:11be giving orders to your agent because
- 10:13you don't want someone who sends you an
- 10:15email and then that email has a prompt
- 10:16and then that prompt is obviously
- 10:18fishing something and then you are just
- 10:19giving away some of your information to
- 10:21that person. You don't want that to
- 10:23happen. And I think most of the time
- 10:24these agents are smart enough to not
- 10:26allow that to happen. And then last but
- 10:28not least, you should see the work where
- 10:30you want to check out the cost of the
- 10:32agents, the tokens, and also you want to
- 10:34see, for example, how much contacts you
- 10:36have used in each chat and eventually
- 10:38finish each one of these tasks. Okay. I
- 10:40hope this intro is helpful and now let's
- 10:43jump right into these five tasks to
- 10:46experience what an AI Chief of Staff
- 10:48would do for us at Clockwork. Okay. So,
- 10:50what I'm going to do now is that I just
- 10:51copied the link for this artifact and
- 10:53I'm going to paste it here on the
- 10:54browser. And because I'm logged in with
- 10:56the same account, I'm able to read it
- 10:57from here. Okay. So, that on my
- 10:59Clockwork I can focus on doing the
- 11:01tasks. So, right now I open the sidebar
- 11:03and I click on plus new, okay, and turn
- 11:06off this sidebar. Uh remember the first
- 11:08thing we do is that we want to set up
- 11:09the computer, make sure we are in the
- 11:11right folder for everything. If you have
- 11:13not done anything like that, make sure
- 11:15you create an empty folder and then you
- 11:17enter this folder like this. And a good
- 11:19habit to have, if you're not technical,
- 11:20is that you should use a cloud.md file.
- 11:23So, I just open it with cursor over
- 11:25here. You can see that the cloud.md is
- 11:28clearly saying that you should only work
- 11:30with certain rules here. For instance,
- 11:32we're demoing this video, so I don't
- 11:34want the rest of the world to see all my
- 11:35emails. So, it should always work with
- 11:38the Gmails with the demo label on it and
- 11:41the handbook lives in Notion. So, later
- 11:44if I tell it to pull in my handbook, it
- 11:46knows because your agent will by default
- 11:48read the cloud.md in the folder that
- 11:51you're working in, okay? So, the first
- 11:52thing is we should plan for tomorrow's
- 11:55meeting. So, let's click into it and
- 11:57scroll down a little bit. We're just
- 11:58going to quickly go through it and then
- 11:59follow the examples and show you some
- 12:01quick aha moments, okay? So, today
- 12:03you're my Chief of Staff. Use only the
- 12:05demo calendar, mail label demo, and
- 12:07Notion pages under Chief of Staff demo.
- 12:09Draft never send confirming one line.
- 12:11And just to confirm that, if we copy
- 12:12this prompt and put it here, this is
- 12:14asking for tomorrow's morning brief,
- 12:16which connected tools will we use and
- 12:17why in one line each. Do not call them
- 12:20yet. So, this is a good way to quickly
- 12:21check what exactly Clockwork is going to
- 12:23use as your AI chief of staff. And uh
- 12:25now it's telling me that it's using uh
- 12:27Google Calendar, Gmail, Slack, Waku
- 12:29Board, Granola, Linear, GitHub, and
- 12:31HubSpot. The work for us is that we want
- 12:33to make my morning brief for tomorrow,
- 12:35November 5th, as an artifact, okay?
- 12:37Remember, artifact is one of the work
- 12:39that uh eventually you want your AI
- 12:40chief of staff to deliver. And we should
- 12:42use only the demo calendar, which in my
- 12:44case, I'll show you, okay? We have a
- 12:46demo calendar right here, and uh we have
- 12:48a few meetings for tomorrow. We have a
- 12:50weekly stand-up, we have a catch-up with
- 12:53a VC uh called Redwood Peak from Dana.
- 12:56These are all made up, by the way. Uh we
- 12:57have a studio intro with Marcus, we have
- 13:00a
- 13:01Meridian discovery call, which is a
- 13:03potential client in our HubSpot. If you
- 13:05can see, we have a company called
- 13:07Meridian, and we are going to work on a
- 13:10pilot with $5,000 of a deal. And last
- 13:13but not least, we have a sponsorship
- 13:14deal review, which had a $4,000
- 13:17collaboration with 60-second
- 13:18integration. I just made things up, it
- 13:20it doesn't exist. It should work in
- 13:21these following orders with decisions I
- 13:23must make, my five meetings with
- 13:25one-line prep each, and anything
- 13:26suspicious, and keep it to one screen,
- 13:28okay? Now, let's just copy this prompt
- 13:30and then paste it here and see what
- 13:31happens, okay? So, now Clockwork is
- 13:33trying to think. You can always click on
- 13:35this little arrow to see what it's
- 13:36doing, all right? And now it's using a
- 13:37skill called Anthropic Skills Morning.
- 13:40So, it knows I'm preparing some kind of
- 13:41morning brief. It automatically
- 13:43triggered the right skill for the right
- 13:44thing. And it says it's reading the demo
- 13:46calendar, the demo labeled mail, and the
- 13:48Monday Scrum page. I should show you the
- 13:51demo emails, too. So, this is the demo
- 13:53tab of the emails, and uh for some of
- 13:55the folks that I mentioned earlier, like
- 13:57the studio collaboration deal is right
- 13:59here. If I click into it, right? And and
- 14:01I'll just mark it as unread still. And
- 14:03then Dana from Redwood Peak Ventures was
- 14:06asking us for a 30-minute catch-up on
- 14:09Waku cuz she potentially could be our
- 14:11investors. So, these are the demo
- 14:12emails, um which are pretty much synced
- 14:14with the calendars here. Taking quite
- 14:16some time. Let's look at what it's
- 14:18doing, right? It's already taken a
- 14:19minute and something. So, it tried to
- 14:21understand, okay, what labels are there
- 14:23in my Gmail and then make sure that
- 14:25we're using the demo label only. And
- 14:27then for Notion, it was reading the
- 14:29Monday scrum. Uh, I do apologize for not
- 14:32explaining that. So, you can see that in
- 14:35our chief of staff related demo, uh, our
- 14:38Clockwise very smart enough to figure
- 14:40out that it's in the Notion, right? And
- 14:41there's a Monday scrum here. Right?
- 14:43Click into it. I have a clear agenda,
- 14:45last week's decision, and the open asks,
- 14:48right? For example, we should draft a
- 14:50pilot proposal for Meridian. Okay. Uh,
- 14:52cool. So, now it had created this, uh,
- 14:55artifact for us, okay? And this artifact
- 14:57is saying that there's a few decisions
- 14:59for me to make, right? I should accept
- 15:01or counter this sponsorship deal from
- 15:04this person, right? Do you want to say
- 15:05yes or no to this studio collaboration,
- 15:08and so on and so forth. And in my five
- 15:10meetings, there is a brief here, and I
- 15:12can look up and something suspicious.
- 15:14One of our emails was an IT help desk
- 15:15urgent email, and this looks like
- 15:17something fishy. So, it marked it out,
- 15:19right? There's a something suspicious
- 15:20here. So, why do I have to do this
- 15:21myself? What I should have is it should
- 15:23be done automatically. So, I'm going to
- 15:25ask it, right? I can say, "We should
- 15:28schedule this for tomorrow morning at
- 15:317:00 a.m." Again, let's check out what
- 15:33it's doing, okay? It's creating a
- 15:35scheduled task for morning brief at 7:00
- 15:36a.m., okay? And you can see the details
- 15:38here that it wrote down. Guys, this is
- 15:40all about context, okay? It knows the
- 15:42context that you know. We have the
- 15:43Google calendar, use the demo emails,
- 15:45okay? And then we have Notion, where
- 15:47there's a Monday scrum under YouTube
- 15:50folder and everything. Without all these
- 15:51context and without all these connectors
- 15:53we talked about, your agent won't be as
- 15:55smart as this. They will have no idea
- 15:56what's going on. So, it's really, really
- 15:58important. So, I think for any agent to
- 16:01be a chief of staff, context is
- 16:02everything. Connecting to these
- 16:04connectors and MCPs are everything. So,
- 16:06now it says that we have scheduled this
- 16:08as a task at 7:00 a.m., okay? I guess
- 16:10what we can do is that we can click on
- 16:12scheduled here. You can see that there's
- 16:14a morning brief at 7:00 a.m. I think
- 16:16this was a previous one I made. This was
- 16:17a new one I just made, okay? And you can
- 16:19see the details here. They have wrote
- 16:21the instructions for me. Uh, the scope,
- 16:23like what exactly are the tools that you
- 16:25should look at, and what are some of the
- 16:26safety things that you should check,
- 16:28okay? And the page should be, uh, in
- 16:30this following order. So, it will run on
- 16:33weekdays at every 7:00 7:00 a.m., okay?
- 16:35Uh, but maybe I just want to run it now,
- 16:37right? I can just click on run now. And
- 16:38you will see this task running right
- 16:39now. Okay, but what happened tomorrow
- 16:41morning at 7:00 a.m. on my laptop?
- 16:42Because now it's doing it for me, okay?
- 16:44I just click into it, and you can see
- 16:46exactly what it did, all right? So, you
- 16:47see that this is the context it just
- 16:49injected here from, um, the scheduled
- 16:51task, and it's using the exact same
- 16:53tools that we did, all right? It's
- 16:54reading the Gmail, we're getting every
- 16:56email thread. Um, now reading the Google
- 16:58Calendar. I know some people are going
- 16:59to ask, "What about the security? What
- 17:00about these issues?" My answer is, I
- 17:02don't know. I think people will just
- 17:04eventually get used to it, and, uh,
- 17:06people don't care as much, uh, compared
- 17:09to maybe a year or two ago, where
- 17:11everyone's like, "I'm not going to share
- 17:13my emails with an AI." And, um, now I
- 17:15feel, at least in the startup world, or,
- 17:18uh, among a lot of the engineers, people
- 17:20are just giving it anyways. And I'm sure
- 17:21a lot of product managers do that, too.
- 17:23And agency owners, business owners,
- 17:25people are doing the same things. So,
- 17:26you can see that we got this report
- 17:28already, and, uh, I would call that a
- 17:31success, okay? So, that was the first
- 17:32task. You can plan for tomorrow, and we
- 17:34used a lot of these tools. So, what's
- 17:36next? We kind of already covered some of
- 17:38them, okay? So, for the inbox, we're
- 17:40going to draft some replies to these
- 17:41emails, uh, for Dana, for Mark and all
- 17:44these people, and then eventually they
- 17:46should be a draft instead of sending
- 17:47anything, so that we are the ones who
- 17:49are making the decision. So, all you got
- 17:51to do is to write a task to your agent
- 17:54who has all the context, and ask her to
- 17:56do the job for you. So, he searched for
- 17:58the tools. He's using Notion. Uh, it was
- 18:01checking the handbook again. It's pretty
- 18:03slow sometimes. So, he just finished
- 18:05writing all the drafts. So, if I just
- 18:07come back to my Gmail, and, uh, you can
- 18:09see all these emails have a little draft
- 18:11here, right? But if we click into
- 18:12drafts, we can see the details here.
- 18:14Yep. Thanks for the note. A pilot is how
- 18:16we usually start, blah, blah, blah. You
- 18:18can just send it. Uh, but these are fake
- 18:20accounts, so they won't exist, okay?
- 18:22But, maybe Dana, who's this fake venture
- 18:24capital, if I click on send, it will be
- 18:26sent here as done. So, I will consider
- 18:29this as done. So, you might be wondering
- 18:30what's the difference between just
- 18:31sending emails using agents. The
- 18:33difference is that your agent now has
- 18:35full context on your company. It knows
- 18:37everything about the contacts, about the
- 18:39CRMs, about the previous records of
- 18:41communications. And when it now writes
- 18:43emails, it's just so much faster. You
- 18:45you don't really need to buy some of
- 18:46those tools out there to help you reply
- 18:49to your clients. All you need to do is
- 18:50kind of, you know, schedule task here
- 18:52and ask it to reply questions. You can
- 18:55do it yourself. This is your AI chief of
- 18:56staff. So, the third one is prepare for
- 18:59your meetings, okay? Because tomorrow we
- 19:01have this meeting with this venture
- 19:03capital firm called Redwood Peak
- 19:05Catch-up. And what we want to do is that
- 19:07we want to prepare for that catch-up
- 19:08with Dana. And we want to use her email,
- 19:10the handbook on Waku Memory because on
- 19:12Waku Memory, I'll assume that you have
- 19:14connected with Waku Memory for quite
- 19:16some time, so it knows about you and
- 19:18knows about knows about the project. It
- 19:20will create a one-pager later telling me
- 19:22who she is, what she asked, what I
- 19:24promised, three questions I should ask
- 19:25her. Okay, let's just ask this question
- 19:27to our chief of staff. Let me paste that
- 19:28in.
- 19:30Okay, so you see that it's calling Waku
- 19:32Memories here, doing some memory calls,
- 19:34pulling the relevant information.
- 19:36Meanwhile, I will show you if you come
- 19:38to waku.one, which is a portable memory
- 19:42layer for agents and teams to
- 19:43collaborate together. We started as an
- 19:45open source product and we hit more than
- 19:472,000 stars in a couple months' time.
- 19:50Feel free to try it out. And the way to
- 19:52set up is really simple. You just come
- 19:54to this main website and click on copy
- 19:56of this prompt and just paste it into
- 19:58your cloud code and it will set things
- 19:59up for you. Or you can do the MCP setup
- 20:01that I showed you earlier, okay? And
- 20:03what I'm trying to show you is that
- 20:04after you sign in, you can click into
- 20:06the console and within memories, I've
- 20:08already got a chief of staff folder here
- 20:10with something related to Dana. And this
- 20:12was a previous memory that I had with
- 20:13her, okay? And basically, this partner
- 20:16asked me for evidence of company usage
- 20:18and Shawn promised some usage examples,
- 20:21okay? So, I think in the next
- 20:24preparation for the meeting, we should
- 20:26definitely prepare for that because
- 20:26that's where
- 20:27that was her ask. If you scroll down to
- 20:30Dana, you can see that the preparation
- 20:32was telling us about who she was, what
- 20:35she asked for, and she was asking for
- 20:37some evidence that people will come back
- 20:39to our product every week, and what we
- 20:41promised is three usage examples. So,
- 20:43this has prepared me for three questions
- 20:45to ask her, something that not checked,
- 20:48and the after call half I don't know.
- 20:50These are just the stuff that the clock
- 20:52came up to me. You should put in the
- 20:53thoughts to really like think about how
- 20:55they should prepare the meetings for
- 20:57you. Feel free to edit something in your
- 20:59Notion pages, your handbooks. You can
- 21:02create another page calling it, "Oh,
- 21:03this is how we should reply to VCs and
- 21:05stuff like that." I guess we'll pass
- 21:07here. The fourth stage is we want a
- 21:09one-page for people and agents to share
- 21:11decisions and memories together. So, the
- 21:13work we're going to do is basically we
- 21:16want to draft the team's team update,
- 21:18what was shipped, was blocked, decisions
- 21:20needed, use only the scrum page and Waku
- 21:22memory, and then list of decisions as
- 21:24separate one-line items I can record,
- 21:26okay? Again, I'll just paste that in.
- 21:28It's searching memories on Waku, got
- 21:30everything connected. It's literally
- 21:31just searching the information. For
- 21:33example, you can see that this one saw
- 21:34that Waku reached 2,000 stars in about 3
- 21:37months with no paid ads, a hosted
- 21:39product with paying users shipped within
- 21:40weeks, and we have about 135,000
- 21:43audience across different platforms. And
- 21:45these memories are just going to be
- 21:47stored there and shared across teams,
- 21:49okay? To be honest, if you're not me,
- 21:50you probably won't see what is really
- 21:52going on here. But if this is your own
- 21:54context, it would make a lot more
- 21:56clearer sense. I remember for instance
- 21:58this company called Meridy. The
- 21:59decisions needed is that we want to
- 22:00confirm the proposal from their pilot,
- 22:02and then there's a price list in our
- 22:04memory saying that our starter and team
- 22:06plans are $2,000 and $5,000. So, these
- 22:09are the stuff that you want your chief
- 22:11of staff to be able to fetch
- 22:13immediately, okay? And these decisions
- 22:15should be shared across your teammates,
- 22:17okay? We do have a feature called Waku
- 22:19team space where you can create a team
- 22:21for your companies, and you can share a
- 22:23lot of these decisions by using this
- 22:25feature called board. Okay, what you got
- 22:27to do is that you can just say, "Hey,
- 22:29tell Sasha about our prep on Dana from
- 22:33Waku board." So, again, it's using all
- 22:35these connectors from the team shared
- 22:37memory and decision layer and then
- 22:39started posting something that uh we
- 22:41have just prepared by our chief of staff
- 22:43all above here. It will ping the message
- 22:44to Sasha. Okay, I'll just pull in the
- 22:46email from Sasha right away because it
- 22:48should have received an email. And I
- 22:50just went to Sasha's email and then
- 22:52check the board and I can see there is a
- 22:54information sent from Shawn about Dana.
- 22:56Okay, this is the board message. What
- 22:58happens is that we also receive an email
- 23:00right here um from [email protected] where
- 23:03my teammate Shawn was telling me me as
- 23:05in Sasha now is that we have a
- 23:07preparation for Redwood Peak uh catch-up
- 23:09with Dana from this venture capital.
- 23:11I'll just be able to see uh you know,
- 23:13these uh team decisions from our
- 23:14teammate. Uh last but not least, from
- 23:16send a quote to the CRM. Okay, there was
- 23:18a person called Tessa which we haven't
- 23:20paid much attention to yet. So, Tessa
- 23:22was asking us that their product ops
- 23:24team at Meridian is looking for a shared
- 23:26memory across their agents and they were
- 23:28asking us for a quote, right? And we
- 23:29just send them an email over here. Okay,
- 23:32but the tasks was that we should read
- 23:33this email, write a one-page pilot
- 23:35proposal for them, and save it as a PDF.
- 23:38We should not invent any price or date
- 23:41or term because the price and date and
- 23:43term are all in our handbook. All right,
- 23:45there's a price list here for how much
- 23:47we're charging these clients. Okay, so
- 23:49then you should draft a reply to Tessa.
- 23:51Okay, we mistakenly send a reply over
- 23:53here because earlier we were just like
- 23:55in a rush replying to emails, but now we
- 23:57should do it properly. And here, we're
- 23:59probably going to use HubSpot. I'm going
- 24:00to copy this prompt and I'm going to ask
- 24:02our chief of staff to work on it. Let's
- 24:04see what happens.
- 24:06Okay, very quickly, so it was trying to
- 24:08fetch information from Notion and
- 24:10failed. And then it was checking some of
- 24:12the PDF libraries and project folders.
- 24:15Fetching from Notion again, found a tool
- 24:18to create PDFs. I'll just leave it to
- 24:21work. Cool, it created this PDF for me
- 24:23and honestly it doesn't look very nice.
- 24:26So, I'll just say, "Can we make it look
- 24:29nicer by following the same design
- 24:32system
- 24:34as wakku.one?" So, again, because we
- 24:36have this shared team space, it kind of
- 24:38remembers that we have certain rules
- 24:40within our team of how we should do
- 24:42things, and uh it's just using these uh
- 24:45these files. You remember how it should
- 24:47function. Built wakku.one style proposal
- 24:49and render to PDF.
- 24:51Create a screenshot here. Oh, so much
- 24:53nicer. I love this a lot more. Oh, wow.
- 24:56Look at this. It looks nice, doesn't it?
- 24:57Yeah, very nice. It's prepared for
- 24:59Tessa. It's a pilot proposal for
- 25:01Meridian Streaming. It's a team pilot.
- 25:03It's shared memory and board for teams
- 25:05and AR agents for the product subs
- 25:06teams, about 40 people, and we're
- 25:08quoting a $5,000 price for 4 weeks.
- 25:11Monthly upkeep is $500, and payment
- 25:13should be net 14 and stuff. Very cool.
- 25:15So, let's check out the drafts.
- 25:17Uh we can see that right here. And
- 25:18somehow it's not attaching it, so we
- 25:20should just tell it uh hey, we should
- 25:22attach it to the draft of the email and
- 25:24then send it over.
- 25:25Oh, this is using the browser for me,
- 25:27and it just attached this proposal. This
- 25:29is a cursor that Clocko is using. What
- 25:31Clocko is doing is just taking
- 25:33screenshots real time and use the
- 25:34computer for me. I think it's sent. So,
- 25:37let's check. Look at this. So, they
- 25:38applied to Tessa for me and then created
- 25:41this quote for us. And I'm just going to
- 25:42say, "Hey, we should update that to uh
- 25:45HubSpot, too. And uh can you turn this
- 25:47into a skill in the future for how we're
- 25:49going to reply to these quote requests
- 25:52and CRM update?" Thanks.
- 25:55Yep, now it's using the HubSpot CRM. It
- 25:57found out about Tessa, Meridian
- 25:59Streaming. It's really, really
- 26:00convenient. Again, remember what we say?
- 26:02It's all about context. It has the
- 26:04context about who the client is, where
- 26:07they are on the CRM, where some of the
- 26:08emails that we have replied. And given
- 26:10all the context regarding the handbook,
- 26:12the records, the memories, and all these
- 26:13connected tools, you will be able to
- 26:15generate this PDF quote for this client
- 26:19and then get things done for us. And
- 26:20then eventually I can ask it to create a
- 26:22skill for me. Okay, you see? Look at
- 26:23this skill. The skill is basically just
- 26:25checking all the contacts we just said,
- 26:27and then there's a clear instruction
- 26:29here. Remember, this is a procedural
- 26:30memory. Read before writing, pick the
- 26:32plan, invent nothing, one page proposal
- 26:34PDF, draft the reply, send on Shawn's
- 26:37yes, and HubSpot update, and eventually
- 26:39have a report, okay?
- 26:41Add a note to CRM, too. Now, it's adding
- 26:43it to the CRM objects. Let's come back
- 26:45to Meridian, and let's take a look
- 26:47again.
- 26:48You can see that in the recent activity,
- 26:49something has been updated. There's a
- 26:51new note here saying that there's a team
- 26:53pilot proposal sent, all right, for
- 26:55$5,000. There's a link here. I click
- 26:57into it.
- 26:58It's the same thing. Okay, that's pretty
- 27:00much it. We have finished the five jobs
- 27:02for this Chief of Staff.
- 27:04Well, I'm really curious about what you
- 27:05think, and I I believe if you're a
- 27:06technical, please bear with us that this
- 27:09video is about showing people, showing
- 27:11product leaders and business owners how
- 27:13they can use Claude code as their Chief
- 27:15of Staff from end to end. I hope this
- 27:16was helpful for a lot of you guys, and
- 27:19uh let me know what else you want to
- 27:20watch, because this was a pretty
- 27:22introductory end to end workflow for how
- 27:24you can set up the Claude code all the
- 27:26way to connect with all these brains,
- 27:28and set up the work, set up the skills,
- 27:30and set up all these connectors in order
- 27:32for things to be functional. And even
- 27:34you can schedule things, and make sure
- 27:35that they tell you things. One more
- 27:37thing that Claude does these days is
- 27:39that we have a Claude tag, which is for
- 27:42teams to collaborate at the same time.
- 27:43So, I can just ask Claude tag, and I can
- 27:45say, "Hey, what's up with Tessa's deal?"
- 27:48at Claude. You're pretty much just
- 27:50running it here over here in Slack using
- 27:52Claude tag.
- 27:53Cool. Okay, it can allow you to read it.
- 27:56Now, it's checking the CRM for the deal,
- 27:57okay? Check the recent emails, send the
- 27:59answer. So, Claude is trying to be this
- 28:02AI employee within Slack for us, and I'm
- 28:05looking forward to how much it will
- 28:06evolve. So far, I think this is what's
- 28:08been helpful for us to run the company
- 28:10using Claude as our sort of AI Chief of
- 28:12Staff to do a lot of things that are
- 28:14non-technical related, but we have used
- 28:16enough context to to simulate how a
- 28:18business are supposed to use Claude as
- 28:20the AI Chief of Staff. And if you're
- 28:22interested in the future, we're happy to
- 28:23show you more things related to
- 28:25technical people, how how our engineers
- 28:27or technical teams are going to use as
- 28:29their executive assistant or Chief of
- 28:31Staff, or however you want to call it,
- 28:33okay? And uh you can see that it got the
- 28:35contacts and uh it's really amazing that
- 28:38you have connected your cloud to all the
- 28:39contacts once, it just knows everything
- 28:41everywhere. Again, I hope this is
- 28:43helpful and tell me if you have any
- 28:44questions. Feel free to try out our
- 28:46tools, try out Claude, and join our
- 28:49community. Uh we have all the links
- 28:51everywhere. Uh feel free to click into
- 28:52them and um I will see you next time.
- 28:54Thank
About this transcript
This page contains the full transcript of You Can Learn AI Chief Of Staff With Claude Code In 29 Min | Founders, PMs, Skills by Sean‘s AI Stories, generated from the public captions YouTube serves with the video. The transcript has 6,385 words across 891 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.