May 20, 2024 • 40 mins
Bringing Coding to the Masses with AI.
Can programming become accessible to all? This week on Things Have Changed podcast we have Filip C. Kozera (CEO) and Robert Chandler (CTO) from Wordware AI, a startup that's simplifying app creation and making it more user-friendly for both experienced programmers and complete beginners!
AI is the hottest topic in tech right now, and for good reason. Filip and Robert break down how they see the software landscape changing – moving away from traditional coding toward simply using plain English instructions with advanced technology.
Imagine you have the next million-dollar app idea – and you're going to build it yourself! But instead of writing complex code, you're simply using...English. That's the vision driving the Wordware team, who are developing these impressive capabilities in a seamless, easy-to-use interface.
Today's episode dives into topics like the rise of "prompt engineering" as a vital new skill, the wide-ranging impacts of easily accessible AI development, and a future where creativity and clear communication are more valued than traditional programming languages. Don't miss this fascinating look into making software development more intuitive and user-friendly!
00:00 The Dawn of AI in Software Development
00:46 The Evolution of Programming: From Code to English
01:50 The Impact of AI on the Software Industry
05:02 Exploring the Future of AI and Software with WordWare
05:18 The Challenges and Innovations in AI Development
07:48 WordWare: Revolutionizing Programming with AI
09:33 The Expanding Role of AI in Various Industries
10:02 The Future of Work: AI Engineers and Domain Experts
12:31 Unlocking Creativity and Efficiency with AI
18:49 WordWare: A New Paradigm for AI Development
25:05 Empowering the Next Generation of Developers and Creatives
30:09 WordWare's Vision: Making AI Accessible to All
36:26 Closing Thoughts and Opportunities with WordWare
THE HIGH-EARNING WOMEN PODCASTThis podcast empowers high-earning women to make informed financial decisions and thrive.
Listen on: Apple Podcasts Spotify
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Jed Tabernero (00:01):
Imagine you've got the next million dollar app
idea.
And you're going to build ityourself.
But instead of writing lines ofcode, You're simply using
English.
Filip Kozera (00:12):
So I think we are inherently working with
something which is nondeterministic, which brings a
lot of difficulty,
Jed Tabernero (00:20):
that's Filip CKozera, EO of Wordware and he's
describing the beauty.
And the difficulty of AI.
Filip Kozera (00:29):
We believe that AI will bring more engineers, not
less.
And the definition of the wordengineer will change to somebody
who is extremely accurate andusing English and knows how to
communicate with these monsterswe call LLMs
Jed Tabernero (00:46):
The shift is central to how new technologies
are being integrated into thesoftware development process,
making it more intuitive andaccessible.
Robert Chandler (00:55):
actually working with LLMs has made
working with AI incredibly easyin comparison.
Prior to LLMs, if you wanted to,automate some process, You had
to gather a data set of millionsof items.
You had to get people toannotate them.
You then had to train a specificmodel for one specific task,
which would require a team ofhighly skilled machine learning
engineers.
Jed Tabernero (01:15):
And that's Robert Chandler, CTO of Wordware.
This evolution is critical as itsignifies a move towards more
efficient and collaborativedevelopment environments where
AI and human creativity converge
Robert Chandler (01:31):
So at its core, it's a new programming language
that is prompt first.
Tools like word where simplifythe creation of AI driven
applications, enablingdevelopers and non-developers
alike to transform ideas intoreality without the traditional
barriers of coding.
Today on things have changedpodcast.
(01:51):
We're learning how prompting isthe new programming with Robert
and Phillip from ward where.
Uh, YC company.
Shikher Bhandary (02:37):
In 2011, Marc Andreessen, who's this renowned
venture capitalist, founder ofNetscape and the most famous VC
firm in the world, A16C, soAndreessen Horowitz, wrote an
article, Software Eating theWorld.
The gist is how software willtouch every aspect of life.
And fast forward, what, 13years, It has everything that we
(03:00):
take in when it comes toinformation, when it comes to
media, when it comes to justdaily life, large aspects of
software in it.
Including the traditionalindustries.
We've been in that phase for awhile and it feels like we are
now getting into that nextphase.
Jed Tabernero (03:18):
So just as you said, the Mark Andreessen
article that was posted in what,2011?
In 2019, somebody who has beenquite consequential in this
space as well, posted somethingelse.
So Jensen Huang, CEO of Nvidia.
And he posts.
yes, software is eating the world, but AI is
(03:41):
going to eat software, right?
So when Shikhar talks about thiskind of next phase, we're
looking at so many things thathappened in the last, I don't
know, two years that hasinvolved AI, right?
Like people have never heard ofAI and its practical
applications are now hearingabout chat GPT.
People are using chat GPT formarketing efforts.
(04:03):
Gemini has come out, there's somany things going on in the AI
space that may be replacing somereally key software development
is one of those spaces.
That's A black hole for folkswho don't really understand
what's going on in the industry,right?
Luckily for Shaker and I, wehave a ton of friends who are
SDEs who are very familiar withthis space.
And so learning about how AI istransforming that space has been
(04:27):
super interesting, right?
So as AI begins to eat software,enhancing and accelerating
certain development processes,there's tools that facilitate
that.
This integration becoming verycrucial enter word, where AI,
right?
A platform that's designed tostreamline and simplify the
(04:50):
creation of AI drivenapplications, which is becoming
quite popular these days, right?
The removing kind of thecomplexities of traditional
programming.
Allowing certain domain expertsto streamline and simplify the
creation of these applications.
Today, we're super lucky onthings have changed podcast to
(05:11):
have Philip Cozzera and RobertChandler CEO and CTO of word
where AI respectively welcome tothings have changed podcast
gentlemen.
Robert Chandler (05:21):
Great to be here.
Filip Kozera (05:22):
us.
Jed Tabernero (05:22):
So we'll get right into it.
This AI development space isquite, again, I was mentioning
quite a black hole to somepeople who aren't in the space.
What do you think's the hardestthing about developing, AI
driven applications?
Filip Kozera (05:39):
Yeah.
So I think we are inherentlyworking with something which is
non deterministic, which bringsa lot of difficulty, we had to
rethink how programming languagewill change once we take English
as the core of it.
And this is what lies in theheart of our product is
rethinking.
(06:00):
How can we work with somethingwhich is inherently non
deterministic and, people talkabout hallucinations and, there
are things which are good aboutthe non deterministic nature.
Like hallucinations actually, inmy vocabulary, it's almost equal
to creativity.
And working with these thingswhich change and shift and we
can no longer can do a verystrict if statement.
(06:22):
That's definitely very difficultfor people to grasp and,
sometimes changing one word isenough to change the whole
output of a large languagemodel.
Robert Chandler (06:31):
Probably the other thing to say is that
actually working with LLMs hasmade working with AI incredibly
easy in comparison.
So prior to LLMs, if you wantedto, automate some process, You
had to gather a data set ofmillions of items.
You had to get people toannotate them.
So it's give you the correctoutput given some set of kind of
(06:52):
noisy inputs.
That would take, months initself.
You then had to train a specificmodel for one specific task,
which would require a team ofquite like highly skilled
machine learning engineers.
And then finally you get thismodel and then you'd have to
keep it up to date and keep itin production.
And so it only meant that likemost high value, most frequently
done tasks could be automated bybig teams.
(07:15):
With LLMs that's changed.
Suddenly you've got thesefoundation models that are
capable of actually reasoning,um, and given almost human level
instructions, they can performarbitrary complex tasks.
There are limitations andthere's and like fighting those
limitations is one of thebiggest challenges of working
with them.
But it is incredibly easycompared to how it used to be.
Shikher Bhandary (07:37):
Y'all have been in the AI space for a while
2015, 2016, where this was stillnovel and compared to today
where everyone's on AI rightnow, what did you see or
experience in your work that ledto this product, the solution?
Robert Chandler (07:53):
Yeah, probably the biggest thing for probably
the listeners that don't yetknow what WordWare does.
So at its core, it's a newprogramming language that is
prompt first.
And it's Exposed in acollaborative web hosted ID.
And that's a lot of words, buteffectively what it means is you
get a place that looks likenotion where actually instead of
creating documents, you'recreating programs, but these
(08:16):
programs are sequences of propsand.
A sequence of prompts isactually what's underlying
everything you see that's calledan agent or everything that you
see that's called an AIapplication.
They're all just chains ofprompts that call other prompts,
that do some self reflection,whatever it might be but it's
natural language.
And so what WordWire does tohelp solve this non determinism
(08:36):
problem, solve this randomnessproblem, Is give people the
ability to iterate faster andfaster.
So if you're changing prompts inthe code base and you have to
spin up some new instance andtrigger that, and then show the
outputs to someone who's maybemore knowledgeable in the field
that you're trying to automate,which is often the PM in, if
it's a company, but it could belike a lawyer or a marketeer or
(08:59):
a salesperson for those kinds ofindustries by putting them in
the driving seat and gettingthem to iterate.
That speeds up the developmentprocess infinitely.
And so they're able to get afeeling of like, when is the
model.
Outputting things in a spacethat I don't like when it's
going off track, being too noisyand they understand which words
to change in their prompt to, tomake that better language model
(09:19):
it's inherently random you'reable to guide it to a much
better set of outputs on abroader set of inputs.
And so it's just having thathuman in the loop in the
development process and beingable to do hundreds, thousands
of iterations.
Even like a thousand iterationsa day that leads to like better
quality LLM applications.
Jed Tabernero (09:39):
You mentioned marketeers, salespeople and
whatnot.
Are they.
The audience I'm sure covers abroad stroke of, software
developers as well.
But when you talk about thesemarketeers and these
salespeople, do you think thatthe goal is someday they could
fully utilize WordWare by promptengineering?
(10:00):
Or do you always think there'sgoing to be an element there of
ML expertise that they need togo, go learn?
Filip Kozera (10:05):
Yeah, so maybe I'll take that one.
Essentially, I think we have aparadigm shift.
And what we are at the core ofthe problem that we are solving
is we believe that AI will bringmore engineers, not less.
And it just the definition ofthe word engineer will change a
little bit to somebody who isextremely accurate and using
(10:26):
English and knows how tocommunicate with these monsters
we call LLMs, because the waythat they actually work, we
don't get it.
So often I imagine this kind ofmulti tentacle monster that's
holding two puppets, which arelooking like human.
And this monster is playing withthese puppets, so these people
need to understand thesemonsters.
(10:47):
And the, by the way, the puppetscome from reinforcement lending
from human feedback.
This monster learned how toimitate humans and how to talk
to us.
And in order to actuallyunderstand the whole monster.
I think there will be engineerswho are using English as kind of
their main programming languagein tools like WordWare.
And we're already seeing theparadigm shift.
(11:08):
We're already seeing people whoare called AI engineers.
And these people will work withdomain experts in order to
create actually fullyfunctioning AI agents.
So it's, I think there's goingto be people who understand
both.
Things they understand bothsales and the kind of that
monster or they will becollaborating with kind of,
(11:29):
sales specialist.
I can change a couple prompts.
And we'll have a engineer whomaybe helps with that.
Exact styling of the prompt orhow this retrieval get into the
whole chains that we're buildingin programs like word where So
overall, I think, we're seeingfrom enterprises.
They are starting promptingteams.
(11:49):
And yeah.
So one of the yeah, one of ourclients, actually a big
enterprise 7000 people.
They are starting a 100 to 150people team that will specialize
in LLM interaction.
The way that it's going to panout in the future, I think it's
hard to say right now.
I think these people need to besomewhat technical, but they
(12:12):
don't need to be coding inPython, they just need to be
very good at communicating theirconcepts clearly, and that's
difficult.
Because our minds are scatteredand, I think even on a podcast
like this, when you have aquestion, answering it
immediately in English in aprecise manner is difficult.
And it will always be.
(12:33):
Communicating between our limbsand humans will always be
difficult.
Shikher Bhandary (12:36):
One of the core visions that stands out
from your website is promptingis the new programming.
So it's fascinating that you'realready seeing that coming into
this call.
I thought, yeah, I can see it.
I put these prompts and I getthese results.
But you're saying that now, ifyou just take out software
(12:58):
engineer, you just take out theprogramming aspect from it, they
are still solving problems andyou can probably solve problems
through the prompts.
Is that the idea?
Filip Kozera (13:12):
Yeah, I would say I would go a little bit further
and currently everyoneunderstands prompting as just a
singular piece of text above ageneration and I would expand it
to say actual future and theequivalent of what you've spoken
at the beginning of likesoftware is eating the world
right now it's going to beagents are eating the world and
(13:34):
maybe I'll just Do a quickexplanation of what agents are
just for the rest of the contextof this because they are at the
heart of a kind of AIrevolution.
So far, we've understood AI assomething that retrieves
knowledge either from theweights of its network that's
been trained or from somedocument we're using where we're
(13:54):
putting that document intovector database.
Which is a more complex databasethat can approximately know the
meaning of each document withoutlooking into the whole reading
all of it and retrievingknowledge.
So those are the use cases we'regetting right now.
Chatbots that can chat with youand give you, good answers based
on some body of text or theweight of the network.
(14:15):
What's AI agents.
So AI agents are actuallycapable of completing more
complex tasks.
So I'm going to give you anexample around essay writing,
even though the whole of AIagents can actually do a lot
more.
They can actually, be like copilots for every part of our
world and our lives.
So if we, Tell an an LLM rightnow, write an essay.
(14:40):
It's like me sitting you infront of a computer, putting a
gun to your face and saying,Hey, you now need to write an
essay about the Roman empire.
You cannot Google, you cannotuse any tools.
You cannot plan it before youcannot talk to your friends and
you cannot reflect, which meansyou cannot use delete backspace.
And I'm putting this gun to yourhead and I'm saying right now,
(15:02):
and the cadence of each wordshould be the same.
And I think that essay would bepretty shit and it's incredible
how well LLMs do at that.
But the most important partright now and what we are
introducing as the core ofWordWare is an ability to create
these agents.
And what are AI agents?
I think they consist ofbasically four scales.
(15:23):
So reflection in this case, it'slike using backspace and
actually being able to go backif you wrote something shitty
planning.
So before you write an essay,you actually are allowed to sit
down and, make sure that thewhole thing works, that you can
do research in particularthings.
The third one is tool use.
So you can use Google and youcan use maybe Grammarly for your
(15:45):
for how, grammatically, butyou're writing.
And the last one is multi agentcollaboration and multi agent
collaboration is the kind ofmore complex one, and it
essentially means you can.
Talk to people who are better atresearch and better at Googling
and better at grammar and style.
And those are more specializedagents.
So I think this is where we seethis kind of a gentle AI coming
(16:09):
into play.
And even though I mentionedessays, Right now it's becoming
workflows and it's becoming,helping with people's work and
making all of these decisions.
Maybe not fully autonomouslyyet, but at least being able to
plan them out in a specificmanner.
Shikher Bhandary (16:23):
I read that to make your answers a bit better
and a bit more human, you got toadd things like please help me.
Thank you for this and stufflike that.
Filip Kozera (16:34):
There's so many of these techniques.
I think those with time will goaway.
Having the ability to just talkto it in a more natural voice.
I think models will improve atthat.
So you don't have to say thesekinds of things, but at the core
of it, planning the whole flowof it being like, Hey, start
with planning, you should thinkin this way, have a thought,
think about an action and, thenconduct the action, those kinds
(16:58):
of logical, almost Almost almostlike logic based steps on how to
think will always persist.
And, I'm actually looking rightnow at a prompting at 12.
Oh, no, it's actually way moreprompting techniques.
So you've got zero shotprompting, few shot prompting,
chain of thought prompting, selfconsistency, prompt chaining,
tree of thought, active prompt,pro program, I did language
(17:19):
models, multimodal CLT graphprompting it's right now it's a
lot.
And that's why we are sayingthat.
It's a full time job, to knowall of this
Jed Tabernero (17:30):
Programming has been an abstraction, isn't it?
It's just more of making thingssimpler to be able to speak to a
computer and to give itcommands, right?
So we've created these language,these languages on top of that,
these programming languages tomake it easier and easier.
If you think about it, Python ismuch easier than the first
programming languages.
I met somebody the other daythat was telling me about, they
would be coming out of thesemassive machines and punching
(17:51):
holes into them or somethinglike that.
I forgot what language that
Robert Chandler (17:54):
Punch card programming.
It's literally like punch cardprogramming.
It's the yeah,
Jed Tabernero (17:58):
that's insane.
It's an abstraction.
Every level we get better and weget better and we get better.
And then when I first saw thisheadline on WordWare AI, I
thought to myself, okay, that'sanother level, again, of
abstraction of saying, look,we're getting closer and closer
to where you just speak to acomputer and it gives you what
you want.
Now, you're talking to me aboutthis kind of agent, this agental
(18:23):
AI concept that's coming up.
I had never thought about thisprior to this call, by the way.
Is that somewhere that wordwhere it's trying to get people
to, where you're able to, givemultiple tasks to this one,
prompt and be able to bring themall together to get the output
that we want.
I think that's a really keydifference here.
And I think Shikhar, when wemake the blog for this, I want
(18:45):
to really focus on a gentle AI.
We haven't covered that conceptbefore, but it's just so novel
for us, but yeah, I guess I'llpause there.
How does that tie into kind ofword work?
Robert Chandler (18:54):
that, that is, is the biggest thing about
WordWare.
It's like giving people theability to make these more
agentic loops in theirprompting.
You can think when you use chatGPT.
You can do the same thing Philipwas talking about with the
essay.
You can give it time to plan,you can paste in extracts from
Wikipedia, but you're there inthe loop, going back and forth
(19:16):
with it.
Whereas if you distill that intoa program, you're able to just
click a button, put in the RomanEmpire and get out a great essay
a few minutes later.
And we when we were explainingto non technical people, we
think about when you're usingchapter GPT and you're doing
that back and forth, you're awizard and you're casting spells
and you're like going back andforth, waving your wand, seeing
the results, putting in some newinputs muttering some new
(19:38):
incantations.
And then when you use word whereyou turn into a potion maker.
And so you're able to distillthat spell into just like a
reusable thing, a potion thatyou can share with your friends,
or you can drink, or you canpour on your data, whatever,
however you want to take theanalogy forwards.
And it basically turns everyoneinto as powerful a prompt
engineer as you are just bydrinking the potion.
Filip Kozera (19:57):
and to add to this, I think what you've
mentioned is assuminggeneralized agents.
It's like, how do you buildthese agents that can think, et
cetera, et cetera.
We are not there yet.
Currently agents like auto GPT.
Our gimmicks and they are cooldemos where right now we're
seeing real business value.
It's these handcrafted agents.
(20:18):
You are actually as a human anda domain expert guiding it
firstly plan, you're giving itthe time to plan and then say,
Critique this plan and you mightuse a different model to
critique this plan.
So to, to, change it, then yousay, Hey, you're have these kind
of tools because of the plan, doresearch, research each one of
(20:41):
them.
And then we do a loop.
So do you see like how.
These kind of classicalprogramming concepts, like
looping and conditionalstatements and calling a
function being like, this is afunction for planning.
It consists of all of thesethings that somebody in WordWare
community has created and itworks super well.
And I would not be able to comeup with the right prompts for
(21:01):
all of this.
And you doing this kind of.
Potion as Robert said, butyou're not saying to an agent
being like, yeah, you can plan,reflect and use tools, go for
it.
You have to guide it in a veryspecific manner.
And that's why we're superexcited about GPT 5.
And by the way, I think, younever want to get involved
yourself with AI companies thatare afraid of GPT 5 or six or
(21:23):
seven.
You want to get yourselfinvolved with the people that
are super excited about it.
And that's why I we're so earlystill GPT 4.
It's amazing, but for what weare using it for, which is not
token generation, but areasoning engine behind
something which is trying totake human creativity and human
(21:44):
productivity and put ittogether, that's going to be the
next frontier.
So for now it's very much humansbehind the the recipe of the
potion, but at some stage itmight be other agents.
Shikher Bhandary (21:55):
I do know there is a promise of AI
touching every aspect of what wedo in our lives, like how
software ate the world.
But for now coming back to likeour first point, are you seeing
more of that AI impact on thesoftware industry first?
And is it down to just the factthat it's already like the
(22:16):
structure data?
It's.
Easier to see the impact rightaway on the software side.
And that's how solutions likeWordWare are being built.
Robert Chandler (22:25):
Software engineers being domain experts
on software.
And so the reason we're seeingloads of software based
applications is because softwareengineers know how to build
software and then validate ifthe output of the AI that built
the software is any good.
The reason we're not seeingloads and loads of applications
in legal tech, in medicine, inlaw, in marketing, in sales, we
(22:49):
are seeing companies spinning upto do this, but it's not as
thriving a community is becausemost lawyers and PMs and
marketeers don't have theability to like import lang
chain into their Pythonenvironment and write a chain of
prompts and, connect it to somekind of internal data store.
That's just, I think living inthe Bay area, you think everyone
(23:11):
can code.
If you leave the Bay area, yourealize that it's not a skill
that people learn as kids justfrom their parents, because most
people's parents can't code.
And that's literally whatWordWare is there to unlock.
It's there are so many creativepeople and we're seeing them in
the community.
They're not technical, but theyare empowered by using wordware
to build, not just toys.
So there's a lot of low codetools that allow you to build
(23:32):
like simple flows, very like youcan chain one prompt to another
prompt and then use Dali orsomething you can't go to
anything actually useful there.
Once you get the ability to havethe same kind of concepts of
programming, but in an interfacethat's familiar to like non
technical people the creativityexplosion is incredible.
And so we're super excited to bethis like GitHub, but for
(23:53):
prompting like we're still superearly days building out all
these community features, but onour discord, there's people
sharing prompts, like cloningeach other's prompts, being
like, Oh, what if we changedthis line?
And Oh, it worked great.
Now, like then someone elsetakes that and improves it.
And it's magical to see.
And these are not like nontechnical people taking their
world knowledge, domainknowledge, taking just
experiments that they're excitedabout and building cool things.
Jed Tabernero (24:16):
think there's absolutely an appetite for that.
Because right now, all thesecreative people using AI is some
form of interaction with a,consumer facing LLM, unlocking
that creativity.
When you talk about unlockingthat creativity, we know there's
groups of people who are willingto learn how to code, who are
willing to learn how to do theselike small things.
(24:37):
And there's always going to be alimit, right?
For example, like going tocollege, I also took a bunch of
coding classes and that's thebig one.
It's not because I wanted tobecome a software engineer, to
be honest with you, just a lotof people that I respected,
encouraged me to take datastructures just because they
told me, look, the world oftomorrow, you're going to need
to understand data structuresand tools like this, I think
(24:58):
will make that much easier forpeople like me.
Who are going into the space whooccasionally I have no, software
development engineer in my jobdescription, but occasionally I
have to write some code.
It's the world that we live intoday.
Tools, I think WordWare makethat a little bit easier to take
it to the next level.
It's not just coding.
No, it's not.
If else in loops, are youfinding there's a ton of use
(25:20):
cases for people who are tryingto get interested and learn more
about your product on how tointeract with it alone.
Filip Kozera (25:24):
Yeah.
And I think one of the problemsthat you haven't mentioned is
that people, even if they havedone a little bit of software
engineering and their brain doeswork in this kind of more
analytical way, and you can puta structure on it.
They can't deal with.
Things like deployment, theycan't show, other people that
they're the results of theirwork, which is really sucks.
(25:46):
If you take a person who can doa little bit of Python, but then
you tell them, Hey, spin it upon the website with a front end
and put a bunch of LLMs togetherand they are just going to
panic.
And this is where kind of wordwhere it comes in is we are
building the first community ofpeople where they can take their
prompting and take their ideas.
Create these these workflows orkind of little blocks of RLM
(26:10):
interactions and share them withthe community.
So we have one click of abutton.
If you, first of all, if you area CEO or an early stage startup,
or you're trying to build aproduct with one click of a
button, you get an API, whichyou can plug into the code, and
this is for some people, this istheir whole backend.
So they spin up a front end,they put the API there, and
because you can run code onWordWare.
(26:32):
You can actually do everythingon top of it, but there is
another way where you just clickshare and that gives you a
hosted application in a similarway to replicate.
com, which allows you topublicly host your machine
learning models and show themoff to the world.
Here you get to show off yourkind of word apps or prompt
(26:55):
cascades with tools, et cetera,to the rest of the world.
And also, very importantly, youallow the community members to
fork that and to duplicate thatflow and to reuse it inside of
their application.
So I think you're very correct.
Identify that there are peoplewho are analytical and you've
even probably wrote more codethan we expect our users to be
(27:19):
able to do.
But yeah, we're giving it to allthe domain experts.
So it's not only going to becode gen right now, a farmer can
write a program which will tryto optimize how he should move
the crop based on the weatherand a bunch of factors that I
have no clue about.
Neither any software engineerhas any clue about.
So that's why we want to forthese AI aficionados who are
(27:41):
early on and they're intelligentanalytical and know their
domain, give them the right toolto be able to to take their
knowledge and instill it andgive it to LLM in the right
manner.
Robert Chandler (27:51):
It's a little bit like Excel in the, Excel is
basically the most usedprogramming language in the
world.
It's probably got the mostdevelopers.
Building with it.
And it's not a program language,but it is a program language.
And the reason it's so popularis there's such a long tail of
applications that there's no waystartups or companies are going
to build the thing for you, foreveryone.
(28:13):
You end up with a regression tothe mean.
On what companies are viable andit's, where is there enough
value and like on a per usagebasis and enough users multiply
that together and you get acompany that you can actually
start and build.
But there's a limited supply ofsoftware engineers.
There's a limited supply ofentrepreneurs.
For now, once, once we get theseAI engineers and software
(28:33):
engineers, that might change.
But yeah, you need, youbasically need to put the tools
in the hands of the people thatknow their problems, know what
kind of thing they would buildif they had the tool to build
it, and then actually give themthe tool and see what they build
and, empower them.
Shikher Bhandary (28:45):
So when you say Excel, Jed's eyes light up
because when you, yeah, he'sthat guy, he's that guy,
everything in his life from likegrocery bills to stock modeling
and performance of an asset isall on Excel.
The first thing.
And it's funny because the firsttime I met him, he was doing all
these calculations of utilityspend and stuff like that.
(29:06):
And how we are trending over 12months.
And I'm like, dude, that's it's.
Up by a dollar, like what areyou doing trend
Robert Chandler (29:14):
Dude,
Filip Kozera (29:14):
you guys
Robert Chandler (29:15):
get a girlfriend.
It's
Shikher Bhandary (29:15):
Yeah.
Jed Tabernero (29:16):
We used to.
Filip Kozera (29:17):
to.
Okay.
Okay.
We live together.
We live together with Robert aswell.
He's downstairs.
Jed Tabernero (29:24):
You just do the same thing.
It's important.
Before you ask the question,Shikhar, I just want to point
out, I just, It's, you mentionedExcel is a coding language,
right?
Py, there are about 8.
2 million people around theworld who know how to code in
Python, right?
If you were to guess how manypeople use Excel in the world,
just throw out a number.
(29:44):
I just want to see what theperspective is here.
Robert Chandler (29:47):
300 million, maybe?
Filip Kozera (29:50):
200 million, something like that.
Jed Tabernero (29:51):
There are 750 million people in the world who
use Excel.
That's
Filip Kozera (29:57):
that's our target market.
That's our target market.
Jed Tabernero (30:01):
And those people are analytical.
They're analytical, right?
Inherently you have to use Excelto do a lot of the business
worlds.
I think in 10 years, people willthink about Excel as, Oh, that's
like some tool that, a lot ofthe older people use, I
disagree.
There's a lot of people who arestill going to continue to use
Excel because they can't get tothat next level of abstraction,
I think word where it can bethat tool for folks who want to
(30:23):
utilize LLMs, for example,anyway, sorry, sugar.
I completely cut you
Shikher Bhandary (30:28):
Yeah no.
I was actually going therebecause I wanted to convey that
I use Excel a certain way andJed being, a financial dude he's
used Excel in a different wayand he loves using it that way.
So I was just coming down tojust wordware as well and how
(30:48):
you've.
Mentioned that you want to belike a notion style,
collaborative environment.
So, one thing that, all our THCresearch is on notion and we got
those templates from thecommunity and it has helped us,
turn those notes into a product,which is this podcast and where
(31:11):
we actually connect founders toVCs and capital and stuff.
So it's so interesting how,You're going at it the same way
where you are creating thiscommunity, the community shares
a certain prompt and someoneelse can build a product based
on that.
automation or workflow.
Robert Chandler (31:30):
Yeah, I think there's a number of reasons.
And some of them are just that,like the connotations of notions
are great.
Everyone loves notion.
It's definitely a benchmark forhow to build great productivity
software.
The sort of more importantreasons are, it's already
familiar to people who are ourtarget audience.
So the PMs, know how to useNotion.
(31:50):
And so if we can leverage thesame affordances, then it will
be more like the learning curvebecause it is, it's still a
learning curve to use WordWare.
Like it's a new paradigm.
You're having to firstly learnhow to do prompting and then how
to use our tool.
I think hopefully the how to useour tool part is easy and you
just basically need to learn howto do prompting.
And that's why we see peoplethat have used like LangChain,
(32:11):
they can move to WordWare and belike, Oh my God, this just makes
things Way lower friction andway better.
And then the people who arelike, have not done any
prompting or agent buildingbefore go still confusing, but I
get the slash command and I getthe how to use the tool.
I just need to work out how do Iprompt the other kind of why
notion and why not somethingelse.
So there's you've probably seena few low code tools that are
(32:32):
flowchart based.
And actually the very firstiteration of word wear was built
at a hackathon.
Was flowchart based the problemyou've reached with flowcharts
is whilst they look very simplewhen you look at them at a high
level blockiness, when it comesto actually like building
something complex where you havethese loops that are going
background, you have branching,suddenly your simple looking
(32:53):
workflow that was like a fewblocks leading one to the next
turns into this horrible mess oflines going all over the place.
Shikher Bhandary (32:59):
Yeah.
Robert Chandler (33:00):
It's actually way easier and way like the
structure of software is thatit's like documents and you have
functions that call functions,which gives you effectively a
third dimension rather than thiskind of 2D grid of a flowchart
based thing.
And so that was really, it was,it came.
Through seeing the deficienciesin the flowchart model that we
were like what if it looks likesomething else?
What if it looked more likeprogramming and maybe the way to
(33:22):
make that accessible is to makeit look like Notion.
And it was very much came out ofa, almost like a design
experiment.
And then it was like, Oh waitthis works.
This is beautiful.
Like it's I can't tell you howmuch I enjoy prompting in
WordWare compared to writingprompts in the code base.
Jed Tabernero (33:35):
Thinking through kind of your target audience,
right?
Like folks who are creative, whowant to get into the technical
stuff and learn just a littlebit enough to build something
beautiful.
I'm thinking about it at anenterprise level where you
mentioned, right?
Like PMs really use Notion alot.
When I initially looked, Iwatched your walkthrough, Robert
on creating just the chat bot orthe first agent, I guess that
(33:59):
just to me was like, okay, if Ican work with them like this at
an enterprise level where, okay,let's say we have some kind of
LLM integration and I need tocommunicate with domain
expertise as a softwareengineer.
That actually makes it a loteasier for me to even
communicate to them in English.
This is exactly, I think whatI'm doing.
This is the structure that Ineed it to be.
(34:20):
I think that'd be a huge unlockjust for people who, don't want
to think about the structure.
Don't have to think about,certain things that, that
software developers have tothink about.
Filip Kozera (34:29):
So you've mentioned kind of two different
ways.
So one is the community sharingtemplates where it's the same as
notion, but actually incorporate setting, the dynamics
a little different.
So in because you can write somecode and in word, where what
these engineers are doing,they're providing you with these
blocks and you don't need tounderstand how these blocks
work, but they will tell you,Hey, this We figured out that
(34:52):
for retrieval of this type oflegal documents, we need to
choose these hyper parametersand these vector database, and
you don't need to understand anyof it, but they're going to come
to you and tell you, Hey, whenyou're working with legal
documents, use this block.
We called it retrieval argumentretrieval for legal documents,
and we've made it workperfectly.
And then there will be differentretrieval for internal
(35:13):
documents.
Just because how much of a muchexpelled retrieval augmented
generation is still and so wesee kind of engineers writing
these blocks and you as aconsultant or as a salesperson,
being able to prompt and playaround with it and see exactly
what the retrieval has outputtedand create these these well
(35:34):
working word apps.
And because you can seeeverything gets at the core of
it.
We are, what is what you getplatform, which means that you
can debug it extremely quickly.
So those kind of, those, this iswhat we are seeing from the
corporate world is that they arecollaborating and it's not.
Like the famous notion template,because the notion template
(35:54):
would be like a fully workingthing, but it's more these two
people working together.
And and it's a lot aboutconsultants as well.
So they go to the client andthey listen to their problem and
they are a little bit better atprompting.
Cause I think consultants jobsare going to change from
creating decks to creating.
Proof of concepts right now.
And I've got my own theory aboutwhy they are freaking out right
(36:17):
now and why is it moving soquickly?
So I can dive a little bitdeeper into that, but they
essentially booked a lot ofrevenue and in order to create
good return on investment forall of these companies, they
need to be doing something elsethan the PowerPoint deck,
Shikher Bhandary (36:31):
Before we close, I want to give you the
stage to shout out the work thatyou're doing, how people can
reach you, hiring or funding,and just want to give you the
stage to shout out your teambecause it looks like an
incredible product and you guysare crushing it.
Filip Kozera (36:47):
thank you so much.
So for any listeners Of thingshave changed podcast.
We've got 250 of credits onWordWare.
So we, we discussed this, sojust enjoy it.
Just try all the differentmodels go to WordWare.
ai.
It's like software, but withWord and just sign up for anyone
right now the signups will befully open and enjoy the
(37:09):
product.
As mentioned, anyone can buildmore complex AI agents.
Make sure you go through theonboarding process.
As we've mentioned, this is anew programming language.
So it takes the 5 10 minutes.
Don't expect to be able to jumpin and start creating amazing
agents.
It is still somewhat complex todo these things.
So that's one thing we are alsoactively hiring.
(37:29):
I'll let Robert do the kind offoundation engineer job
description, but whoever is veryinterested in the community
about agents, just reach out tous at founders at wordware.
ai.
We have multiple roles open andit's actually quite in a very
early stage startup.
It's all kind of fluid and mixestogether.
But on the engineering note ofhiring Robert will say a little
(37:51):
bit more and lastly we arehelping a bunch of enterprises.
If there are any enterprisecustomers listening to us we're
more than happy to give you freeconsultancy at the beginning and
help you figure out how todeliver on that ROI and all that
book revenue and for many otherpeople build co pilots and
workflows that will make yourenterprise more efficient.
Robert Chandler (38:12):
Yeah.
I think founding engineer wisethere's all sorts of incredible
things to build.
We're a full stack TypeScriptstack.
We love serverless.
So we love doing things asefficiently as possible worrying
about building the value addstuff, not the undifferentiated
things that, everyone has tobuild.
If you love building things, ifyou love making things, if
you're excited about the worldof agents hit me up on Twitter,
(38:35):
Bertie underscore AI or email.
Robert at WordWare.
ai and tell me why you'reexcited to build on WordWare.
Jed Tabernero (38:44):
Beautiful.
Guys,
Shikher Bhandary (38:46):
want to mention one thing.
You guys stack team so well.
Where Philip takes a certainsection and then Robert takes
the other section.
Even on the message front.
So
Robert Chandler (38:56):
Years of practice.
It's we've known each other for10 years.
It's.
Jed Tabernero (39:00):
yeah, no, it's, it was really fun guys.
So much in this world that we'rereally curious about, but really
appreciate that.
The information and opinionsexpressed in this episode are
for informational purposes only.
And are not intended asfinancial investment or
professional advice.
Always consult with a qualifiedprofessional before making any
(39:20):
decisions based on the conceptprovided.
Neither the podcast, nor iscreators are responsible for any
actions taken as a result oflistening to this episode.
Imagine you've got the next million dollar app
idea.
And you're going to build ityourself.
But instead of writing lines ofcode, You're simply using
English.
Filip Kozera (00:12):
So I think we are inherently working with
something which is nondeterministic, which brings a
lot of difficulty,
Jed Tabernero (00:20):
that's Filip CKozera, EO of Wordware and he's
describing the beauty.
And the difficulty of AI.
Filip Kozera (00:29):
We believe that AI will bring more engineers, not
less.
And the definition of the wordengineer will change to somebody
who is extremely accurate andusing English and knows how to
communicate with these monsterswe call LLMs
Jed Tabernero (00:46):
The shift is central to how new technologies
are being integrated into thesoftware development process,
making it more intuitive andaccessible.
Robert Chandler (00:55):
actually working with LLMs has made
working with AI incredibly easyin comparison.
Prior to LLMs, if you wanted to,automate some process, You had
to gather a data set of millionsof items.
You had to get people toannotate them.
You then had to train a specificmodel for one specific task,
which would require a team ofhighly skilled machine learning
engineers.
Jed Tabernero (01:15):
And that's Robert Chandler, CTO of Wordware.
This evolution is critical as itsignifies a move towards more
efficient and collaborativedevelopment environments where
AI and human creativity converge
Robert Chandler (01:31):
So at its core, it's a new programming language
that is prompt first.
Tools like word where simplifythe creation of AI driven
applications, enablingdevelopers and non-developers
alike to transform ideas intoreality without the traditional
barriers of coding.
Today on things have changedpodcast.
(01:51):
We're learning how prompting isthe new programming with Robert
and Phillip from ward where.
Uh, YC company.
Shikher Bhandary (02:37):
In 2011, Marc Andreessen, who's this renowned
venture capitalist, founder ofNetscape and the most famous VC
firm in the world, A16C, soAndreessen Horowitz, wrote an
article, Software Eating theWorld.
The gist is how software willtouch every aspect of life.
And fast forward, what, 13years, It has everything that we
(03:00):
take in when it comes toinformation, when it comes to
media, when it comes to justdaily life, large aspects of
software in it.
Including the traditionalindustries.
We've been in that phase for awhile and it feels like we are
now getting into that nextphase.
Jed Tabernero (03:18):
So just as you said, the Mark Andreessen
article that was posted in what,2011?
In 2019, somebody who has beenquite consequential in this
space as well, posted somethingelse.
So Jensen Huang, CEO of Nvidia.
And he posts.
yes, software is eating the world, but AI is
(03:41):
going to eat software, right?
So when Shikhar talks about thiskind of next phase, we're
looking at so many things thathappened in the last, I don't
know, two years that hasinvolved AI, right?
Like people have never heard ofAI and its practical
applications are now hearingabout chat GPT.
People are using chat GPT formarketing efforts.
(04:03):
Gemini has come out, there's somany things going on in the AI
space that may be replacing somereally key software development
is one of those spaces.
That's A black hole for folkswho don't really understand
what's going on in the industry,right?
Luckily for Shaker and I, wehave a ton of friends who are
SDEs who are very familiar withthis space.
And so learning about how AI istransforming that space has been
(04:27):
super interesting, right?
So as AI begins to eat software,enhancing and accelerating
certain development processes,there's tools that facilitate
that.
This integration becoming verycrucial enter word, where AI,
right?
A platform that's designed tostreamline and simplify the
(04:50):
creation of AI drivenapplications, which is becoming
quite popular these days, right?
The removing kind of thecomplexities of traditional
programming.
Allowing certain domain expertsto streamline and simplify the
creation of these applications.
Today, we're super lucky onthings have changed podcast to
(05:11):
have Philip Cozzera and RobertChandler CEO and CTO of word
where AI respectively welcome tothings have changed podcast
gentlemen.
Robert Chandler (05:21):
Great to be here.
Filip Kozera (05:22):
us.
Jed Tabernero (05:22):
So we'll get right into it.
This AI development space isquite, again, I was mentioning
quite a black hole to somepeople who aren't in the space.
What do you think's the hardestthing about developing, AI
driven applications?
Filip Kozera (05:39):
Yeah.
So I think we are inherentlyworking with something which is
non deterministic, which bringsa lot of difficulty, we had to
rethink how programming languagewill change once we take English
as the core of it.
And this is what lies in theheart of our product is
rethinking.
(06:00):
How can we work with somethingwhich is inherently non
deterministic and, people talkabout hallucinations and, there
are things which are good aboutthe non deterministic nature.
Like hallucinations actually, inmy vocabulary, it's almost equal
to creativity.
And working with these thingswhich change and shift and we
can no longer can do a verystrict if statement.
(06:22):
That's definitely very difficultfor people to grasp and,
sometimes changing one word isenough to change the whole
output of a large languagemodel.
Robert Chandler (06:31):
Probably the other thing to say is that
actually working with LLMs hasmade working with AI incredibly
easy in comparison.
So prior to LLMs, if you wantedto, automate some process, You
had to gather a data set ofmillions of items.
You had to get people toannotate them.
So it's give you the correctoutput given some set of kind of
(06:52):
noisy inputs.
That would take, months initself.
You then had to train a specificmodel for one specific task,
which would require a team ofquite like highly skilled
machine learning engineers.
And then finally you get thismodel and then you'd have to
keep it up to date and keep itin production.
And so it only meant that likemost high value, most frequently
done tasks could be automated bybig teams.
(07:15):
With LLMs that's changed.
Suddenly you've got thesefoundation models that are
capable of actually reasoning,um, and given almost human level
instructions, they can performarbitrary complex tasks.
There are limitations andthere's and like fighting those
limitations is one of thebiggest challenges of working
with them.
But it is incredibly easycompared to how it used to be.
Shikher Bhandary (07:37):
Y'all have been in the AI space for a while
2015, 2016, where this was stillnovel and compared to today
where everyone's on AI rightnow, what did you see or
experience in your work that ledto this product, the solution?
Robert Chandler (07:53):
Yeah, probably the biggest thing for probably
the listeners that don't yetknow what WordWare does.
So at its core, it's a newprogramming language that is
prompt first.
And it's Exposed in acollaborative web hosted ID.
And that's a lot of words, buteffectively what it means is you
get a place that looks likenotion where actually instead of
creating documents, you'recreating programs, but these
(08:16):
programs are sequences of propsand.
A sequence of prompts isactually what's underlying
everything you see that's calledan agent or everything that you
see that's called an AIapplication.
They're all just chains ofprompts that call other prompts,
that do some self reflection,whatever it might be but it's
natural language.
And so what WordWire does tohelp solve this non determinism
(08:36):
problem, solve this randomnessproblem, Is give people the
ability to iterate faster andfaster.
So if you're changing prompts inthe code base and you have to
spin up some new instance andtrigger that, and then show the
outputs to someone who's maybemore knowledgeable in the field
that you're trying to automate,which is often the PM in, if
it's a company, but it could belike a lawyer or a marketeer or
(08:59):
a salesperson for those kinds ofindustries by putting them in
the driving seat and gettingthem to iterate.
That speeds up the developmentprocess infinitely.
And so they're able to get afeeling of like, when is the
model.
Outputting things in a spacethat I don't like when it's
going off track, being too noisyand they understand which words
to change in their prompt to, tomake that better language model
(09:19):
it's inherently random you'reable to guide it to a much
better set of outputs on abroader set of inputs.
And so it's just having thathuman in the loop in the
development process and beingable to do hundreds, thousands
of iterations.
Even like a thousand iterationsa day that leads to like better
quality LLM applications.
Jed Tabernero (09:39):
You mentioned marketeers, salespeople and
whatnot.
Are they.
The audience I'm sure covers abroad stroke of, software
developers as well.
But when you talk about thesemarketeers and these
salespeople, do you think thatthe goal is someday they could
fully utilize WordWare by promptengineering?
(10:00):
Or do you always think there'sgoing to be an element there of
ML expertise that they need togo, go learn?
Filip Kozera (10:05):
Yeah, so maybe I'll take that one.
Essentially, I think we have aparadigm shift.
And what we are at the core ofthe problem that we are solving
is we believe that AI will bringmore engineers, not less.
And it just the definition ofthe word engineer will change a
little bit to somebody who isextremely accurate and using
(10:26):
English and knows how tocommunicate with these monsters
we call LLMs, because the waythat they actually work, we
don't get it.
So often I imagine this kind ofmulti tentacle monster that's
holding two puppets, which arelooking like human.
And this monster is playing withthese puppets, so these people
need to understand thesemonsters.
(10:47):
And the, by the way, the puppetscome from reinforcement lending
from human feedback.
This monster learned how toimitate humans and how to talk
to us.
And in order to actuallyunderstand the whole monster.
I think there will be engineerswho are using English as kind of
their main programming languagein tools like WordWare.
And we're already seeing theparadigm shift.
(11:08):
We're already seeing people whoare called AI engineers.
And these people will work withdomain experts in order to
create actually fullyfunctioning AI agents.
So it's, I think there's goingto be people who understand
both.
Things they understand bothsales and the kind of that
monster or they will becollaborating with kind of,
(11:29):
sales specialist.
I can change a couple prompts.
And we'll have a engineer whomaybe helps with that.
Exact styling of the prompt orhow this retrieval get into the
whole chains that we're buildingin programs like word where So
overall, I think, we're seeingfrom enterprises.
They are starting promptingteams.
(11:49):
And yeah.
So one of the yeah, one of ourclients, actually a big
enterprise 7000 people.
They are starting a 100 to 150people team that will specialize
in LLM interaction.
The way that it's going to panout in the future, I think it's
hard to say right now.
I think these people need to besomewhat technical, but they
(12:12):
don't need to be coding inPython, they just need to be
very good at communicating theirconcepts clearly, and that's
difficult.
Because our minds are scatteredand, I think even on a podcast
like this, when you have aquestion, answering it
immediately in English in aprecise manner is difficult.
And it will always be.
(12:33):
Communicating between our limbsand humans will always be
difficult.
Shikher Bhandary (12:36):
One of the core visions that stands out
from your website is promptingis the new programming.
So it's fascinating that you'realready seeing that coming into
this call.
I thought, yeah, I can see it.
I put these prompts and I getthese results.
But you're saying that now, ifyou just take out software
(12:58):
engineer, you just take out theprogramming aspect from it, they
are still solving problems andyou can probably solve problems
through the prompts.
Is that the idea?
Filip Kozera (13:12):
Yeah, I would say I would go a little bit further
and currently everyoneunderstands prompting as just a
singular piece of text above ageneration and I would expand it
to say actual future and theequivalent of what you've spoken
at the beginning of likesoftware is eating the world
right now it's going to beagents are eating the world and
(13:34):
maybe I'll just Do a quickexplanation of what agents are
just for the rest of the contextof this because they are at the
heart of a kind of AIrevolution.
So far, we've understood AI assomething that retrieves
knowledge either from theweights of its network that's
been trained or from somedocument we're using where we're
(13:54):
putting that document intovector database.
Which is a more complex databasethat can approximately know the
meaning of each document withoutlooking into the whole reading
all of it and retrievingknowledge.
So those are the use cases we'regetting right now.
Chatbots that can chat with youand give you, good answers based
on some body of text or theweight of the network.
(14:15):
What's AI agents.
So AI agents are actuallycapable of completing more
complex tasks.
So I'm going to give you anexample around essay writing,
even though the whole of AIagents can actually do a lot
more.
They can actually, be like copilots for every part of our
world and our lives.
So if we, Tell an an LLM rightnow, write an essay.
(14:40):
It's like me sitting you infront of a computer, putting a
gun to your face and saying,Hey, you now need to write an
essay about the Roman empire.
You cannot Google, you cannotuse any tools.
You cannot plan it before youcannot talk to your friends and
you cannot reflect, which meansyou cannot use delete backspace.
And I'm putting this gun to yourhead and I'm saying right now,
(15:02):
and the cadence of each wordshould be the same.
And I think that essay would bepretty shit and it's incredible
how well LLMs do at that.
But the most important partright now and what we are
introducing as the core ofWordWare is an ability to create
these agents.
And what are AI agents?
I think they consist ofbasically four scales.
(15:23):
So reflection in this case, it'slike using backspace and
actually being able to go backif you wrote something shitty
planning.
So before you write an essay,you actually are allowed to sit
down and, make sure that thewhole thing works, that you can
do research in particularthings.
The third one is tool use.
So you can use Google and youcan use maybe Grammarly for your
(15:45):
for how, grammatically, butyou're writing.
And the last one is multi agentcollaboration and multi agent
collaboration is the kind ofmore complex one, and it
essentially means you can.
Talk to people who are better atresearch and better at Googling
and better at grammar and style.
And those are more specializedagents.
So I think this is where we seethis kind of a gentle AI coming
(16:09):
into play.
And even though I mentionedessays, Right now it's becoming
workflows and it's becoming,helping with people's work and
making all of these decisions.
Maybe not fully autonomouslyyet, but at least being able to
plan them out in a specificmanner.
Shikher Bhandary (16:23):
I read that to make your answers a bit better
and a bit more human, you got toadd things like please help me.
Thank you for this and stufflike that.
Filip Kozera (16:34):
There's so many of these techniques.
I think those with time will goaway.
Having the ability to just talkto it in a more natural voice.
I think models will improve atthat.
So you don't have to say thesekinds of things, but at the core
of it, planning the whole flowof it being like, Hey, start
with planning, you should thinkin this way, have a thought,
think about an action and, thenconduct the action, those kinds
(16:58):
of logical, almost Almost almostlike logic based steps on how to
think will always persist.
And, I'm actually looking rightnow at a prompting at 12.
Oh, no, it's actually way moreprompting techniques.
So you've got zero shotprompting, few shot prompting,
chain of thought prompting, selfconsistency, prompt chaining,
tree of thought, active prompt,pro program, I did language
(17:19):
models, multimodal CLT graphprompting it's right now it's a
lot.
And that's why we are sayingthat.
It's a full time job, to knowall of this
Jed Tabernero (17:30):
Programming has been an abstraction, isn't it?
It's just more of making thingssimpler to be able to speak to a
computer and to give itcommands, right?
So we've created these language,these languages on top of that,
these programming languages tomake it easier and easier.
If you think about it, Python ismuch easier than the first
programming languages.
I met somebody the other daythat was telling me about, they
would be coming out of thesemassive machines and punching
(17:51):
holes into them or somethinglike that.
I forgot what language that
Robert Chandler (17:54):
Punch card programming.
It's literally like punch cardprogramming.
It's the yeah,
Jed Tabernero (17:58):
that's insane.
It's an abstraction.
Every level we get better and weget better and we get better.
And then when I first saw thisheadline on WordWare AI, I
thought to myself, okay, that'sanother level, again, of
abstraction of saying, look,we're getting closer and closer
to where you just speak to acomputer and it gives you what
you want.
Now, you're talking to me aboutthis kind of agent, this agental
(18:23):
AI concept that's coming up.
I had never thought about thisprior to this call, by the way.
Is that somewhere that wordwhere it's trying to get people
to, where you're able to, givemultiple tasks to this one,
prompt and be able to bring themall together to get the output
that we want.
I think that's a really keydifference here.
And I think Shikhar, when wemake the blog for this, I want
(18:45):
to really focus on a gentle AI.
We haven't covered that conceptbefore, but it's just so novel
for us, but yeah, I guess I'llpause there.
How does that tie into kind ofword work?
Robert Chandler (18:54):
that, that is, is the biggest thing about
WordWare.
It's like giving people theability to make these more
agentic loops in theirprompting.
You can think when you use chatGPT.
You can do the same thing Philipwas talking about with the
essay.
You can give it time to plan,you can paste in extracts from
Wikipedia, but you're there inthe loop, going back and forth
(19:16):
with it.
Whereas if you distill that intoa program, you're able to just
click a button, put in the RomanEmpire and get out a great essay
a few minutes later.
And we when we were explainingto non technical people, we
think about when you're usingchapter GPT and you're doing
that back and forth, you're awizard and you're casting spells
and you're like going back andforth, waving your wand, seeing
the results, putting in some newinputs muttering some new
(19:38):
incantations.
And then when you use word whereyou turn into a potion maker.
And so you're able to distillthat spell into just like a
reusable thing, a potion thatyou can share with your friends,
or you can drink, or you canpour on your data, whatever,
however you want to take theanalogy forwards.
And it basically turns everyoneinto as powerful a prompt
engineer as you are just bydrinking the potion.
Filip Kozera (19:57):
and to add to this, I think what you've
mentioned is assuminggeneralized agents.
It's like, how do you buildthese agents that can think, et
cetera, et cetera.
We are not there yet.
Currently agents like auto GPT.
Our gimmicks and they are cooldemos where right now we're
seeing real business value.
It's these handcrafted agents.
(20:18):
You are actually as a human anda domain expert guiding it
firstly plan, you're giving itthe time to plan and then say,
Critique this plan and you mightuse a different model to
critique this plan.
So to, to, change it, then yousay, Hey, you're have these kind
of tools because of the plan, doresearch, research each one of
(20:41):
them.
And then we do a loop.
So do you see like how.
These kind of classicalprogramming concepts, like
looping and conditionalstatements and calling a
function being like, this is afunction for planning.
It consists of all of thesethings that somebody in WordWare
community has created and itworks super well.
And I would not be able to comeup with the right prompts for
(21:01):
all of this.
And you doing this kind of.
Potion as Robert said, butyou're not saying to an agent
being like, yeah, you can plan,reflect and use tools, go for
it.
You have to guide it in a veryspecific manner.
And that's why we're superexcited about GPT 5.
And by the way, I think, younever want to get involved
yourself with AI companies thatare afraid of GPT 5 or six or
(21:23):
seven.
You want to get yourselfinvolved with the people that
are super excited about it.
And that's why I we're so earlystill GPT 4.
It's amazing, but for what weare using it for, which is not
token generation, but areasoning engine behind
something which is trying totake human creativity and human
(21:44):
productivity and put ittogether, that's going to be the
next frontier.
So for now it's very much humansbehind the the recipe of the
potion, but at some stage itmight be other agents.
Shikher Bhandary (21:55):
I do know there is a promise of AI
touching every aspect of what wedo in our lives, like how
software ate the world.
But for now coming back to likeour first point, are you seeing
more of that AI impact on thesoftware industry first?
And is it down to just the factthat it's already like the
(22:16):
structure data?
It's.
Easier to see the impact rightaway on the software side.
And that's how solutions likeWordWare are being built.
Robert Chandler (22:25):
Software engineers being domain experts
on software.
And so the reason we're seeingloads of software based
applications is because softwareengineers know how to build
software and then validate ifthe output of the AI that built
the software is any good.
The reason we're not seeingloads and loads of applications
in legal tech, in medicine, inlaw, in marketing, in sales, we
(22:49):
are seeing companies spinning upto do this, but it's not as
thriving a community is becausemost lawyers and PMs and
marketeers don't have theability to like import lang
chain into their Pythonenvironment and write a chain of
prompts and, connect it to somekind of internal data store.
That's just, I think living inthe Bay area, you think everyone
(23:11):
can code.
If you leave the Bay area, yourealize that it's not a skill
that people learn as kids justfrom their parents, because most
people's parents can't code.
And that's literally whatWordWare is there to unlock.
It's there are so many creativepeople and we're seeing them in
the community.
They're not technical, but theyare empowered by using wordware
to build, not just toys.
So there's a lot of low codetools that allow you to build
(23:32):
like simple flows, very like youcan chain one prompt to another
prompt and then use Dali orsomething you can't go to
anything actually useful there.
Once you get the ability to havethe same kind of concepts of
programming, but in an interfacethat's familiar to like non
technical people the creativityexplosion is incredible.
And so we're super excited to bethis like GitHub, but for
(23:53):
prompting like we're still superearly days building out all
these community features, but onour discord, there's people
sharing prompts, like cloningeach other's prompts, being
like, Oh, what if we changedthis line?
And Oh, it worked great.
Now, like then someone elsetakes that and improves it.
And it's magical to see.
And these are not like nontechnical people taking their
world knowledge, domainknowledge, taking just
experiments that they're excitedabout and building cool things.
Jed Tabernero (24:16):
think there's absolutely an appetite for that.
Because right now, all thesecreative people using AI is some
form of interaction with a,consumer facing LLM, unlocking
that creativity.
When you talk about unlockingthat creativity, we know there's
groups of people who are willingto learn how to code, who are
willing to learn how to do theselike small things.
(24:37):
And there's always going to be alimit, right?
For example, like going tocollege, I also took a bunch of
coding classes and that's thebig one.
It's not because I wanted tobecome a software engineer, to
be honest with you, just a lotof people that I respected,
encouraged me to take datastructures just because they
told me, look, the world oftomorrow, you're going to need
to understand data structuresand tools like this, I think
(24:58):
will make that much easier forpeople like me.
Who are going into the space whooccasionally I have no, software
development engineer in my jobdescription, but occasionally I
have to write some code.
It's the world that we live intoday.
Tools, I think WordWare makethat a little bit easier to take
it to the next level.
It's not just coding.
No, it's not.
If else in loops, are youfinding there's a ton of use
(25:20):
cases for people who are tryingto get interested and learn more
about your product on how tointeract with it alone.
Filip Kozera (25:24):
Yeah.
And I think one of the problemsthat you haven't mentioned is
that people, even if they havedone a little bit of software
engineering and their brain doeswork in this kind of more
analytical way, and you can puta structure on it.
They can't deal with.
Things like deployment, theycan't show, other people that
they're the results of theirwork, which is really sucks.
(25:46):
If you take a person who can doa little bit of Python, but then
you tell them, Hey, spin it upon the website with a front end
and put a bunch of LLMs togetherand they are just going to
panic.
And this is where kind of wordwhere it comes in is we are
building the first community ofpeople where they can take their
prompting and take their ideas.
Create these these workflows orkind of little blocks of RLM
(26:10):
interactions and share them withthe community.
So we have one click of abutton.
If you, first of all, if you area CEO or an early stage startup,
or you're trying to build aproduct with one click of a
button, you get an API, whichyou can plug into the code, and
this is for some people, this istheir whole backend.
So they spin up a front end,they put the API there, and
because you can run code onWordWare.
(26:32):
You can actually do everythingon top of it, but there is
another way where you just clickshare and that gives you a
hosted application in a similarway to replicate.
com, which allows you topublicly host your machine
learning models and show themoff to the world.
Here you get to show off yourkind of word apps or prompt
(26:55):
cascades with tools, et cetera,to the rest of the world.
And also, very importantly, youallow the community members to
fork that and to duplicate thatflow and to reuse it inside of
their application.
So I think you're very correct.
Identify that there are peoplewho are analytical and you've
even probably wrote more codethan we expect our users to be
(27:19):
able to do.
But yeah, we're giving it to allthe domain experts.
So it's not only going to becode gen right now, a farmer can
write a program which will tryto optimize how he should move
the crop based on the weatherand a bunch of factors that I
have no clue about.
Neither any software engineerhas any clue about.
So that's why we want to forthese AI aficionados who are
(27:41):
early on and they're intelligentanalytical and know their
domain, give them the right toolto be able to to take their
knowledge and instill it andgive it to LLM in the right
manner.
Robert Chandler (27:51):
It's a little bit like Excel in the, Excel is
basically the most usedprogramming language in the
world.
It's probably got the mostdevelopers.
Building with it.
And it's not a program language,but it is a program language.
And the reason it's so popularis there's such a long tail of
applications that there's no waystartups or companies are going
to build the thing for you, foreveryone.
(28:13):
You end up with a regression tothe mean.
On what companies are viable andit's, where is there enough
value and like on a per usagebasis and enough users multiply
that together and you get acompany that you can actually
start and build.
But there's a limited supply ofsoftware engineers.
There's a limited supply ofentrepreneurs.
For now, once, once we get theseAI engineers and software
(28:33):
engineers, that might change.
But yeah, you need, youbasically need to put the tools
in the hands of the people thatknow their problems, know what
kind of thing they would buildif they had the tool to build
it, and then actually give themthe tool and see what they build
and, empower them.
Shikher Bhandary (28:45):
So when you say Excel, Jed's eyes light up
because when you, yeah, he'sthat guy, he's that guy,
everything in his life from likegrocery bills to stock modeling
and performance of an asset isall on Excel.
The first thing.
And it's funny because the firsttime I met him, he was doing all
these calculations of utilityspend and stuff like that.
(29:06):
And how we are trending over 12months.
And I'm like, dude, that's it's.
Up by a dollar, like what areyou doing trend
Robert Chandler (29:14):
Dude,
Filip Kozera (29:14):
you guys
Robert Chandler (29:15):
get a girlfriend.
It's
Shikher Bhandary (29:15):
Yeah.
Jed Tabernero (29:16):
We used to.
Filip Kozera (29:17):
to.
Okay.
Okay.
We live together.
We live together with Robert aswell.
He's downstairs.
Jed Tabernero (29:24):
You just do the same thing.
It's important.
Before you ask the question,Shikhar, I just want to point
out, I just, It's, you mentionedExcel is a coding language,
right?
Py, there are about 8.
2 million people around theworld who know how to code in
Python, right?
If you were to guess how manypeople use Excel in the world,
just throw out a number.
(29:44):
I just want to see what theperspective is here.
Robert Chandler (29:47):
300 million, maybe?
Filip Kozera (29:50):
200 million, something like that.
Jed Tabernero (29:51):
There are 750 million people in the world who
use Excel.
That's
Filip Kozera (29:57):
that's our target market.
That's our target market.
Jed Tabernero (30:01):
And those people are analytical.
They're analytical, right?
Inherently you have to use Excelto do a lot of the business
worlds.
I think in 10 years, people willthink about Excel as, Oh, that's
like some tool that, a lot ofthe older people use, I
disagree.
There's a lot of people who arestill going to continue to use
Excel because they can't get tothat next level of abstraction,
I think word where it can bethat tool for folks who want to
(30:23):
utilize LLMs, for example,anyway, sorry, sugar.
I completely cut you
Shikher Bhandary (30:28):
Yeah no.
I was actually going therebecause I wanted to convey that
I use Excel a certain way andJed being, a financial dude he's
used Excel in a different wayand he loves using it that way.
So I was just coming down tojust wordware as well and how
(30:48):
you've.
Mentioned that you want to belike a notion style,
collaborative environment.
So, one thing that, all our THCresearch is on notion and we got
those templates from thecommunity and it has helped us,
turn those notes into a product,which is this podcast and where
(31:11):
we actually connect founders toVCs and capital and stuff.
So it's so interesting how,You're going at it the same way
where you are creating thiscommunity, the community shares
a certain prompt and someoneelse can build a product based
on that.
automation or workflow.
Robert Chandler (31:30):
Yeah, I think there's a number of reasons.
And some of them are just that,like the connotations of notions
are great.
Everyone loves notion.
It's definitely a benchmark forhow to build great productivity
software.
The sort of more importantreasons are, it's already
familiar to people who are ourtarget audience.
So the PMs, know how to useNotion.
(31:50):
And so if we can leverage thesame affordances, then it will
be more like the learning curvebecause it is, it's still a
learning curve to use WordWare.
Like it's a new paradigm.
You're having to firstly learnhow to do prompting and then how
to use our tool.
I think hopefully the how to useour tool part is easy and you
just basically need to learn howto do prompting.
And that's why we see peoplethat have used like LangChain,
(32:11):
they can move to WordWare and belike, Oh my God, this just makes
things Way lower friction andway better.
And then the people who arelike, have not done any
prompting or agent buildingbefore go still confusing, but I
get the slash command and I getthe how to use the tool.
I just need to work out how do Iprompt the other kind of why
notion and why not somethingelse.
So there's you've probably seena few low code tools that are
(32:32):
flowchart based.
And actually the very firstiteration of word wear was built
at a hackathon.
Was flowchart based the problemyou've reached with flowcharts
is whilst they look very simplewhen you look at them at a high
level blockiness, when it comesto actually like building
something complex where you havethese loops that are going
background, you have branching,suddenly your simple looking
(32:53):
workflow that was like a fewblocks leading one to the next
turns into this horrible mess oflines going all over the place.
Shikher Bhandary (32:59):
Yeah.
Robert Chandler (33:00):
It's actually way easier and way like the
structure of software is thatit's like documents and you have
functions that call functions,which gives you effectively a
third dimension rather than thiskind of 2D grid of a flowchart
based thing.
And so that was really, it was,it came.
Through seeing the deficienciesin the flowchart model that we
were like what if it looks likesomething else?
What if it looked more likeprogramming and maybe the way to
(33:22):
make that accessible is to makeit look like Notion.
And it was very much came out ofa, almost like a design
experiment.
And then it was like, Oh waitthis works.
This is beautiful.
Like it's I can't tell you howmuch I enjoy prompting in
WordWare compared to writingprompts in the code base.
Jed Tabernero (33:35):
Thinking through kind of your target audience,
right?
Like folks who are creative, whowant to get into the technical
stuff and learn just a littlebit enough to build something
beautiful.
I'm thinking about it at anenterprise level where you
mentioned, right?
Like PMs really use Notion alot.
When I initially looked, Iwatched your walkthrough, Robert
on creating just the chat bot orthe first agent, I guess that
(33:59):
just to me was like, okay, if Ican work with them like this at
an enterprise level where, okay,let's say we have some kind of
LLM integration and I need tocommunicate with domain
expertise as a softwareengineer.
That actually makes it a loteasier for me to even
communicate to them in English.
This is exactly, I think whatI'm doing.
This is the structure that Ineed it to be.
(34:20):
I think that'd be a huge unlockjust for people who, don't want
to think about the structure.
Don't have to think about,certain things that, that
software developers have tothink about.
Filip Kozera (34:29):
So you've mentioned kind of two different
ways.
So one is the community sharingtemplates where it's the same as
notion, but actually incorporate setting, the dynamics
a little different.
So in because you can write somecode and in word, where what
these engineers are doing,they're providing you with these
blocks and you don't need tounderstand how these blocks
work, but they will tell you,Hey, this We figured out that
(34:52):
for retrieval of this type oflegal documents, we need to
choose these hyper parametersand these vector database, and
you don't need to understand anyof it, but they're going to come
to you and tell you, Hey, whenyou're working with legal
documents, use this block.
We called it retrieval argumentretrieval for legal documents,
and we've made it workperfectly.
And then there will be differentretrieval for internal
(35:13):
documents.
Just because how much of a muchexpelled retrieval augmented
generation is still and so wesee kind of engineers writing
these blocks and you as aconsultant or as a salesperson,
being able to prompt and playaround with it and see exactly
what the retrieval has outputtedand create these these well
(35:34):
working word apps.
And because you can seeeverything gets at the core of
it.
We are, what is what you getplatform, which means that you
can debug it extremely quickly.
So those kind of, those, this iswhat we are seeing from the
corporate world is that they arecollaborating and it's not.
Like the famous notion template,because the notion template
(35:54):
would be like a fully workingthing, but it's more these two
people working together.
And and it's a lot aboutconsultants as well.
So they go to the client andthey listen to their problem and
they are a little bit better atprompting.
Cause I think consultants jobsare going to change from
creating decks to creating.
Proof of concepts right now.
And I've got my own theory aboutwhy they are freaking out right
(36:17):
now and why is it moving soquickly?
So I can dive a little bitdeeper into that, but they
essentially booked a lot ofrevenue and in order to create
good return on investment forall of these companies, they
need to be doing something elsethan the PowerPoint deck,
Shikher Bhandary (36:31):
Before we close, I want to give you the
stage to shout out the work thatyou're doing, how people can
reach you, hiring or funding,and just want to give you the
stage to shout out your teambecause it looks like an
incredible product and you guysare crushing it.
Filip Kozera (36:47):
thank you so much.
So for any listeners Of thingshave changed podcast.
We've got 250 of credits onWordWare.
So we, we discussed this, sojust enjoy it.
Just try all the differentmodels go to WordWare.
ai.
It's like software, but withWord and just sign up for anyone
right now the signups will befully open and enjoy the
(37:09):
product.
As mentioned, anyone can buildmore complex AI agents.
Make sure you go through theonboarding process.
As we've mentioned, this is anew programming language.
So it takes the 5 10 minutes.
Don't expect to be able to jumpin and start creating amazing
agents.
It is still somewhat complex todo these things.
So that's one thing we are alsoactively hiring.
(37:29):
I'll let Robert do the kind offoundation engineer job
description, but whoever is veryinterested in the community
about agents, just reach out tous at founders at wordware.
ai.
We have multiple roles open andit's actually quite in a very
early stage startup.
It's all kind of fluid and mixestogether.
But on the engineering note ofhiring Robert will say a little
(37:51):
bit more and lastly we arehelping a bunch of enterprises.
If there are any enterprisecustomers listening to us we're
more than happy to give you freeconsultancy at the beginning and
help you figure out how todeliver on that ROI and all that
book revenue and for many otherpeople build co pilots and
workflows that will make yourenterprise more efficient.
Robert Chandler (38:12):
Yeah.
I think founding engineer wisethere's all sorts of incredible
things to build.
We're a full stack TypeScriptstack.
We love serverless.
So we love doing things asefficiently as possible worrying
about building the value addstuff, not the undifferentiated
things that, everyone has tobuild.
If you love building things, ifyou love making things, if
you're excited about the worldof agents hit me up on Twitter,
(38:35):
Bertie underscore AI or email.
Robert at WordWare.
ai and tell me why you'reexcited to build on WordWare.
Jed Tabernero (38:44):
Beautiful.
Guys,
Shikher Bhandary (38:46):
want to mention one thing.
You guys stack team so well.
Where Philip takes a certainsection and then Robert takes
the other section.
Even on the message front.
So
Robert Chandler (38:56):
Years of practice.
It's we've known each other for10 years.
It's.
Jed Tabernero (39:00):
yeah, no, it's, it was really fun guys.
So much in this world that we'rereally curious about, but really
appreciate that.
The information and opinionsexpressed in this episode are
for informational purposes only.
And are not intended asfinancial investment or
professional advice.
Always consult with a qualifiedprofessional before making any
(39:20):
decisions based on the conceptprovided.
Neither the podcast, nor iscreators are responsible for any
actions taken as a result oflistening to this episode.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.