Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Isar Meitis (00:00):
Hello and welcome
to the Leveraging AI Podcast,
(00:03):
the podcast that sharespractical, ethical ways to
leverage AI to improveefficiency, grow your business,
and advance your career.
This is Isar Metis, your host,and we have a fascinating
episode and topic for you today.
So I'm gonna start with aquestion.
What is the most importantresource you have as.
Yourself as a head of a companyor a business?
Well, some would say money, someother would think human capital.
(00:26):
But the reality is there's onlyone resource that we cannot get
more of, and that is time, whichbasically means the better you
are with time management, eitherfor yourself or for other people
in your company, you can achievea lot more without paying for
additional resources.
Now, there's ample of researchshowing that most of us people
(00:47):
and business managers and CEOsof large companies and what,
regardless of our positions areusually pretty bad at time
management, which is really sad.
We get distracted very easily.
We don't necessarily haveagendas for things and so on.
Now, the good news are is thatAI can be an incredible asset,
helping us with managing ourtime better.
(01:09):
So if you feel that you canbenefit from better.
Time management.
You will find incredible valuein this episode.
Now, as always, we're going toshow you exactly how to do this
from a very practicalperspective.
So by the end of the episode,you will have the exact tools,
prompt steps, processes that youneed to use and implement in
(01:29):
order to better manage your timewith the assistance of AI tools.
By the way, if you don't thinkyou need help with time
management, please drop me anote and tell me how you do this
because you're probably the besttime manager in the world.
I've never met a person who'sbeen really effective, uh, with
managing their time.
But now to our topic, our guesttoday, Matt Martin, is the
co-founder and the CEO.
(01:50):
Of Clockwise, which is a companythat has developed software
products to help people bettermanage their time.
Now, this name by itself isbrilliant.
Calling the company clockwisefor a product that actually
helps you manage your time.
But in addition to coming upwith great names, he's also
extremely knowledgeable in howto better manage our time using
(02:10):
software applications because hehas been doing this for nine
years now.
So clockwise has been around waybefore the AI craze, and now
it's on.
Obviously upgraded a lot of whatit's doing with AI capabilities
in connection with AI tools thatwe use in our day to day.
So in this episode, Matt isgoing to teach us exactly how to
use connectors such as calendarconnectors, connectors and email
(02:34):
connectors, as well as theclockwise software into large
language models such as Claudeand chat pt, using either the
built-in connectors in them, orusing MCP connectors, which by
the way, just knowing thesethings, how to use connectors
and how to connect MCP serversare incredibly powerful skills
that are worth learningregardless of the time
(02:54):
management benefit.
But we are also going to go overspecific use cases that can
dramatically improve your timemanagement using AI in
connection with these tools.
And since this is.
Extremely valuable and I'm sureyou would agree.
I am really excited to welcomeback to the show.
Matt, welcome to Leveraging ai.
Matt Martin (03:58):
Sorry, what a great
intro.
Thanks so much for, uh, havingme on and um, this is a topic
obviously that's near and dearto my heart.
Uh, two things off the, off thejump though.
So one is if anybody out thereis responding to Issa's Prompt
to send them an email if youdon't need help, time
management, shoot that my waytoo.
I'd love to hear.
And second is, uh.
(04:19):
I wish I were as good at namingas clockwise would indicate.
I have to, I have to give thanksto my co-founder for that one.
I'm so bad with names.
One of the toughest thing incomputer science naming things.
Isar Meitis (04:30):
Yeah, a hundred
percent.
But, but anyways, I, I, I'mreally glad we're here.
I really think it's a very, veryimportant topic, right?
I didn't say that in a joke.
It's literally the only asset wecannot get more of.
Uh, there are multiple books onhow to manage our time better.
There's different systems,processes, and so on.
Uh, but, but if we can connectit to tools that we're using
anyway in our day to day, thenit's a huge benefit.
(04:51):
So now it's the stage is yours.
Show us, show us the magic, andshow us how we can do this
better.
Let's do it.
Matt Martin (04:58):
And I, I love the
format of this where we just
kind of get in and get dirty,uh, to show people exactly how
to do this stuff.
Um, so let me just share myscreen for a moment.
I've, I've queued up a fewthings that I want to take.
Your audience through, um, soclockwise, as you noted, we've
been doing this for a while.
Um, and one of the advantages wehave is we have a massive corpus
(05:19):
of data.
Um, so we connect to some of thebest companies on the face of
the planet, places like Uber,Netflix, Atlassian, and help
power their schedules.
And we've taken kind of thebrain and hooked it up to a
model context protocol server sothat you can use it instead of
cloud, uh, or, or chatt BT as itwere.
Um, the first thing though thatI I'd like to go through is just
(05:40):
if you don't connect clockwise,you're gonna be missing out on
some power, but there's stilltons that you can do in the time
management space when it comesto MCP servers, time management,
and I'll give a shout out tothis How to Geek article.
Um, if you just Google ClaudeManaged Calendar, how to Geek,
you'll find this article and,uh.
(06:01):
Diba car.
I like this article.
It's a little click Beatty, butI like it because he's gone so
in depth with what you can do,and it, it illustrates both the
power of these tools and some ofthe limitations.
So I'm gonna just take thisprompt and kind of show you a
really in depth usage of Claudeand Calendar.
(06:21):
Awesome.
So, scooting over to Claudehere.
Uh, I'm gonna run over a couplethings with, with everyone if
you're not familiar.
This is claude.ai.
Um, I'm sure your audience isfamiliar with that much.
I, I, I hope so.
Yeah.
Or else you haven't done yourjob right as Sarah.
Yes, exactly.
I, I fail miserably if I don't.
(06:41):
Um, and.
Under here, you have tools andone of the ones that they have
is, so I'm gonna pause you
Isar Meitis (06:46):
just for one
Matt Martin (06:46):
second for
Isar Meitis (06:47):
obvious.
Those of you who are notwatching, uh, we're looking at
Claude and at the bottom, at theprompt section of Claude.
By the way, it's very similar inall the tools.
There's this little, kinda likeTuggle buttons, which is
basically, uh, opens a wholemenu.
And in that menu you havemultiple things that you can do.
One of the segments there isconnectors, which are different
things you can actually connectto Claude.
It's a very similar setup, uh,in Chate as well, which we may
(07:09):
or may not, uh, see afterwards.
Uh, and then now I'll let youcontinue from here.
By the way, the other thing thatwe did is Matt took the prompt
from that article that herecommended.
We're gonna put a show, uh, alink to that in the show notes
so you can jump straight to thatfrom your phone, uh, or your
computer or whatever it is thatyou're listening, uh, to the
podcast.
Uh, and you'll be able to seethe prompt as well.
And we're gonna read a shortsnippet out of it afterwards,
but continue now that we'reseeing the menu with the
(07:30):
different connectors.
Matt Martin (07:31):
Great.
Yes.
And, and good reminder that noteverybody's watching this on
YouTube, so.
Uh, Claude has and Chate does aswell, the ability to read from
your calendar, so you canconnect your Google calendar.
Unfortunately, they don't haveOutlook yet, um, but you can
read from your Google calendarand it'll, it'll be able to suck
down that information and giveyou some insights based off of
it.
So I have toggled on the GoogleCalendar connector.
(07:52):
Um, you just OAuth into yourGoogle account, and I've copied
and paste the prompt, um, thatwas provided from the article.
And there are a few things herethat I'll, I'll just run through
though, uh, for the audience.
So it's gonna, it's asking for,it's a very highly structured
prompt.
Um, and I like these kinds.
When you, when you see peopleare really into ai, really into
(08:12):
MCP, you see people who give theLM a tremendous amount of
guidance on what they want outof it.
And you really do get betterresults by doing that.
Isar Meitis (08:19):
And I'll pause you
just one second before we dive
into the prompt.
Yeah.
Uh, for those of you who are newto the show or hasn't been
listening to all the episodes,what is an MCP?
Uh, an MCP is basically a USBconnector for third party
software into AI tools.
So about roughly a year ago, uh,maybe a little less, uh,
anthropic themselves.
So the company behind Claudecame up with a new protocol that
(08:42):
now everybody's using thatbasically said, instead of.
Every company in the world nowdeveloping connectors of
different pieces of softwareinto large language models or
agents and so on.
What if we develop a standardprotocol that once one person or
the company itself creates theconnector, you can plug it in to
your AI environment in threeminutes, which again, we're
(09:04):
gonna see afterwards how to do.
So that became an extremelypowerful capability to enhance
the abilities of, uh, these AItools, chatbots agents, and so
on, because they can now connecteither just as a read only or as
read and write, depending howthe MCP connector was developed
with whatever third partysoftware in this particular
case, your Google calendar.
(09:25):
But this could be as we'll see,many, many other different
things.
So closing parentheses back toyou, Matt, to talk about the
prompt.
Matt Martin (09:31):
Yeah, no, and it's,
it's, uh, a great, uh, thing to
explain a little bit because CPSare kind of like.
Plugins, um, like you said,they're the USB connector and
it, as we'll see, the ecosystemis still kind of developing on
how exactly to connect them withthe level of support is, but
very, very powerful if you wanta, to allow the LM access to,
(09:52):
you know, third party tools.
Um, and so here, uh, theconnector is to calendar and it,
we're just asking for a fewdifferent things here.
Um, we're asking for our initialcalendar overview, time
analysis, schedule, insights,and then we're gonna ask it to
help us plan out our, uh, week,uh, with event details and
possibly even makemodifications.
(10:12):
So having entered that, what'samazing about this is that, um,
you know, Claude will use thatconnector to search all of my
calendar events and it'll gimmea rundown of the next 30 days of
stuff that I have going on.
And, you know, look like you canget that sort of stuff out of
your calendar in, I would arguea much more consumable fashion
than just text.
But what it can do next with allthat information is to then, uh,
(10:35):
identify insights.
So some of the patterns that youhave, some of the conflicts that
you have when you're busier,when you're lighter, time zone
considerations, specialattributes, and then the real
power is next steps.
So now that it's analyze allthat, it can help, you know,
resolve triple bookings, managetime zone challenges, improve
work life balance.
(10:55):
Now, I, I could dive into thisand start to have a conversation
with Claude.
Um, but one of the limitationsabout current calendar, uh,
connectivity, either in Claudeor chat GBT, is that it's read
only.
Um, so it can't actually, youknow, it will tell me things
that it thinks it can resolve,but, um, it has some pretty
severe limitations.
One is that not only is it readonly such that it can't write or
(11:19):
schedule anything or moveanything, but the other, um,
severe limitation, which I don'tthink people appreciate a lot of
the time is that.
You know, if it's trying to helpme resolve a double booking, it
cannot see the other attendeescalendars.
So its insights are gonna belimited to just what it sees on
my calendar, which is, you know,it's helpful, but it's not, it's
not everything.
But, you know, one of the, oneof the prompts that I do love,
(11:39):
which just one vanilla Claude,is with calendar connected is to
send a message to it saying,Hey, could you take a peek at my
schedule for the next two weeksand find events that I've
organized that don't have anagenda?
(12:01):
And what's great about this isthat Claude can not only
identify that, but then it'svery good at helping me
construct that agenda if I wantto.
So it can identify those events.
But then, you know, with alittle bit of information, um,
it can actually help me.
Construct that agenda for it.
Isar Meitis (12:20):
So I, I wanna add 2
cents to what you said, just to
expand on the capabilities ofthese tools and how these things
become a lot more interesting.
One of the things that you canbuild in Claude and in Chachi
pt, it doesn't matter, isrepetitive information that has
background information to it.
So in, in Claude it's calledProject in Chachi pt, it's
called Custom GPTs.
And you can create a cloudproject in this particular case
(12:41):
that will have the way you craftagendas.
So here's the template foragendas.
We must have all these differenttopics.
Uh, so it needs to say, how longis the meeting?
Who's gonna speak, who'sattending?
What's the goal?
Uh, what are the differentthings we're gonna cover?
How much time is for like,whatever it is that you want?
That will be the standardagenda.
The other thing that you candefine is what kind of agendas
you want for what types ofmeetings, depending on the
(13:02):
people you're meeting with, likewhatever additional information
you can provide.
And then the kind of answersyou're gonna get out of Claude
when you ask these kind ofquestions is gonna be
significantly more aligned withwhat you and your company's
expecting to have in thosemeetings because we'll follow
your rules and guidelinesbecause you place them in this
project before you started, uh,the conversation.
So just another food for thoughtand another expansion of what
(13:24):
you can do with these kind oftools.
Matt Martin (13:26):
Killer tip.
Killer tip.
And I'm not gonna belabor thisuse case too much, but you know,
we see that there are a varietyof meetings that don't have
agendas that's identified.
And I could use, you know, weall know how great LMS are at
generating text.
I could use them to dynamicallycreate that agenda.
Um, with a little bit more work.
I could also connect.
Uh, you know, either GoogleDrive or notion, um, using that
(13:47):
same MCP connector to pull outany sort of associated agendas
or create them in those areas,which is what I love to do.
Um, I'm not gonna do that'causeI actually, I'm gonna switch
gears here a little bit, uh, andshow folks exactly how to
connect, uh, external connector,um,'cause that's kind of the
next step of the process.
So let's pop over to a new tabhere.
(14:08):
And so I have just a freshClaude instance, cla.ai,
anthropics, uh, chat model.
And here under that same togglewhere you have search and tools,
which is right underneath theprompt.
You can open that.
And then there's a, a littleplus button at the bottom to add
connectors, and this opens upClaudes, well, rather
anthropics, um, kind of list ofpre-approved connectors and it's
(14:30):
growing all the time.
Um, there are some really,really cool things you can do
here with things like Canva orDatabricks or fireflies, you
know, pull in your recordings.
I'm of course going to go toclockwise, um, and connect
clockwise.
And so we're featured herealphabetically.
Um, you can connect to clockwiseand I'll just, it'll pop open a
(14:52):
new window, sign me intoclockwise, and then I'm good to
go.
Um, and actually this is aprompt from a different window,
so we'll close.
We will, we'll reset this for asecond.
And so now if I open up thatsame dropdown menu, and this is
gonna be the case for any, uh,connector that you add.
And by the way, connectors are.
Philanthropics friendlier wordfor an MCP server, um, MCP
(15:15):
server, and
Isar Meitis (15:15):
I'll say, I'll say
2 cents.
You can connect.
So there's, I dunno how manyAnthropic already have approved.
So there's 20, 30 of them Ithink ish, uh, that are tools
that you use, that most of ususe all the time.
But you can connect literallyany MCP server that is out there
by going to the MCP server.
Getting the code that it needsin order to connect and dropping
(15:36):
it into Claude's backend.
It sounds really complex.
Watching a YouTube video forabout three minutes, you can
literally do the processyourself.
Uh, and if you don't know how todo this, which I'm one of the
people to not know how to dothis, I ask Claude to do it for
me.
I literally drop the code in clain Claude.
I said, this is what I, what Iput in your config file.
Uh, write the code for me.
(15:57):
It will write the code.
I will paste it back into theconfig file.
Uh, and then it will add anotherone of these connectors, even
if, even if it's not showing upby default in that plus button
of approved connectors.
So even you have a connectorthat is not approved, uh, you
can still add it in the back anddo really, really magical things
with this.
But again, closing parenthesesagain, back to this.
So now we're looking at theclockwise connector.
Matt Martin (16:20):
Yes.
And, uh, I, I, uh, I love thedigression there because there
are so many great tools thataren't featured in their
connector list.
Um, if you, as you're addingtools, you'll see that you can
kind of dive one level deeper inthis dropdown and you can see
all the tools that a specificconnector has given you access
to.
So, um, in MCP world, what a,uh, MCP server exposes are a set
(16:44):
of tools that give specificabilities to the lm.
Um, and so the clockwise MCPserver, for example, uh, one of
the tools that it has is tocreate a meeting proposal that's
one of 16 tools.
If you connect, um, notion, oneof the tools is going to be
create Notion page, um, or readNotion Doc.
And these specific tools arewhat give the underlying LLM the
(17:07):
ability to add additional poweras additional functionality.
A tip that I'll give folks,especially as you're playing in
MCP world, um, as you addadditional connectors slash CP
servers, um, it's, it's best tostart playing actively with
these toggles for better orworse, as you add a lot of them,
(17:29):
the LM can start to get a littlebit bogged down in which tool to
call into and get a little bitconfused.
Um, especially as you're playingand getting confident with
these, it's best to say, youknow, if you're using clockwise
to switch off calendar search,'cause those, those can
conflict.
Um, if you're, you know, if youboth use Asana and Jira, um, you
know, flip off Asana and useJira, or flip off Jira and use
(17:52):
Asana.
'cause these, they can kind ofcollide and they can also make
the LM trip up.
Isar Meitis (17:56):
Yeah.
So to explain two seconds, andit's a great point.
Uh, you don't call the tool asyou're writing the prompt,
you're literally writing theprompt in regular English.
So if I'll use your Asana andJira example, if you ask it, oh,
look for all the tasks that X itknows on its own that it has an
MCP connector, and then it willuse the connector to find the
tasks that are assigned to aspecific person in a specific
(18:18):
date range.
And if there's two tools that dothat, it will potentially get
confused or get some of it fromone, one, some of it from the
other.
So it's just, uh, you, you'rebasically opening yourself to
more errors.
And I agree with you a hundredpercent if you have things that
might be contradicting, uh, justturn the toggle button off on
one, work on the other or viceversa.
Matt Martin (18:36):
I've also found,
uh, in, sorry if you found
something differently, uh,definitely corrects me here.
But I've also found that evenwhen you, it definitely when you
have overlap in domains, it'sgood to flip them off because as
you said, they can collide.
But even if you havenon-overlapping domains, I find
that it's helpful just to selectthe tools that I'm working with.
And I don't, I don't alwaysremember this, but it, I find
(18:58):
that, um, Claude, she, pt, otherLMS can get bogged down just the
more tools that has access to,even if they don't conflict.
Isar Meitis (19:05):
Yeah, it makes
sense.
Think about it.
They gotta kind of go throughthe list of all the different
functions in each and every oneof the tools with every prompt
that you write, uh, you, youwanna minimize that just to, uh,
let it run smoother.
Matt Martin (19:15):
Yeah.
Yeah.
I think it's one of the areasthat they're still kind of
developing the underlyingsystems.
So I'm, now that we've doneClaude, I'm actually gonna
switch to chat.
Gcpt.
I know, you know, Che, GBDA, uh,just from a pure numbers
perspective, is much more widelyused.
Unfortunately, Chatt B t'ssupport for MCP servers is a
little bit lagging behindAnthropics.
Um, that's being very gentlewith open ai.
(19:39):
Well, I, you know, look, I knowthey'll get there.
They're obviously building intoit more, and it was Anthropics
spec.
Um, but yeah, it's a little bit,it's a, it's a little bit
disappointing that they're, theydon't have it full featured yet.
Um, so this is kind of funky.
I'm gonna, I'm gonna walk peoplehow to through how to do this
and, um, oh, my, my annotatebutton for Zoom is actually in
the place that I'm trying toclick here.
(20:01):
It's annoying.
Um, so if you're in chatt PT inthe bottom left, you'll see your
profile, and that's where youget to settings.
Um, so you're gonna go intosettings, and I, and you do need
for this in Chatt PT, you needeither a, a Chatt PT, uh, I
forget what they call theirbusiness accounts, either
business or enterprise.
But you need a business.
Yeah.
There's,
Isar Meitis (20:19):
there's a thing.
Plus then teams, thenenterprise.
Yeah.
And I think all three of themwill have access to
Matt Martin (20:25):
that.
Yeah, I think that's right.
I think that for, on theindividual side, you just need
plus.
Yeah.
Plus or
Isar Meitis (20:30):
what's the, uh,
plus or pro.
And then on the, on the businessside, you need either teams or
enterprise.
Yep.
Yep.
It's, and, and I think
Matt Martin (20:38):
that, you know,
their pricing rubrics are
getting more and morecomplicated, but I think that's
basically everything exceptfree.
If you're on the free tier, youwon't see this.
Yeah.
Um, but otherwise you should seeit.
So you go into settings, you gointo connectors, so they're also
calling MCP servers, connectors,and you'll see under the
connectors tab.
So I'm in settings if you're notfollowing along visually, I'm in
(20:58):
settings and there's a list ofconnectors that are available.
Now the thing about chat CBTSconnectors is their search only.
Um.
And, and they, they can't reallydo much.
And we ran into this when wewere developing clockwise.
They really don't have multipletools right now.
They just have search.
So, you know, if I connectGoogle Drive, it can search
(21:20):
through MRI documents and accessthem, which is really helpful.
But it can't create one.
It can't, you know, add anupdate.
Um, it's just search.
They're obviously moving towardsfull MCP support, uh, because
under this, it's so buried.
You go into connectors and youhave to scroll to the bottom,
make sure you scroll all the waydown.
It's hidden.
You go to advanced settings, andthen you enable developer mode,
(21:41):
which is, says this in beta, andthis is how you get access to
adding a full MCP server.
Now if I go back to that sameconnectors page and I click
create, um, I can create acustom tool.
And, uh, as, as Sarah mentionedearlier, there are so many great
MP servers out.
You will have to go find the URLfor this.
Um, but if you search, you know,for example, clockwise, MCP,
(22:04):
linear, MCP, Atlassian, MCP,you'll find the instructions
pretty quickly.
And so you fill in a few fewterms, uh, here.
Um, and what you really need isthe M-C-P-U-R-L, uh, and you
need to know how toauthenticate.
Ours happens to be OAuth, noteverybody's is, and then you
create it.
And, uh, this is now going toenable the same MCP server
(22:26):
access that I had on quad side,except that it's in chat GPT.
Now, I will also note that chatGT still has some limitations
even in this mode.
First and somewhat annoyingly,you're gonna see a red box to
indicate that you're in devmode.
It always freaks me out.
I'm like, what did I do wrong?
It's just like, I, I have toremember that I'm in dev mode.
Um, the second is that
Isar Meitis (22:46):
when, when, just a
second, when, when Matt's saying
a red box, the, the.
Prompt box just has a redoutline instead of the regular
black slash gray outline that ithas usually.
Matt Martin (22:56):
That's right,
that's right.
And the other, the other thingto note about chat t's uh,
implementation is right nowmemory is disabled when you're
using dev mode and memory iswhat stores, uh, kind of the
past information about yourchats in chat chippie, I find it
(23:17):
to be just insanely, insanelyuseful.
And so this is kind of a blockerfor me.
I, I often have to go toggle offdev mode because memory is so
helpful in of itself.
So I will cop to the fact that.
I, I wanna show people how touse MCP servers in chat GBT, but
I use it almost exclusively inClaude right now just because
chat GT's implementation is soconstrained and we
Isar Meitis (23:36):
are on the same
page.
Yeah.
It's extremely annoying thatchat memory is not working, and
I don't understand if this is atechnical limitation, which
doesn't make any sense to me ora security limitation, which
also doesn't make much sense tome.
Uh, but, but that's thesituation right now, as you
said, like the memory's notworking, like the red circle,
(23:56):
whatever.
No, I don't care.
Matt Martin (23:58):
Yeah.
Isar Meitis (23:59):
But not having
memory available on both sides,
so it doesn't know what happenedand he doesn't remember anything
new is really annoying.
So a little
Matt Martin (24:06):
bit of a nerdy
digression here.
I don't know if you have anyinsights, but I, I've been
wondering why they're a littlebit behind and they have these
limitations as well.
And I think part of the reasonhas to be that.
Tool calling has to be part ofthe model.
And so Anthropic developed thespec.
And so when they were developingClaude and 4.5 sauna just came
out today.
But, but these models are verygood at tool calling, whereas
(24:28):
I've seen that Chet PT fiveisn't quite as good at tool
calling and when you go backinto later models, like anything
trained off of four, four, oh,they're just not very good at
tool calling.
Yeah.
So I wonder if it's modeltraining that they're behind on
potentially.
Okay.
Let's get into the fun stuff.
So we, we went through setup,which, you know, isn't super
fun, but if you can't set it up,you can't do anything.
Uh, so I wanted to make sure torun that through, run through
(24:50):
that with everyone.
So I'm gonna pull up a new, um,uh, a new window for cloud ai.
Again, this is what I prefer forany time that I'm working with,
um, connectors or MCP servers.
I'm gonna check that clockwiseconnected.
So again, opening up that littletoggle underneath the, the
search prompt, uh, clockwise isconnected.
(25:10):
Uh, calendar searchesdisconnected, so the full power
clockwise will be in here.
And I'm just gonna paste in avery simple prompt.
Um, can you find me a time tomeet with June, June, Mike and
Casey next week.
And these are just three folksthat I work with.
Um, I'm just using them plainnames, just like I would with an
assistant.
And I'm gonna fire this off andI'm gonna note a couple things
(25:34):
that, yes, to brag a little bit,this is about clockwise, but,
uh, it really showcases theadvances in MCP server.
And what tool calling enables isthat, you know, Claude by itself
wouldn't know who Juju, Mike andCasey are, but it can use
clockwise to search people thatI attend meetings with,
disambiguate that to find theright people.
(25:55):
Then it can create in kind ofone shot.
A proposal for times that Imight meet with them.
And so what we're seeing in thechat here is that it's using a
tool to find who Juju and Mikeand KC are.
Then it's using that informationto create a meeting proposal.
Um, and then it's coming back tome to say, uh, what some
limitations are on finding thetime with them.
(26:16):
Because again, it knows thosespecific individuals.
So it's saying, you know, Hey,I, I didn't find any times'cause
it looks like everybody's busy.
Um, somebody might be outtaoffice.
Can we have a conversation abouthow to extend this?
And then so I'm gonna say, uh,you know, go ahead and extend
the time range to find somethingthat works.
(26:39):
With my typos.
Um, hopefully it'll understandwhat I'm saying.
And it, again, part of thevirtue of using a system like
Claude to engage with this isyou might be on the go using
this on mobile and it, you canbe confident in the conversation
that you're having with itbecause it's actually giving you
the information about, you know,your colleagues availability,
what works, what's not working,and explaining it to you as you
(27:01):
go.
Um, and I'd also note that, youknow, one of the things that
using MCP servers often enablesis that instead of relying on
kind of a generic system like.
Claude or chat GBT.
Um, we're using a morespecialized system that's
enabled by clockwise to findthose right times.
And so it found with thatconversation of, you know, okay,
we need to extend the timerange.
(27:22):
It found some, uh, I found a toppick on Monday, October 6th,
4:00 PM Um, it's noting thatthat loses, uh, five hours of
focus time, but there's anothergood time on Tuesday.
Or if I want the soonest time, Icould do it at 10 30 at October.
It's also helpfully provided mewith a link where I can view,
um, all the details about thatrequest if I'd like to,
including the times that areavailable.
(27:44):
Um, visualize nicely on mycalendar.
Um, so I'm just gonna say, youknow, uh, the top pick looks
great.
Let's schedule that and so.
Right there.
I've, these are three people whoit's actually quite difficult to
find time with.
And, uh, using Claude with sometools connected, uh, I'm able to
(28:07):
get that schedule right away.
Um, you know, reading thecalendars, seeing attendees,
seeing their availability,seeing their preferences, seeing
their time zones, and thenactually scheduling the meeting.
So that kind of completes a fullloop.
Um, I'd love to show a coupleother use cases unless, sorry,
you have any questions?
Isar Meitis (28:23):
No, this is
awesome.
I would love to see additionaluse cases and I have like four
or five questions, but theymight, you might already answer
them in the other use cases.
So let's go into these usecases.
Let's do it and then we'll see,uh, if my questions are still
relevant,
Matt Martin (28:34):
let's do it.
I'm gonna, I'm gonna fire offmaybe a couple in parallel and
then we can, um, go through.
So one that I love here is I, somy kids are just, um.
Uh, going back to school.
Um, and I got a newsletter, uh,from the school, um, that is
helpfully at least posted on apublic URL here.
(28:56):
So we can look at this.
This is a real world use case.
This is actually, it's cute.
Um, but it varies A bunch ofinformation in here about
schedules, um, and, andimportant events.
And so what I'm asking is forClaude to take a look at this,
uh, newsletter, Claude can, youknow, go crawl that URL, uh, and
then schedule'em on my calendar.
So I'm gonna, I'm just gonnafire that off and I'm gonna go
to the next use case and we'llcome back to see the results.
(29:18):
The next one here is, um, Isometimes like just, I don't
have time to think about it.
I just want help fixing my week.
Um, like, just make it better.
Like, I don't care.
Like take a look at it.
You do the work.
Just make it better.
So I'm gonna ask Claude, Hey,can you look at my week next
week?
(29:39):
And ask clockwise to optimizeit.
Um, so I'm just gonna say, Hey,look at my week next week.
Ask clockwise to optimize it andwe'll see what it does.
This one's super cool.
Um, so here I'm actually gonnahave to take folks through one
more step, which is, again, youcan, you know, connect multiple
(29:59):
connectors.
So here I'm actually going totake you through the flow of
connecting a custom MCP server,which we were talking about
earlier.
CP is linear is not on thislist.
And so when I add a custom one,there's a handy link up here.
When I open the toggle that wementioned before, add connector,
I want linear.
Linear is not on this list.
(30:20):
Um, it's nice, nicely, oh, it ison the list.
I'm sorry, I thought I was gonnatake you through a custom
connector right now, but linearhas already been added to this
list as an actual connector, so.
Let's take the easy route.
I'm just gonna connect itthrough here.
Um, again, if you do want to doa custom connection, just Google
that.
It's really easy to find.
So here I'm gonna paste in, lookat the top priority linear tasks
(30:42):
for Scott and Nikita, and thenlook at their focus on the
clockwise and see if they cancomplete everything this week.
So my engineering team, uh, youknow, we're developing software
all the time, improving it.
Uh, we work off of linear forour tickets about what's what we
need to do.
I do a fair bit of projectmanagement here.
We have a project manager whoactually does that, but I just
wanna look to see if, you know,this team has the capacity to do
(31:03):
that.
And so now it's gonna be usingnot just clockwise, but a couple
tools.
Alright, I fired those off.
And I, that's a pattern thatisn't unusual for me.
Sometimes when I have multiplethoughts, I'll let these all go
at once.
I'm realizing immediately whatone of the problems is with
firing these all off at once isthat it's probably gonna have
asked me.
Um, for permissions as I go.
Isar Meitis (31:24):
So, let me explain
this as we're looking for it.
Yeah.
When you're running any MCPserver, uh, on Claude, every one
of the functions, so we talkedabout this, that there's
different tools within that MCPconnector for each and every one
of the tools.
It's gonna ask you, are we, areyou allowing it to use it?
And you can say yes once or yes,always.
And if you click yes once, then.
Every, every time you're on thetool, you'll have to do the same
(31:47):
thing and it won't finish doingits thing because it may need to
use three or four differentthings in order to complete the
request that you gave it to do.
Because every one of those, whatwe call tools is, is a small
function.
So one of them could be search,one of them could be fetch, one
of them could be update, one ofthem could be, uh, right one.
Like whatever the case.
So if it needs to do all thesethings to complete what you
(32:08):
requested, it needs to make allthese calls and it's gonna stop
at each one unless you toldthat, yes, you're allowed to do
this.
Always.
Uh, one thing that I don't, it'sactually a good question to you.
Maybe you'd know, can I allowall of them kinda like carte
blanche as I install the MCP,you're allowed to do anything,
or it's always gonna ask you thefirst time it's doing the
function and then you have toapprove it.
Matt Martin (32:27):
I, to my knowledge,
um, let's take a peek at our
connectors here for a second,uh, to see if we can.
Find that.
Oh, actually, yes.
Okay.
So if you go in, this is new,I'm glad I checked.
So if you go in that same togglethat we've been talking about,
um, at the very top there,underneath the toggle, there is
a managed connectors link.
You can open this up and Ibelieve, I'd have to check this,
(32:52):
apologize, apologies to theaudience.
This managed connectors may onlybe available if you are the
admin of the team or enterpriseaccount.
So I'm using this in a businesscontext and I admin, if you're
on a personal context and youhave a pro plan, you're gonna
have access to this.
But if you're in a workenvironment, you may not have
access, but if you do, you cango to this connectors
(33:12):
management, you can clickconfigure and it you can allow
on supervised, you have to dothis tool by tool, which is kind
of annoying.
Isar Meitis (33:20):
Um, yeah, there,
there's no select all, but at
least you can do this once andthen every time you use this,
it's gonna run for everybody andyou don't have to wait to
actually run this function inorder to allow it.
That's
Matt Martin (33:29):
right.
And now I'll say this, uh, eventhough we went through this
process, I'm kind of glad thatwe discovered this.
Um, I, I kind of appreciate thatthat anthropic is asking tool by
tool.
It's annoying for a power userwhen I'm using these again and
again.
But for everybody in theaudience, this gives you the
opportunity to play.
You don't have to worry.
You're, a tool is never gonna beenabled to be called by Claude
(33:52):
without your express permissionat least the first time.
So if you're a little bitworried and it's, you know,
frankly reasonable to beworried, like, what's this gonna
do?
How's it gonna access my data?
How's it gonna connect?
Um, you can still play and knowwith confidence that at least
the first time it's called,you're gonna get asked whether
or not to grant permission.
Um, okay.
Going back to our workflows herea little bit.
(34:13):
So as I mentioned, uh, with thisback to school example, I have
my kids' newsletter.
Uh, Claude went and it crawledthat newsletter.
It found all the events and thenit used clockwise to create a
meeting proposal.
And what's great about usingclockwise is again, clockwise
can actually write to mycalendar, um, and it can write
in bulk.
(34:33):
And the way that we'vearchitected clockwise is it
always returns a proposal that Iexplicitly have to approve.
So it's not gonna go and modifymy calendar unless I say yes,
but it's found to all thoseevents, it's added them to my
schedule and I can just go aheadand save it.
And then I have all of theseback to school events on my
calendar in one quick pass.
Isar Meitis (34:51):
I love this.
So I literally did this manualwork.
A few weeks ago, uh, because Iwas tasked by my boss, hence, in
other words, my wife and say,Hey, go and check all the kids'
activities that are coming thiscoming year based on the
school's calendar and add themto both our calendars.
So I went manually and did thiswork and then had to invite her
to all the meetings.
This would've saved me about anhour of my life that I had to do
(35:14):
this thing.
So this is absolutely awesome.
Matt Martin (35:17):
Yeah, and you can,
I mean, and the great thing too
is like you can run this andrerun it, you know, if you have
it, um, you know, you can createa project as you mentioned
earlier, and save the chat, andthen when you get the next
newsletter, drop it in and justlet it go again.
Isar Meitis (35:28):
But I, I'll say
something that it pops into my
head, uh, of how.
Broad, this can be used not forschool stuff.
So I do a lot of publicspeaking.
Uh, I get invited to speak onconferences about ai and most of
the conferences I speak on arenot AI conferences, but just
regular conferences.
And they bring me in as an AIspeaker.
And so I'm always looking fordifferent relevant conferences,
different locations, and beingable to do something like this,
(35:50):
go and search for conferencesthat might be relevant, find
them, look at my calendar, seethat I'm not traveling at that
time for a different conference,or I don't have any other big
conflict.
Uh, and then list those andsuggest what is absolutely
amazing.
Like it's again, just gonna savehours and hours and hours of
work for myself and or myassistant, uh, lo looking for
(36:10):
these kind of things.
And the same thing is true forany other third party data that
AI is really good at crawling,but then that needs to be
verified against your actualavailability.
Matt Martin (36:20):
Yeah, it, there
was, uh, a fun one internally
where somebody was, one of, oneof our employees was playing
with this and said, you know,was using Claude and said, Hey,
can you, uh, check for when thenext time Bona Air or Taylor
Swift or Beyonce is in town, andif so, just add it to my
calendar.
Um, and, you know, it searchedthe web, it did it, found it,
and then used clockwise at thecalendar.
Isar Meitis (36:41):
Yeah.
Matt Martin (36:42):
Um, okay.
So one of the other use casesthat I, uh, wanted clockwise
help with is to help optimize myweek, um, uh, in the coming
weeks.
And so again, we're goingthrough the tool calls and I do
have to approve these, but it'slooking at my calendar events
and then it's going to call atool called Optimize Calendar
Schedule.
(37:03):
And that tool is actually apretty robust one.
It's using all of clockwiseintelligence, all of clockwise.
Power of its scheduling engineto look across all my events of
that week and identify based onmy availability, the attendee's
availability, my schedule, mypreferences, their preferences,
what are some possible ways thatclockwise can help clean up that
(37:25):
week to make it better for meand for all of the attendees.
And it's, um, you know, it is afair amount of compute on our
side, but the brilliant thingabout it is that, you know,
we're not relying on theunderlying LLM to understand all
of the attendees data, which itdoesn't have access to.
But also, um, as we've found indeveloping a lot of these
(37:46):
systems, uh, even the cuttingedge LMS right now are, they can
get tripped up pretty badly whenthey have intersecting calendar
math, um, time zones trip themup.
Time zone math is quite tricky.
And so it's come back with acouple suggestions on how I
could clean up my week.
Um, it's found that.
Um, moving my meeting withClaire would be very helpful.
And moving my, uh, meeting withAaron is very helpful and that
(38:08):
doing that actually adds atremendous amount of heads down
time for each individual.
And just like all otherclockwise requests I can pop in.
It gives me a link that helpsvisualize these changes.
I can see those changes, writingcontext of my calendar and I can
approve them or decline them asI see fit.
Um, but just a nice way to cleaneverything up, um, especially on
a week when I'm traveling.
Isar Meitis (38:28):
All right.
Now quick, quick question aboutthe optimization.
Yeah.
Are there backend rules that youset up on how to optimize?
Like what are the things youprioritize?
What are the things that otherpeople are pro?
Like, I want at least two hoursof deep thinking time on my
calendar, uh, every day.
I do not want to touch that timeregardless of anything.
I have a board meeting thatcannot be like, stuff like that.
(38:51):
Can you set your own rules ofwhat does optimize means, what's
acceptable and not acceptable asfar as optimization?
Matt Martin (38:57):
Yes.
The, the short answer isabsolutely yes.
The, the slightly longer answeris that, um, when you, so you do
have to create a clockwiseaccount with Google Calendar,
um, and it, it's using just thesame login as Google Calendar.
Very simple, verystraightforward.
When you onboard, what we'redoing is we're crawling your
calendar to understand a littlebit of your calendar preferences
(39:19):
by inference.
But then we'll also ask you avariety of questions about which
events are flexible, whicharen't.
Um, what, what are your workinghours?
When do you prefer to havelunch?
When do you prefer to go headsdown?
How much heads down time do youwant?
Um, kind of basic stuff.
And, uh, will you sane defaults?
And then you can go deeper anddeeper down that rabbit hole as
you see fit.
Um, the, the other thing thatI'll note is that this has been
(39:41):
one of the delightful surprisesabout working in MCP land and
working with lms is that becausethese are conversational in
nature, um, you know, we usedto, when we were first
developing this, kind of try tojust shoot back to the lm the
answer.
Um, but what we're doing now iswe're sending the LM a bunch of
information about what we thinkcould happen and other options.
(40:02):
So you can have thatconversation if you're seeing
something that you're like, oh,I actually don't really want to
move that one.
You can go back and forth on it.
Isar Meitis (40:08):
Cool.
Matt Martin (40:09):
So in, in this, uh,
use case, I did get a little bit
delayed by approving the toolrequests here.
Um, but what we're doing here isthat it's hitting, you'll see
it, uh, for those who can't seethis visually.
Um, Claude is looking for thetop priority linear tasks for
two of my colleagues.
(40:30):
And it's using the linear MCP tolist all those issues for those
individuals.
And then it's using clockwise tocross reference what not just
their availability.
Um, you can actually doavailability with Google
Calendar for yourself.
You wanna be able to do withcolleagues'cause doesn't have
access to colleagues.
But if you're looking foryourself, you look at peer
availability.
(40:50):
Here I'm looking for colleaguesand um, I'm using clockwise and
uh.
Our calendar analytics tool touncover, you know, how much deep
work time do they have, howmuch, uh, fragmented time do
they have, um, you know, whatare their preferences or out of
offices, et cetera.
So I'm getting a much richerview into their overall
availability.
And then, uh, Claude can usethat information to stitch
(41:12):
together and analyze, you know,what, what is the availability
for these folks?
Can they accomplish all thetasks that have been assigned to
'em?
And I love this one.
It, it takes a little, as you'llnotice, um, whether you're
watching this on YouTube oryou're listening, um, this one
takes a while because it has tohit multiple tools and it, so
this is one that I like to runin the morning.
On Mondays, I'll just shoot inthe background and then wait for
(41:33):
it to run.
But it's a really great way tohave an understanding of my
team's capacity and whetheranything needs to adjust,
because at the end of the day,it'll give a overall assessment.
And it said yes, you know.
Uh, both Scott and Nikita havesolved the amount of time to
accomplish her tasks and I, I,having run this one, you know,
dozens and dozens of time, um,it's pretty good.
(41:55):
I mean, they'll come back to meand say, Hey, warning, you know,
this person is actually out ofoffice for the back half of the
week.
Um, they're not gonna have asmuch time.
You might want to adjust somethings really helpful for kind
of managing and coordinating ateam.
Isar Meitis (42:07):
Yeah, I think, I
think the, the interesting thing
here and now, now I'm thinking alittle bit into the future,
right?
When these tools will notconflict and they will be able
to run 20 of these tools at thesame time and just run as a
centralized brain.
Think about how incrediblypowerful this thing is.
It can look into all the dataacross.
In theory, your entire techstack, and make logical
(42:30):
suggestions based on a set ofrules that you have
pre-assigned, either in thetools themselves or just by
speaking to Claude Chachi, pt,whatever tool you are using to
run this.
And so you can have a.
Strategic level assistant or avery tactical assistant,
assigning tasks to the relevantpeople who have time on the
calendar to actually completethe tasks that are relevant and
(42:50):
have the skills, like there's somany, think about the HR tool,
the hr, uh, system combined witha scheduling system combined
with a task system.
And now across all of these witha few sentences or as you're
driving the car, just having aconversation with it and being
able to literally make dramatic,significant, important changes
to how operation is going to runin the next few hours, few days,
(43:11):
few weeks, depending on how deepyou want to go.
And I find this absolutelymagical.
I do very basic stuff like thisalready and it blows my mind
when I do these kind of things.
Like, oh my God.
Like, go and find.
And again, just a simpleexample.
I talk to these ais all the timeand I go and say, okay, go and
check, uh, my meetings today,later on, because I'm driving
back from the gym and I know Ihave several meetings this
morning.
I say, okay, well you have ameeting with this so and so, so
(43:32):
and so, and so like, okay, canyou research so and so and tell
me.
Who they are, which company theywork for, and a little bit about
the company and about theperson.
And it will go and it will comeback, invoice and will tell you
these kind of things.
Yep, yep.
And, and this is without fancyconnectors.
Think about once it can actuallydig into all the data in your
company, uh, the amount ofthings you'll be able to achieve
with a predefined amount of timejust amplifies dramatically.
Matt Martin (43:56):
No, I, I, uh, just
a riff on that.
'cause I, I, it, it, youcompletely nerd sny me.
I mean, I think that where thisheads, you know, for us that are
kinda living on the cuttingedge, and I think anybody who's
listening to your podcast is,you know, well ahead of the
pack.
Um, just, you know, a,forgetting these tips, but b,
just'cause you've self-selectedinto interest in this, um, you
know, we're seeing what thepossibilities.
(44:18):
A lot of this is just notreadily accessible to the
average person because, youknow, connecting a, you have to
know to use these tools tostart, which, um, it's, it's.
A lot of people are using chat,GPT.
A lot of people are using CLAai, but I'm still surprised how
few people are digging deep.
But then connecting MCP serversand all this tool connecting,
it's kind of fiddly.
I mean, it just is.
Um, and then knowing how tostitch it together is even more
(44:40):
complex.
And it's kind of slow.
But like, I think that this willget to a place where it's so
seamless that it's easy for theaverage person to ram into it.
And then to your point, are whenit starts to get proactive, man,
that's where some of the realmagic is gonna be involved.
'cause like, I have to thinkabout how to connect this, how
to do all of it, how to plug itin, when to run it.
But like imagine, you know, itknows I'm a product manager at a
(45:01):
mid-size software company and itjust runs that for me on Monday
morning and sends it mydirection.
Um, I I just think there's somuch on the horizon here.
Isar Meitis (45:08):
A hundred percent.
I'll say one thing about thetime it takes, because I, I was
thinking about it and askedabout it a lot, and my answer is
that it doesn't matter.
And the reason it doesn't matteris I don't have to do this.
So let's say, and I see this nowa lot, I, I do more and more
vibe coding and my favoriteplatform as of right now, this
may change next Monday, but asof right now, it's Repli and
(45:32):
Repli.
Now with the new tools, theyhave introduced agent three,
which is amazing.
It really does like crazy stuffwhen it comes to writing code,
testing its own code.
Like it does really amazingthings.
It takes significantly longerthan the other tools.
Like if I run the same prompt inRept and in lovable, lovable
will finish, and rep will stillwork on it sometimes for 40
minutes and people, and I'mlike, is this worth it?
And the answer is, I don't carebecause I didn't have to do that
(45:55):
40 minute minutes of work.
I can at the same time do emailor have a meeting with somebody
or play with my kids or play mybass, guitar, whatever it is
that I wanna do, because thisthing is running on its own.
So whether it could have beenmore efficient in doing the
thing.
Yes, I would love it to gimme ananswer in five seconds instead
of in 30 minutes.
(46:15):
But if I can do other stuff inthose 30 minutes and it dings,
it actually dings the, the, the,the chrome browser goes ding.
And I'm like, oh, it finishedthe thing and I can go back and
look at it.
Um, then, then I don't care howlong it takes.
I mean, again, I care, but a lotless because it doesn't take any
of my time while it's doing thework.
And so going back to what yousaid before, if we are each
gonna have 5, 10, 50, a hundredof these agents running in the
(46:38):
background doing stuff for us,and they're just gonna notify us
when it's done and just thinkabout it as another employee.
As soon as this thing connectsto your notion or Clickup or
Jira or Asana or whatever it isthat you're using, and it's
gonna report just like any otheremployee, their progress and
what they're doing and doing thehandoffs and it's gonna go to a
different agent to complete thestep.
(46:58):
And then the task is gonna movebetween the different steps of
the task on your Jira or Asana,whatever.
There's full transparency ofwhat's actually happening.
You're managing it just likeyou're managing today.
Nothing changed.
Only you have other employeesthat are cost$20 a month that
can actually do the work.
And so this is where it's allgoing and it's just mind
(47:19):
blowing.
You know, both you and me are atthe cutting edge of this and we
are just scratching the surfaceYeah.
With the possibilities of what'shappening.
Matt Martin (47:27):
Well, and and, and
I'll note something that you
said, which we've noticed aswe've been developing software
in this ecosystem, which is thatas you, the leverage that you
can get outta the systems.
Goes up dramatically, the moreconfidence you can have in the
output.
You know, you take your repexample and if, if it were
spending extra 40 minutes and itwasn't any better than lovable,
well that would be a problem.
You know, then, then, then, youknow, it's taking up time and
(47:49):
it's, I code quite a bit aswell.
And the difference between, youknow, having what is effectively
like a junior intern that youhave to sit and pair code with,
which is a huge productivityboost.
I mean, like, I don't wannaundersell that like, you know,
can do a bunch of stuff for meversus being able to say, Hey,
just run at it for an hour andI'll check back in, is
dramatically different.
But the, but you have to haveconfidence in the output of that
(48:10):
agent and it has to be goodenough that you can let it run.
And there's a big, there's a biggap right now in, in overall in
terms of what's available andthe confidence that you can get
out of it.
Isar Meitis (48:18):
I agree a hundred
percent, but I think the jumps
in the past year and definitelytwo years have been incredibly
impressive.
Astronomical.
The trajectory is very clear.
Yeah, it's crazy.
Matt, this was.
Really fascinating, really wellexplained.
I'm sure people will learn a lotfrom this episode across
multiple aspects, not just timemanagement.
Uh, with ai, if people want tofollow you, use your product,
(48:43):
work with you, connect with you,what are the best ways to do
that?
Matt Martin (48:46):
So the, the first
and most important plug is
clockwise ai.
Um, that's clockwise.ai.
Um, you can go there, you canstart with clockwise in 30
seconds.
It doesn't take long at all.
Um, you can also start in CloudConnect.
Use the connector, we will getyou hooked up.
Um, it's very easy if you wannaconnect with me.
Best play, best place to do thatis LinkedIn.
(49:06):
Um, I'm LinkedIn uh, dot com.
Search for Matt Martin or slashin slash vox.
Mat, uh, V-O-X-M-A-T-T is myhandle.
Most places you can find me onMastodon.
I occasionally am still on x,I'm not on there as much.
But, um, would, would love tocontinue the conversation in any
forum you find me.
So reach on out.
Awesome.
Thank you so much.
This was absolutely great.
(49:27):
Yeah.
Thank you so much for having me.