How AI Is Changing Developers and What They Build—with Ray Deck

Is software dead? Are engineers going to disappear?

In this episode of Futureproof, Xano CEO Prakash Chandran talks with Ray Deck, founder of State Change and longtime technologist, about how software creation is evolving in the age of AI. Ray explains why the future isn’t about replacing developers or building the right SaaS — but about rethinking how we define both of those terms. Together, they explore the evolving profile of a developer and why software might be ephemeral, but the data and customers behind it are constants.

• AI as amplifier: AI helps shorten the path between concept and creation, expanding the definition of what a developer is, emphasizing the importance of AI literacy and putting a premium on creativity.
• Data Is the foundation: Your most durable advantage isn’t code — it’s the data and customer context that power your products.
• Think big, build small: Big ideas will increasingly be delivered through targeted, temporary tools designed to solve precise problems.

Episode ID: 18126513-how-ai-is-changing-developers-and-what-they-build-with-ray-deck

Subscribe to Futureproof wherever you get your podcasts.

From Xano - The fastest way to create a production-ready backend for any app or agent. Xano unifies AI speed, code control, and visual clarity, so you never trade reliability for velocity. Sign up for free today.

Episode Transcript

Ray Deck (00:00):
My my spiky opinion is that SaaS is dead.
The idea of SaaS is you have anapp.
People are going to come useyour app.
They're going to put their datainto your app, and then they're
going to get value from havingto put data into your app.
That's like the classic sort ofcrud structure.
The agentic aspect of what youwere just describing is what
really changes things up.
I don't want to go to yourapp. I want your app to come to me.
Hi, my name is Prakash, CEO of Xano.com.

(00:40):
Today I'm joined by Ray Deck.
Ray is a veteran technologistand founder of StateChange.ai.
With over 25 years in software,he got his start in low code
back in the day when it wascalled rapid application
development, later training as adata scientist and leading
implementations fororganizations from Fortune 500s
to NASA.
Along the way, he's foundedcompanies, run large consulting

(01:02):
projects, and built a reputationfor helping others become
software creators.
At State Change, Ray nowfocuses on the hardest 5% of
building software, the trickychallenges that still remain
even in the age of no-codeautomation and AI.
Beyond all that, I'm happy tosay that I consider Ray a friend
and a mentor and a formermastermind member.

(01:23):
So Ray, thank you so much forbeing here.
Thank you so much, man.
I'm so glad you're doing thispodcast.

Prakash Chandran (01:27):
Yeah, it's gonna be a lot of fun.
Um I thought we'd maybe startthe conversation for those that
don't know you on a little bitof background, your origin
story.
So I'll let you go ahead andkick it off.

Ray Deck (01:37):
Uh yeah, sure.
So, I mean, I went to schoolfor what was supposed to be
political science and then gotuh, you know, roped in by this
guy named Ed Tufty, who I didnot know was gonna be one of the
founding fathers of a sciencewe would later call data
science, uh term I didn't reallyget to learn for the first 10
years of my career.
Um, but learned pretty quicklycoming out that like uh I did
some consulting, did some ofthis and some of that, really

(01:58):
got bit by the startup bug, butwanted to go build my own
businesses, and then juststarted building one after
another over the course of mycareer.
Um, and the um, you know, builta pretty successful uh uh legal
tech company um that focusingon you know large, you know,
helping large law firms.
Um and then uh really, reallyaround the pandemic, which is
around when we met, you know,was trying to figure out what my

(02:19):
next act was going to look likeum and did uh did a little bit
of a of an experiment and umdiscovered a couple different
ways that like, you know, thatpeople were finding value out of
technology.
Crypto was one, AI was one.
That's actually what led me tothe NASA thing.
Uh, and then uh and also thisworld of you know, low-code, no
code, which sort of ties intosomething I've always really

(02:40):
enjoyed, which is teachingpeople and growing people and
like the helping uh you knowpeople who saw themselves as
non-developers and non-technicaluh become you know successful
and be able to turn their ideasand their the the engines and
their mental models into uh intoproducts that they could have
in the in the world.
And uh I've done some of thatwith Xano, some of that with
some other tools, um, and sortof that's what I've made into a

(03:03):
practice that called StateChange now, which is a
combination of a uh communitymentorship service and uh and
absolutely and now AI bot, uh,that is about helping people,
you know, apply these mentalmodels that I've learned over
the course of the last quartercentury.
I feel old when I say that outloud, uh, to um to so they can,

(03:23):
you know, move faster.
And sort of our line is it'syou know, 80% of this stuff is
easy and 10% of it seems hardand 5% of it is impossible.
And then we're about crackingthat last 5%.

Prakash Chandran (03:34):
Yeah.
Um, you know, I think I got thebenefit um and the joy of
watching you kind of build statechange.
And um, I think that you have aunique ability to bring like to
bear all of your experience asa computer scientist and an
engineer and an architect anddistill down those learnings in
a very like understandable andaccessible way to help people

(03:56):
get over that 5% hurdle.
So maybe let's talk about statechange a little bit before we
talk more broadly about AI.
How many people do you feellike you've served at this
point?
And are there any common themesaround what uh problems people
uh tend to solve in that last 5%mile as they're building
applications or business valuefor themselves?

Ray Deck (04:18):
Yeah, sure.
So uh I think state changeitself as a practice has served
maybe 500 different logos atthis point, um, you know, teams
of people within institutions.
Um, most of them probably like,you know, entrepreneurs who are
sort of you know transformingthemselves, uh, as well as, you
know, some, you know,medium-sized and some, you know,

(04:38):
larger, uh, larger institutionsas well.
Um, and the um, and but but thebut the nature of the problems
they run into have a lot more incommon uh than they are, you
know, super different.
And it really kind of boilsdown to machines aren't like us.
Uh and there are aspects ofbuilding software that are very
intuitive because that's thepart where the machine is more

(04:59):
like us.
Or part of Barrier to say, thisis where we're bending the
machine to our will.
And a lot of front-enddevelopment looks like this,
right?
Where we're trying to createthings that a human being will
know how to use, and there areissues of user experience and
design that are actually fairlyintuitive when you just sort of
lay them out for people.
Um, like, you know, keepingsimplicity, white space,
organization, though those kindsof things.
But then there's a design formachines.

(05:21):
And machines don't think theway that people do.
Uh, even though they arethinking things and we are
thinking things, uh, the waythat humans think or animals
think in general is just verydifferent from the way the the
machine works and the way that amachine computes, right?
Because we we call themcomputers.
They do math of a very specifickind.
Uh and the and when someonetries to attack it as if it were

(05:41):
just sort of a person doing it,you wind up getting uh getting
a little bit uh sideways.
So it is when we think aboutthat in terms of like front end
versus back end, it's oftenabout the back end.
It's often about like how doesdata get managed?
What you know really drivesmemory?
How does a computer work thatit can go faster and what causes
it to go slower?
And you know, a lot of thesethings boil down to you know

(06:03):
mental models we built along theway that we also have inside
state change.
But that but being able to dealwith the alien when it's acting
in its most alien way.
And AI has sort of you knowredoubled that because uh, as
much as natural languageprocessing looks normal, it
actually is a very weird beastunderneath.
Uh, and so getting our armsaround that becomes an

(06:24):
interesting source of uh ofchallenges, you know, as well.
So I find that's usually wherethe hardest 5% is, is getting
people to think more intuitivelyas the machine does.
Although I will say that overthe course of the, I guess about
three years I've been doingstate change, the um the number
of people who come in and sortof intuitively understand more
about how the machine works hasreally increased as I'm seeing

(06:47):
more people who are coming inwho have, you know, maybe they
did a bit of coding, uh, theylearned how to do JavaScript or
had a Java course or somethingback maybe when they were in
school 15 years ago.
And so they can have anintuitive sense for what could
be going on here, but not quitehow to articulate it to the
machine in a way that's gonnaallow them to make the kind of
progress they want to.

Prakash Chandran (07:05):
Yeah, this actually leads me to one of the
questions that I wanted to askyou, um, which is well, let me
frame it.
I, you know, I think thatespecially with the introduction
of these uh vibe coding tools,a la bolt, lovable, et cetera,
we've seen like the widening ofthe aperture for uh these new
types of builders that are ableto build, prototype, and express

(07:26):
their ideas.
Uh, kind of amongst them, wekind of have people that are
trying to tackle that last 5%, alot of the people that come to
statechange.ai.
In your mind, like what is thedefinition of like a developer
these days?
Like, how would you how do youthink about uh the new type of
developer, the people that arecoming to State Change?
And uh how do you see thatchanging over the course of the

(07:50):
next couple of years?

Ray Deck (07:53):
Yeah.
Uh so there were there was thisline, you know, when the iPhone
first came out, um, or orshortly after when the App Store
first became a thing, theiPhone.
There's an app for that.
And now pe and I think it'sworth noting that the iPhone's
been around long enough that ifit were born the day it was
introduced, it'd be old enoughto drive today.

(08:14):
Yeah.
People who have had it fortheir entire professional
careers, their idea of softwarehas gone from like my
grandfather's idea of software.
He actually worked with likethese these big mainframes and
insurance companies.
Uh, and the um, and like so hisidea was like these machines
that would fill like the floor,right, of a big skyscraper
building.
Um, you know, my my my my dad'sidea of computers, you know, he

(08:37):
was actually a little bit of apioneer like first early on to
like using like the original,you know, Apple Macintosh and
even the one of the firstversions of the IBM PC that came
out.
And his idea was like it's arelatively heavy piece of metal,
right?
With an ugly screen or whateverthat's sort of sitting in front
of you that you are that thatyou're then trying to bend
yourself to make work andcommunicate through this very
loud, clackety keyboard.

(09:00):
But the people who have beenworking with technology for the
last 15 years have had it intheir pockets and it responds to
touch.
And more recently, in thecourse of the last decade, it
responds to voice.
It is something that they havecommand over.
So now instead of a, oh, it'sthat big thing that I couldn't
possibly hope to understand, Imight only ever hope to wrangle

(09:20):
it a bit.
They say, no, not only there isan app for that, but there
should be an app for that.
And that's a small thing.
I should be able to make theapp for that.
That should exist and be ableto serve me rather than it being
the giant colossus that I needto serve.
And that attitude towardstechnology, what Malcolm
Gladwell calls entitlement, thatI think is the big difference

(09:43):
in what I've been seeing, evenjust increasing over the course
of the last few years with theway people have been uh uh using
and approaching technology.
They come in with this ideathat there should be software
that does this, uh, and thatthey and that they themselves
are empowered to do it and theyshould be able to make it
happen, and there should betools that they can have command

(10:04):
over that will allow them to dothese things as well.
That change in attitude is, Ithink, what defines a developer.
A developer is someone who canmake software to do their will.
And that um, and and then thatdescribes a much broader set of
people, partially because ofwhat's been in the air, the
technology that the you know,the the the technology over the

(10:24):
course of the last you knowquarter century of this um, you
know, the start of thismillennium.
The uh the way it's become somuch cheaper, so much smaller,
it can be fitting in ourpockets.
The way that um the factthey've had some training with
this, usually in school, if notprofessionally, the fact they're
using these things every day,both at work and personally.
Uh, and then they say, yeah,sure, I can do that too.

(10:46):
And that's um, and and thatidea of um again, you know,
using that word entitlement, Ithink that is what allows many
more people to becomedevelopers.
And then the question is, youknow, who's gonna be able to
help them be able to uh torealize and execute on that
vision.

Prakash Chandran (11:04):
Um, and then the the person or the persons
that would help those that areentitled, like this new type of
developer that wants to kind ofwill that app into existence.
That's kind of at least what Ithink about as like software
engineers, people who have hadall of the education and
literacy to be able to kind oftake the initial spark of

(11:24):
creation and then get it overthe hump.
Would you sit, do you see theworld in that same way?

Ray Deck (11:29):
Well, I think software engineers, as we might have
understood them before, peoplewho were trained to do this and
says software engineering is myjob, right?
And is it was how I definemyself.
And like, you know, the what isgoing to be used for, that's
someone else.
I but my specialization is Iknow how to make software,
right?
And the the gag is that the newdeveloper says, I understand

(11:50):
this business process, Iunderstand this domain, right?
And I can bridge thatunderstanding to software, maybe
mediating it.
Previously, you have said Ihave to mediate this through the
software engineer who will thenbe able to take my ideas and
express it in the form ofsoftware.
And now it's I can, I can moredirectly express it as software,
you know, myself, maybe throughsomething as simple as doing it

(12:12):
through an Excel formula, orsomething as complicated as, you
know, creating a, you know, bigapp or automation or AI agent
or whatever it is, right?
Uh, but that they don't that itI mean, this this is a term
that showed up a lot in like thedot-com boom, right?
The idea of disintermediation.
We don't need the middle personanymore to create the software.

(12:32):
We can instead say, I have anidea.
I can then express that idea assoftware and have that come
directly.
So the, the, the, you know, youI think you're right to say
that like, you know, if we weretalking about this a decade ago,
we would have said softwareengineers are the people who are
doing this.
They were trained to do thisand they learned how to do this.
They could put in the timecommitment to do this.
But now the uh the ex antetraining required is less and

(12:56):
the time required is less, whichmeans software engineering goes
from being a, you know, adefinition of a professional to
a thing that I'm doing onTuesday.

Prakash Chandran (13:07):
I think that makes a lot of sense.
Um, I want to touch onsomething that you you said
earlier, which is around kindof, you know, the people
increasingly that you speak withhave a more intuitive sense of
how to kind of wrangle uh themachine.
And this kind of touches onlike the general topic of like
AI literacy and um what you needto know now in order to be

(13:31):
relevant and to, if entitled, beable to use these machines in
the most productive way.
Let's talk a little bit aboutyour thoughts on that.

Ray Deck (13:41):
Yeah, sure.
Uh the the I I I I think I'veI've made the analogy before of
AI to literacy, right?
Literacy really changed the waythat you know people could uh
create value, you know, insociety and the the way that
ideas could create value insociety.
Because previously, an idea wasonly one that you could share
with somebody, and if theydidn't remember it, the idea

(14:02):
died, right?
Then you have the ability towrite it down, the ability to
read, be able to build build onother people's ideas.
And I sort of see AI providingthat same kind of value into
society, particularly like thethe uh natural language uh you
know flavor of it.
Um I'm teaching my children,you know, I homeschool my
children, um, and we're workingon some fairly, you know,

(14:22):
complicated ideas, sometimesideas I don't totally get.
And, you know, I would say fiveyears ago, in order to
understand this, I would like goto Google and I was okay,
where's a good article orsomething on the subject?
I'm dependent on like who mighthave read uh uh put something
together and like I have to reada bunch of pieces in order to
sort of understand the topic.
Now I can go to AI.
AI will take those, whateverthose resources are, distill

(14:46):
them into the aspect of myquestion, trying to help me
understand just the thing that Ineed to know at that moment.
And my access to ideas as aresult is really much more
refined.
And I can learn a lot more andwork with a much bigger set of
ideas in a much smaller space oftime.
And that time thing, like withsort of using my Tuesday joke

(15:09):
from before, is the reason why Ino longer need the is the
difference between needing to bea lifelong student of a
discipline and being able toincorporate that discipline into
whatever it is that you'redoing today, right?
And that ability to remix ideasand disciplines is what I think
is really um, you know,transformative here, right?

(15:29):
In terms of being able to applyAI to create a whole lot more
value.
Like one thing they they werejust researching, like, you
know, debt equity ratios and howit relates to like, you know,
the largest companies in theworld.
And like the the the insightthey were able to get very
rapidly was just astounding tome.
One, of course, they're sparkkids.
Um, but because they could, youknow, make use of these

(15:51):
resources, they had access to alot of data.
But rather than having to siftthrough all that data
themselves, they could use theleverage of the AI on top of it,
and then use the leverage oftheir own insight on top of that
to come up with like reallyremarkable understanding in a
very short period of time, whichmeans they can combine it with
other kinds of understandingalong the way, which allows for
much bigger intellectual unlock.

(16:13):
And that's that that's sort ofthe the uh the the literacy
aspect, I think, of that, thatgets me really excited.

Prakash Chandran (16:18):
Yeah, that's fascinating.
But a part of that isunderstanding like how you can
properly interrogate um the AIto kind of narrow in on exactly
what you're looking for.
And this kind of speaks towhether it be software
development or just kind of morebroadly, a paradigm or
framework around like how youcan work with the machine to get
the results that you need outof it.

(16:40):
Yes.
I know you have a lot offrameworks around how you think
about this.
Maybe there's one that youmight share for any like
technical builders or people whoare really starting on the
frontier of leveraging some ofthese tools to work with them.

Ray Deck (16:51):
Yeah.
Uh, you know, the I the um thethe the the framework we keep on
coming back to most often instate change, and especially in
the context of softwaredevelopment, are you will build
this again.
The idea that you are going tooperate iteratively and you're
trying to get to the nextanswer.
I run into so many people, bothin the world of software, but

(17:12):
also just like, and this happenslike just in the world of
ideas.
You're trying to get to thefinal answer.
How can I rifle shot this rightnow and get my final answer?
Um, but the thing is, you'renot ready for the final answer.
You have ideas that need to getinto your head in order to get
from here to there.
You're not even ready to askthe question yet.
And so the the what I find isthat being willing to ask more

(17:34):
questions that then get you thatone step further.
And that's another questiongets you another step further,
is a model that uh I think whenpeople are are afraid it's gonna
take a long time to do allthis, they become much more
reticent to ask those questions.
And this is where like AI andthe speed aspect really changes
that, because now I can affordto ask more stepwise questions
and then get to the next place.

(17:55):
Because that wasn't the wayresearch worked previously,
people are more reticent to dothat.
They're like, oh, that thing'stoo far away.
There's no way I can get that.
I want to go make that somebodyelse's problem.
But you can make it yourproblem and get to the get to
that like at the end of an hour.
It is amazing what you can do.
I think the the most importantskill is going to be the

(18:16):
willingness uh to be the the thethe willingness and certain
amount of technique of askingsemi-open questions, right?
Questions that like are in yourdomain, right?
You're not allowing things togo completely flop open, but
that are uh they're focusing youforward, but that are keeping
open the the weirdpossibilities, you know, that

(18:38):
are in front of you.
And that's a balancing act.
And that's something that Idon't think that many people are
good at right now, uh, but isdefinitely one of the things we
teach in like the mental modelsthat we do in SageChain,
certainly something I'm tryingto develop because something
provides even more value todaythan did yesterday.
I would say it's probably truethat like you for an
intellectually curious personshould be asking questions in

(18:59):
general.
Yeah.
But I think the technique ofasking those questions is
shifting as we use AI.
The uh the the the speed atwhich you can ask questions goes
way up.
The types of questions that'sable to answer well shifts a
bit, as you might imagine.
That's also somewhat sensitiveto like how these models work
and what have you, uh especiallylike in the multimodal aspects

(19:19):
of it.
Um, but the um, but the valueyou can get from just being able
to go all the way down that uhpath is just extraordinary.
Like another line I have islike, you know, data is like the
new oil, right?
And that AI is a refinery forthat oil.

Prakash Chandran (19:35):
Yeah.

Ray Deck (19:36):
And that you you now get access to more of the value
from that oil, that data, thatknowledge that your company has,
that you have, that like theworld has, and that you can put
together, you can put to put itto work for you in a much more
refined way in a much shorterperiod of time because of this
technology uh that is, you know,growing every minute while

(19:56):
we're having this conversation.

Prakash Chandran (19:58):
So you mentioned something there, like
data uh kind of as the new oil.
And you know, we were at theGartner conference a couple of
weeks ago, and uh there's a lotof talk around AI readiness and
how that relates to datareadiness.
Um, you know, I think that whenthey talk about that,
oftentimes they're talking aboutdata quality and there's a
cultural component of that.

(20:19):
I'm curious as to your thoughtsaround, you know, the broader
term of AI readiness, theimportance of data as it kind of
relates to kind of gettingorganizations to leverage AI in
the right way, to get theanswers that they need.
Um, yeah, talk about your uhyour thoughts around data or AI
readiness and data.

Ray Deck (20:39):
Yeah.
So um, you know, this is notour first turn of the wheel for
thinking about data and theselarger organizations.
Big data was a topic of the daya decade ago.
Right.
Uh that had a lot to do withthe fact that these big
companies, like Amazon's a goodexample of this, right?
It has tons of data about howpeople engage, go through their
stores, what they buy, um, andthey try to use it in various,
you know, semi-competent ways tobe helping people buy more,

(21:02):
find the stuff they're really,you know, looking for, um, et
cetera.
Uh, and the the trick is thatso much data is terrible, right?
That the the the it is it is umyou know, it's it's exhaust
data, and sure we have access toit, but like what of it really,
really matters in here and whatis driving um value.
So a lot of the the data Ithink is is most exciting is

(21:24):
data that was not exciting inthe big data era, but like is
for what we now call maybe youknow uh unstructured or
semi-structured data, you know,like text, things that reflect
knowledge.
Um and the and and I think oneof the when when I think about
like AI readiness for theselarger organizations, it's where
they suddenly realize, oh mygoodness, there's all this stuff

(21:44):
that we're not capturing thatis actually valuable, right?
That we're capturing the wrongthings and we're keeping the
wrong things.
Um, I'm imagining uh, you know,that you, like many sales
professionals, are doing thingslike recording your meetings,
right?
And then you're running throughthem through analysis.
Now, the analysis is fine.
That's the refinery part,that's the AI part, but the way

(22:05):
you get value from that is byrecording it in the first place.
And those recordings would havebeen verboten in most of these
companies coming into this.
Like, oh no, how do we getvalue from this?
How do we get value from all ofthese interactions of our
knowledge workers that we'vebeen doing all the way through
here?
Previously, like big data wasreally about what were our
customers doing, uh, maybe whatwere our line employees doing,

(22:25):
things that could be turned intonumbers relatively easily,
because that that kind ofstructured data fit easily into
databases and also could be runthrough more traditional
statistical techniques.
But one of the neat thingsabout AI, something that's sort
of you know discontinuous aboutit, is that this unstructured
data, images, video, um, youknow, the words people have

(22:46):
written, or probably the mainmedia that we usually think
about, um, the sound video andtext, um, those those create a
lot of value and allow us tofigure out, okay, what else was
this customer doing?
And when were they doing?
How can we associate that withtime?
Uh and the uh and what was whatwere what were we doing?
Can we clone our people?
Can we take some of our peopleare really in the business of

(23:08):
like just restating corporatepolicy?
Do we need people to do that?
Can we have a machine that isable to just restate that policy
and be able to help the client,you know, work through our
policy to figure out how to begetting the most value that
they, that they, that they can?
All of this requires capturingmore of that unstructured data.
So when I go into a lot ofthese companies, what I'm seeing

(23:29):
is that they have some of thisdata and then a good fraction of
it, they just don't have.
And the most important thingfrom an AI readiness point of
view is being able to catch thedata that when they last did
this a decade ago, they weren'tcatching, right?
Because all of a sudden therelative value has shifted
between these things.
So the and and like the it'sit's a little bit like, hey, if

(23:51):
you got oil in your field, yougot to put in, you got to put in
the pumps, right?
We need to turn it from justlatent value into actually
captured value.
Because if you haven't pumpedout the crude, it won't do any
good at the refinery.

Prakash Chandran (24:01):
Yeah.
So the importance, there's likewhere data may have been uh not
as important.
You don't, you didn't haveleverage over it now becomes
exceedingly important.
So if there's one takeaway,it's get the data, you know, uh
start attaching as much data asyou can because the more context
that you have, the more you canleverage the machine to help
you.

Ray Deck (24:20):
Right.
I I I think that that that'sthe key.
For most folks, I have foundthey focus a lot on how can we
refine it better, how can wehave better models, how can we
have better prompts, whatever.
But it's all a multiplier, amultiplier on the data.
And when they're not collectingit and they're just letting it
fritter away, which many are,uh, that is, I think, the uh,
the the key missing part.
That's where like a lot of youknow opportunity is for, you

(24:43):
know, uh, you know, the robotsto be able to come in and be
able to make sure the stuffdoesn't um uh doesn't float away
because traditional datawarehouses haven't been solving
this problem so far.
Not because they'restructurally incorrect, but
because they've been saving thewrong stuff.

Prakash Chandran (24:57):
Fascinating.
Um, I want to talk aboutbasically maybe accessing the
data and maybe through the lensof um just kind of something
that I think we've seen justshift dramatically.
So obviously, search behaviorhas changed quite a bit.
We've gone from potentiallyusing something like Google to
now just asking Chat GPT or uh,you know, perplexity or

(25:18):
something, give me the answer.
And it's amazing across theindustry how much top of funnel
has dropped, intent hasincreased, but wow, how that has
shifted.
And we're it's not really faraway, and you can already see
this happening.
It's not just asking, hey, giveme the answer, but do this
thing for me, right?
In this world where we areslowly moving to like when we

(25:41):
talk about agencera, I'mwondering what you think about
that modality, like the new waypeople are kind of going to
consume the data and consume theinformation.
And what does agentic mean toyou in that regard?

Ray Deck (25:55):
Yeah.
Um I the the my um my my spikyopinion is that SaaS is dead.
Um the um because the the ideaof SaaS is you have an app,
people are gonna come use yourapp, they're gonna put their
data into your app, and thenthey're going to get value from
having put data into your app.
That's like the classic sort ofcrud structure, right?

(26:17):
Of of one of these SaaSapplications.
Um the agentic aspect of whatyou were just describing is what
really changes things up.
So I'm not, I don't want to goto your app.
I want your app to come to me.
I don't want the the theenvironment in which I work with
software would have been uh,you know, 20 years ago, it's
your desktop, it's your Windowsdesktop, right?

(26:37):
Because Mac is sort of nowhere20 years ago.
Um in uh 10 years ago, it'syour browser.
Right.
And you know, the way youaccess software is by opening up
the browser and going to acertain URL and then getting to
there.
And that's one of the reasonswhy search engine marketing made
so much more sense becausepeople are looking for websites
and now you're able to find thewebsites, and that's what SEO
and that's what you know,PageRank stuff and whatever all

(26:59):
that stuff was about.
But now I don't want to use anyof that.
Like I want to just getsomething done.
And how do I get that thingdone?
I get it done by working withmy AI, right?
I can go just talk with Claude.
Hey, Claude, go take care ofthis problem for me.
Or Chat GPT, or Dia, which isactually uh an AI that's built
into a browser, is what I'mtalking to you on right now.

(27:20):
Um, or Perplexity, which has anoffer called Comet, which is
its own built-in browser to beable to integrate these things
so that you can be doing thework.
And that idea of of bringingthe workdoer to the professional
or to the consumer is, I think,the the the process that we're
still in right now.
Uh and that is the and andthat, and if that's gonna be

(27:42):
where work gets done, thenthat's probably also where
commerce gets done.
Uh there's a there's there's agreat book called Uh When
Machines Become Customers.
It was a really pressure bookthat was written a couple of
years ago about this subject,about how AIs would be able to
start buying things on behalf ofpeople, you know, to to to do
work for them, right?
Uh and the um and and and youcan just have the AIs do these

(28:06):
things for you.
And the question like, how dowe give them permission?
That becomes a problem tosolve, right?
How do we give them a wallet towork with if they want them to
do commerce on our behalf?
How do we give them access tothe tools that we need?
And the uh and the standardright now that sort of is most
exciting in this regard thatallows these AIs to go talk to
these various services is calledmodel context protocol, MCP.

(28:29):
And MCP is, at least right now,I think the the most exciting
thing in agentic AI because itexposes tools and resources and
prompts.
Those are the three big thingsin MCP, to um to empower an AI
that's connected to a person todo things on their behalf.
So that my chat GPT isn't yourchat GPT.

(28:50):
My chat GPT is chat GPT plusaccess to my notion, plus access
to the books I read, plusaccess to the My Gmail or other
things I care about.
And it's able to both bring meknowledge that I care about, you
know, either that is private tome, contextual to my company or
from the world, and be able todo things on my behalf that uh

(29:12):
through through my being able toissue commands.
And now that becomes the waythat I'm interacting with
software.
Instead of from an app on mydesktop or a site on my browser,
it becomes a tool that I can beinteracting with, you know,
through the agentic interface,which could either be in the
form of straight text or becauseof a newer technology called
MCP UI, which is a reallyexciting development from 2025,

(29:33):
um, being able to start tobring, you know, micro apps that
are then being uh uh uhinterleaved in with the um with
the experience of working withthe AI and the um and then like
software that's just made ondemand to help me solve this
problem right now.

Prakash Chandran (29:49):
Mm-hmm.
Talk to me in this world thatwe're moving uh towards and kind
of like the foundation that youlaid out, what is the job of
the or what does the front endlook like?
Like, Is it a mobileapplication?
Is it a web application?
Is it a uh is it somethingelse?
Is it just that we're gonna doeverything through cloud?
Like how do you see the frontend evolving these days?

Ray Deck (30:11):
U well, uh that MCP UI thing that I alluded to is
actually what I think is themost exciting development in in
in user interface.
Because now instead uhpreviously it was a, hey, you
either have this flow of text,right, that's coming through the
application, uh, or you aredealing with like a traditional
app.
Or sometimes you have like theco-pilot modality where like
you've got the you know, thesidebar of like, you know, I've

(30:31):
got my chat going on over here,and then maybe the other 75% of
the screen is working with theapp.
Like that's uh pretty commonsince like the days of like
GitHub Copilot working withVisual Studio Code, and you see
that in some other apps now too.
Um the but the but but I thinkthat the way we're gonna see it
is the is the reverse, uh, wherethe the primary mode is going
to be talking and working withthe AI, and the AI is able to

(30:53):
bring micro apps in context, youknow, to you.
So we'd have like a, you know,be able to have a window that
maybe could expand up to be ableto have the podcast going, uh,
be able to have informationthat's being delivered, you
know, over the top, that is, youknow, as it has identified
opportunities for me, kind ofclouly style, uh, being able to,
but the but the but but but thethe main controller for all

(31:16):
this is an AI who's working forme.
Then that's important that'sgot to be working for me because
you don't want to be giving upcontrol.
But the um, but like I'mimagining I gotta tell you, like
when I'm using my computerright now, and I'm a fairly
sophisticated user of thesemachines, um, most of my tabs
that are open in DIA are to thechat, not to people's sites.

(31:37):
And that's because that's wherea lot of the value is.
And I think what we're gonnastart seeing is through the use
of this kind of micro app, youknow, integration, uh, which
will initially be decried asthat's not a real app and that's
not a real front end, which wasthe same thing as what people
said about web apps 20 yearsago.
Uh, I think we're going to seethe same sort of you know
evolution of like instead ofsaying, oh, well, the browser is

(32:00):
just an app that's inside thedesktop, why would you want to
have apps inside that?
People will be saying the samething for a while.
Traditionalists will be sayingthe same thing about like things
are integrated into the AI.
But I really think that's sortof where the where the
modalities are shifting.
And we'll see that both on, youknow, on the desktop, probably
with a more textual interface,because desktops tend to have
more of that.
And then on the phone, uh,we'll see it probably with some

(32:21):
text, but I'm expecting voice tobecome a very important part of
that because key typing is areally hard thing to do on
mobile, uh, as well as uh makinguse of the camera.
So I'm expecting really the um,you know, the the the
microphone and the camera tobecome uh the most important
ways to be interacting with aphone um, you know, heading into
the next couple of years.

Prakash Chandran (32:41):
You know, we've um we kind of touched on
this briefly, and I'd love foryou to dig in a little bit more
um around kind of this uhpotential around like ephemeral
software.
Uh in this world, maybe it'senabled by something like an MCP
UI, where it's really use casedriven um and it serves its
purpose.

(33:01):
Um, how does the world ofsoftware look when you have
basically these one-shots oflike, hey, we're gonna give you
something and then maybe it'snot needed anymore?
Talk a little bit more about umthat because I thought it was a
really interesting take fromyou.

Ray Deck (33:16):
Yeah.
So the the why do we usesoftware?
We use software to get jobsdone, right?
And like a lot of the timewe're using some piece of
software that was written with,you know, a million other people
in mind uh in order to do thejob that we care about.
And then we are the frictionpoints we're dealing with are
trying to get through the, youknow, the weirdnesses of like

(33:37):
whatever Microsoft decided wasgoing to be good for me and be
good for, you know, 500,000other people as well.
Right.
And that had a lot to do withlike the cost of building
software.
And you know, software used tobe something like we would talk
about like software engineersbeing a being a career path,
right?
And by the way, that careerpath is not going away.
But like, you know, for uhsince almost all software had to

(33:59):
be built by professionalsoftware engineers, uh, that
became expensive to build.
It would take months or yearsto build a good piece of
software.
And then once you set then onceyou built it, you have to sell
a lot of it in order to pay forthat back.
So that means it's got to begood for a lot of people.
And that means instead of beingreally excellent for you, it's
got to be good for you, me, andthe five, 498,000, 499,000 other

(34:20):
people.
So how do we um the the theopportunity is that uh instead
of having to be downstream ofmass-produced software, we can
be downstream of bespokesoftware.
The most expensive, the highestend suits in the world are made
on Seville Row.
And Seville Row makes each suitcustom.

(34:40):
It's called bespoke clothingmanufacture.
And the, and if you uh if yougo there, you will get one that
fits you perfectly and will beuseful for just you.
And then you'll probably wantto keep it for a while.
But that idea of like havingsoftware that's just for you or
just for a smaller audience ishas been sort of a holy grail
for software for a while, right?

(35:01):
That's what like the rat uh theum rapid application
development stuff was reallyabout.
It's like driving down thatcost of software so I can solve
it for a smaller audience that'suseful for a shorter period of
time.
And then we sort of see thathappen again with like low-code
software is like again trying todrive down the cost, be useful
for a smaller set of people fora shorter period of time,
basically shorter depreciationschedule is what accountants

(35:22):
would say.
And then low-code note, andthen like, you know, the no-code
movement of like 2022, thiskind of thing I associate with
Xano again, shrinking that down,shrinking that down.
And then AI, which has beenshrinking that down in like five
waves over the course of thelast two and a half years, has
been just madhouse how fast it'sbeen going.
And it keeps on shrinking down,shrinking down.

(35:44):
So that now you can be, youknow, creating meaningful
software in an hour, which mighthave previously, which in the
previous modality might havetaken a couple of months.
Now, that's not software that'sgoing to be good for everybody.
And it's my main softwarethat's gonna be good for you for
a really long period of time,but it's based on your superior
understanding of this is the jobthat I need to do, and I need a

(36:06):
machine to help me do it.
And then I can execute that,and it's good for that job.
And at that point, I don't carewhat's good for after that
because we drove down the costof creation by so much.

Prakash Chandran (36:17):
Yeah.

Ray Deck (36:18):
And the difference between disposable versus
long-term.
Yeah, sorry, go ahead.

Prakash Chandran (36:22):
No, it's it's interesting in that world of
kind of Malcolm Gladwell'sentitlement.
It's like, this should exist.
I'm gonna make it.
Okay, now I'm gonna throw itaway now that I'm done with it.
Um, that's where we're movingtoward.
No good.

Ray Deck (36:35):
When software is when software is something you create
on Tuesday, then you're gonnause it on Wednesday.
And on Thursday, some otherproblem comes up, right?
Yeah.
Your your your view of it, thethe same thing that creates that
level of entitlement alsocreates this ephemerality to it.
Because now it's not about Ineed to go make a piece of
software.
Like I I've still run into somepeople who say that my big who

(36:57):
who describe themselves as theirbig ambition is to go make a
SaaS because that's what theybelieve software
entrepreneurship looks like.
But when I look at people whoare more sophisticated about
this stuff, those are people whoare looking to get jobs done,
who are thinking in terms ofservices and the way value gets
created.
And then they're asking, howcan we apply machines to be
doing this?
And when you start thinkingabout it from those terms, a

(37:19):
software-enabled business asopposed to a software business,
you start to free yourself tothink differently about like uh
how the software is going towork, but also what software
gets created, how much softwaregets created.
And then I think we get toreally unlock a lot of economic
opportunity.

Prakash Chandran (37:36):
So, you know, there's going to be tactical
founders, um, applicationdevelopment leaders listening to
this.
They might be building SaaS orthey have SaaS, right?
Like, let's let's pretend thatthere's a fictitious SaaS
company that finds the bestrestaurants in your area.
You're able to kind of makereservations.
They have a mobile and webapplication.
They are listening to this andthey're asking themselves, what

(38:00):
should I be doing right now?
I've got a web, a mobileapplication, I've got a back
end.
What should I be thinkingabout, right?

Ray Deck (38:08):
You should be thinking about your customers and your
data, uh, and be thinking morecreatively about how to be
getting leverage from both ofthose because those two things
are assets.
And the the software that'sbeen created so far, don't get
me wrong, it'll probably stillcontinue to create value for a
while.
But like, do we think the nextturn of the wheel looks like the
last one?

(38:29):
Is it you're evolving thesoftware that you have today to
be useful in like the nextcouple to be useful more than
say, I don't know, 12 monthsfrom now?
Um, or is it like what you'rereally doing is creating and
nurturing, right, this asset youhave of your brand in the
market, because you've beenproviding a good service to
people, of your customers whohave a certain amount of trust

(38:50):
and that are also providing youwith data, as well as like the
vendor relationships youprobably have, like the actual
restaurateurs or what have you,right?
And then those relationships,that knowledge you have, the
secrets you have, the the um thereputation that you have in the
market, the value of the brand,and being willing to think more
creatively about how you willthen uh uh you know filter those

(39:11):
through software to deliverthose assets to your customers
because your software was neverthe asset.
And that's I think a key thingabout most SaaS.
That's one of the reasons why Isort of am more skeptical about
SaaS overall, is when I lookunderneath at SaaS businesses,
it's like stone soup.
I don't know if you ever heardthat story, but like yeah, the
um the the idea is that what itwas a lure for the customer to

(39:35):
get their data in.
The data was the asset, therelationship with the customer
is the asset, the value they getout of it is the asset, but the
SaaS is usually just amechanism to go to connect those
things.
Got it.
And the uh and and and and uhwhen, and that's one of the
reasons why Micro SAS has beenso successful, because it's not
like technical excellence meansthat one SaaS rules them all.

(39:57):
It's going to be being able toconform that SaaS to the
particular needs of a smallerand smaller segment of people.
That means it does a better jobfor those people.
If you're able to do, and thethe promise of AI in a more
ephemeral form of software isthat it's able to conform around
the person in a way that is um,you know, different by an order

(40:17):
of magnitude from what camebefore, that implies that your
real value is going to be in thedata and relationship assets as
opposed to in the software codebeasts that you have today.

Prakash Chandran (40:29):
Love it, Ray.
Could talk to you all day.
Um, but in in uh moving to aclose, I wanted to ask you a
couple kind of more lightning uhround questions.
Um talk a little bit aboutwhat's in your like AI toolbox.
What are the tools that you'reusing that you're excited about
that are potentially underrated?

Ray Deck (40:46):
The tool I am most excited about AI-wise is Suno.
Um Suno is a software for beingable to make songs with AI.
Their models are extraordinary.
But more importantly, I cantake like a meeting that I just
did, maybe a conversation likethis one.
I can have Claude, I can have ahave a have a pipeline where

(41:07):
Claude is able to, you know,take that uh meeting and turn it
into uh identify the key ideasout of it.
Take those key ideas, turn theminto lyrics, have those lyrics
turn into an earworm in Suno,and I will then play those songs
in the background.
My my kids quote those songs.
Um, and and it's feeding theideas back into my brain.

(41:29):
The opportunity to be usingdifferent modalities, right, to
get the the this the thisuniverse of ideas that surround
us to feed them back in so thatwe can be making better use of
them is I think the theunder-index part.
It's not about developingsoftware, it's about developing
ourselves.
And that's why uh, and and solike I I know the the Suno one
is a little bit funny when Iknow I've sent you a couple Suno

(41:49):
songs before, um, but like thatthat kind of tech that tech has
done, you know, more to sort ofyou know infuse these things
into my brain than just aboutany of these other uh chat-based
systems.

Prakash Chandran (42:01):
I love it.
I was just gonna say I've beenon the receiving end of some of
those songs.
It's it's pretty incredible.
Um, did not think you weregonna say that, and I love that
you did.
Um talk a little bit about anunderrated like practice that
developers should be honing orworking on right now.

Ray Deck (42:23):
Um the White Room, uh, I think is the one that it
seems it seems like the hardestthing for me to get people to
do, which is to, you know,you're working on a hard
problem.
Don't try to edit your softwareto do it.
Clean room, or if you're ifyou're if you're in a chat,
clean chat, work just this oneproblem.
Um it is the it is somethingthat like professional software

(42:46):
engineers get taught in schooland then routinely forget.
I it I I find myself sayingthis to people at at whatever
level of education they're in,because especially when you're
getting most frustrated, it'smost tempting to say, that
didn't work, try one more time,right?
That that's the kind of thingwe've done with AIs before.
It's something we do withourselves all the time.
But the willingness to say,nah, you know what, this is the

(43:06):
one problem, let's isolate it,let's work this problem, and
then when we have betterunderstanding of it, let's bring
it back in is a hard thing todo from an ego point of view
because it's saying that youweren't almost there.
Right.
But then if you are willing todo it though, you will make a
lot more progress, youunderstand it much more quickly,
and you can tack a bigcomplicated problem as a series
of small and much simpler ones.

(43:28):
And that tends to uh, you know,crumble much, much faster uh
and provide a whole lot morevalue for you.
So I think as a mental modeland as a practice, the can I
take this to the White Roomwould be the question I actually
certainly find myself asking alot in like state change office
hours.
And that I would encourageanybody listening here at home
to ask, should they be doingthat today?

Prakash Chandran (43:47):
Yeah.
Um, that's a great one.
And then finally, you know,we've talked about a couple of
these in the process of ourconversation, but any bold
predictions um or just kind of aglimpse of what you think the
world is going to look like withAI in the next three to five
years.
And I know it's hard to thinkthat far out just because things

(44:08):
are moving so fast.
Um, but anything that you mightshare around uh what you think
things will look like then.

Ray Deck (44:15):
In 2012, anybody who knew anything would tell you
there's no future in AI.
Uh in 2017, anybody and andthen uh that was the year, uh
I'm pretty sure that was theyear the Alexnet came out and uh
transformed and just sort oflit the firecracker for the
world of of computer vision uhand uh and that little machine

(44:36):
learning.
In 2017, anybody who wasanybody would have told you that
like AI is good for computervision, but has no play in um in
in in natural language.
We were like working withthings like called RNNs.
It was like it was really akind of a go-nowhere segment.
It was in that year that umattention is all you need, which
was the paper that firstdefined the technology behind

(44:56):
what we now call LMs, uh, waspublished.
Uh and people have told you,like, you know, there, there's
you know, people in 2022 wouldhave said, uh, it is so obvious
that like the the you know, theuse of just sort of creating
like this this spacer foam andand chilly stuff uh for you know
any of this natural languagestuff that is being made out of
GPTs.
And that, of course, is theyear the Chat GPT came out and

(45:16):
caused us to really rethink justby uh changing up the packaging
and some improvements to themodel.
Um the world will keepchanging.
I don't know what's going to becoming down the road, but
something, but more things aregonna come, you know, down the
road.
And it's not gonna be aboutlittle stuff, it's gonna be
about big stuff.
I do think that there's goingto be an interesting play for
energy uh when I think about theeconomic implications, because

(45:38):
we now have much more demand forcompute, and compute requires
energy in a way that like SaaSthat were built in 2017 did not
require energy because they werejust about storage, right?
Rather than being about, youknow, compute.
And the uh the thing thatcaught that that drives the
biggest caloric intake in ourbodies uh is this brain that

(45:58):
we've got up here.
And machines are the same way.
When they're thinking, they runhot.
You put your hand on yourlaptop, when it's thinking hard,
you can tell that's runninghot.
And that's that's the reasonwhy.
And that the the uh a demandfor energy is going to be
changing, I think, significantlyfor that.
Uh, and and will probably havea significant impact on like the
the the way labor spreads uh aswell and the kinds of jobs
people do.
So those are the kinds ofthings I would look for in terms

(46:20):
of like the way I expect youknow markets to change and the
way that like the nextgeneration of people are even
more native, not even to theworld of software, the way they
were from the previous 20 years,but they're going to be native
to you know AI and what AndreKarpathy calls software 3.0, the
natural language-drivensoftware, as we head into the
next few years.

(46:40):
And that's leaving aside likebig weird possibilities like
quantum computing, et cetera,that could, you know, really
change up the way that like, youknow, compute is allowed to
work.
Um, but um, but anyway, thoughthose none of those were like
super huge you know predictions,but those are things that I'm
looking for in terms of how AIwill affect economics in the
course of the next, you know,36, 60 months.

Prakash Chandran (47:00):
Ray, as always, fascinating
conversation.
I really appreciate your timeand your wisdom.
Learned a lot, took copiousnotes as you were talking.
Um, thank you so much, myfriend.
Absolute pleasure, sir.