On with Kara Swisher - How To AI: A Practical Business Q&A With Three Experts
Episode Date: September 15, 2025As more companies push AI in their workplaces, the technology is rapidly reshaping the way many of us do our jobs. But a lot of people — from entry-level employees to the C-Suite — are still in th...e dark about the limits of AI, its best uses, and how to make it work for them. We called in a panel of AI experts to answer some our listeners’ burning questions about how to use it at work: Sayash Kapoor, co-author of the book AI Snake Oil: What Artificial Can Do, What it Can’t, and How to Tell the Difference and the Substack AI as Normal Technology; Rajeev Kapur, CEO of 1105 Media and author of the book AI Made Simple: A Beginner's Guide to Generative Intelligence; and futurist and author Amy Webb, founder and CEO of the consulting firm Future Today Strategy Group. Kara, Sayash, Rajeev and Amy break down everything from how vibe coding works to thornier questions around privacy and regulation. They talk about how young people can prepare themselves to enter the workforce, and how all of us can develop skills to stay relevant. And, of course, they weigh in on the question so many of us are asking right now: Is AI coming for my job? Questions? Comments? Email us at on@voxmedia.com or find us on YouTube, Instagram, TikTok, and Bluesky @onwithkaraswisher. Learn more about your ad choices. Visit podcastchoices.com/adchoices
Transcript
Discussion (0)
We're going to have to manage humans, AI agents, or whatever that format takes, and at some point, robotics.
So that's a whole different type of skill set.
I don't even want to manage humans.
Hi, everyone, from New York Magazine and the Vox Media Podcast Network.
This is On with Kara Swisher, and I'm Kara Swisher.
Today, we're answering your questions about how to use AI at work or for your business.
We're tackling everything from how vibe coding works to the big societal issues around regulation and privacy.
And, of course, we'll get into what so many of us want to know right now, will AI take my job?
A lot of you sent us some really great questions via email threads and blue sky, and we've called in a panel of experts to help answer them.
Syash Kapoor is the co-author of the book AI Snake Oil, What Artificial and Television,
can do, what it can't, and how to tell the difference.
He also writes the substack AI as normal technology.
Rajiv Kapoor is CEO of 1105 Media, a business-to-business tech marketing and events company.
He's the author of AI Made Simple, A Beginner's Guide to Generative Intelligence.
And Amy Webb is a futurist and the founder and CEO of the Future Today Strategy Group.
She teaches at New York University's Stern School of Business, and she's the author of multiple
books. And her latest is The Genesis Machine, our quest to rewrite life in the age of synthetic
biology. I think it's really important for us to do this a lot and ask all kinds of questions
as the technology develops. We did not have this opportunity to have podcasts when the internet
first started and I spent a lot of time answering these questions myself when I ran into people.
And so I think it's really important to keep asking questions as this stuff rolls out because
it's really in that phase where we're not really sure what's going to happen. All right,
Let's get into my conversation with Syash, Rajeev, and Amy, which is brought to you by SmartSheet.
Stick around.
Support for this show comes from SmartSheet.
There's a common saying out there, Jack of all trades, Master of None, but SmartSheet is changing that.
With its intelligent work management platform, SmartSheet can embed AI into your workflow
so you can tackle any project without needing to be an expert.
Whether you're setting up your first project, building a complex formula, or analyzing your data,
the AI-powered features of SmartSheet are there to work alongside you to help you move faster and get work done.
Go to smartsheet.com slash Vox to learn more.
It is all.
Syash, Rajiv, and Amy, thank you for coming on on.
My pleasure. Our pleasure. Thank you.
Thank you so much for having us.
Hey, everybody. Nice to be here.
All right. Let's start with a broad question.
and then we'll get into specifics.
So how has AI already changed the way businesses operate?
And in the next 12 months or so,
what are the most consequential impacts you expect to see from AI?
Let's start with Rajiv and then Saish and then Amy.
Yeah, I think for me, Kara,
I think what we're seeing here is that AI is starting to really start to take tasks.
And I think that's what's a key distinction between this idea of taking tasks versus taking jobs.
So I think you're going to start to see a lot more tasks.
being automated, for example, in my company, other companies I've been advising and working with,
the idea of taking every single thing that's manually written down, and how do you not take that
and AIify it, is really what's going to be consequential. And I think going forward, it's going to
really help really drive that opportunities of productivity, gains, efficiencies, and those kinds of things.
And I think that's probably going to be the big win. And I think you're seeing that a lot more
in the SMB space than you are in the big, larger Fortune 1,000 type of spaces, where the
SMB guys, actually, what I'm seeing, have a lot more flexibility and are willing to invest
and willing to take a longer-term view of how AI can impact their business.
Isn't taking tasks, taking jobs? Because those were jobs.
Well, I think what you're saying is that, look, we may not need to hire more people.
And I think what we're doing it, we're saying, okay, we're going to become more efficient
with what we have. And I think you're starting to see more and more organizations, start
to train their organizations as least that's what I'm seeing in the area where I'm focused on,
which is more of that consumer SMB space. We're starting to see that.
Okay, Saish.
So I really wanted to come back to your point about whether taking tasks is the same as taking jobs.
And I would say if you look back at the history of general purpose technologies,
these are technologies that can be applied broadly.
That has actually not been the case.
So, for example, in the 1970s, lots of people thought that ATMs would make bank tellers obsolete.
And what we've seen instead is that at least for the first four decades after ATMs were installed,
the number of bank tellers employed increased.
And that was because it was so much cheaper for banks to open new branches.
with the rise of ATMs,
that they started opening more of these.
But having opened a new branch,
they realized that, look,
we often do still need bank tellers.
Maybe their job is not cutting checks anymore.
Maybe their job is not to sort of hand over money,
but they do still need bank tellers
to maybe handle the customer relations side of things.
And so oftentimes that's what makes it hard.
Like sometimes new technology makes it so much cheaper
to sort of do the functions you were doing before,
that the demand for that sort of service or product increases, and that's what we've seen
with ATM. So as we have new technology automate certain tasks, the definition of the job
changes to be about everything that has not yet been automated. And what would be the most
consequential impact right now that you're seeing initially? So I think the main jobs that are
being affected already, are jobs which were actually already sort of reduced to carrying out
one specific task at a time. Most commonly, this is tasks where people had like contract work
or where the task itself has been neatly allocated away. So people just had to translate a piece of
text or transcribe a certain audio recording. And these jobs comprised of primarily solving one specific
well-defined tasks. Those are the sort of jobs where I think we're seeing the most impacts already.
I think there has been a reduction in the need for translation and transcription. We've also
seen routine work for artists get reduced because of AI because you no longer need to hire someone,
let's say, to create a logo that you want to use for your website. All right, Amy. I might have a little
bit of a contradictory viewpoint on this. Sure. The main thing that I'm seeing is a growing delta between
expectations and reality. We're working with a lot of CEOs and executive leadership teams who are
getting contradictory feedback. So a board of directors or the street, they want to see AI being
implemented at scale primarily to address the bottom line, which takes a lot of upfront investment.
Meanwhile, the street does not reward companies, at least not that we've seen, for implementing
AI solutions, and in some case, it can drive stock prices down. We've actually seen this with a few
companies already, and it's a shame because there are leaders taking strategic measured risks,
which is to say not throwing obscene amounts of capital at products and services before they
have a plan in place or a long-term, long-term perspective.
And they're making salient decisions, and they're getting spanked.
And why is that?
I think because, as we've seen with every other consequential technology for decades,
these technologies take a long time to develop and to implement them at scale.
Even at a small and medium business, you have things like compliance and insurance and
thinking through workflows and change management.
So the technology on its own isn't enough.
You need all of the other structures in place.
And you need a plan for implementation and then a plan for what comes after that.
And just that all takes time.
So technology may be developing at breakneck speed,
but the reality is that business moves at the pace of business.
Right.
And therefore, slower than people think or more measured.
And government's even slower.
Right.
Yeah.
And I think that was my point earlier, which is I'm seeing a lot more adoption at the SMB level.
Smaller.
Because they just don't have that same overhang that.
you might get from the street or other places, and they're willing to take risk, they're willing
to give it the time it needs to be successful. Right. Okay, let's type into the big question.
Everyone's asking, which is either, will AI take my job or the other side of the coin,
how can I use AI to reduce my labor costs? I do talk to a lot of CEOs, and especially in the
tech area, they're like, oh, I'd like to cut my coders from 6,000 to 2400. I heard this a year ago.
And I hear it a lot from different people, even people who are tech, but
tech adjacent to AI, essentially.
They're not AI companies, per se.
They're doing other things, like travel, et cetera.
So, Rishi, what industries do you think benefit the most from adapting AI quicker?
And what should they be doing in those fields to make sure they're deploying it intelligently
so they don't get over their skis?
As Amy was talking about lay off their workers, they actually need.
Yeah, in terms of a very specific industry, I mean, I'm seeing it being adopted across a lot of different key industry.
you know, just that I can't point to one particular one that says this one's going to benefit over over one particular one.
I mean, one could argue like the legal industry is going to go through a lot of change because of what's happening with AI and you may not need as many paralegals, et cetera, you know, things of that nature.
But I think in general, in terms of how you start to think about this, how you start to implement AI across the board, what I've been advising people is you start small.
You pick one process, whether that might be customer support, scheduling, whatever it might be, you know, marketing areas.
You deploy it, you start to measure the ROI.
You measure the ROI in weeks, not years.
And you really need to look at this from a top-down perspective.
And what we're seeing is a couple things.
Number one is that that ROI issue is a real issue for the CFOs.
And they really are the ones who are really tend to be a little bit of the roadblock
in terms of understanding and letting people breathe a little bit with this stuff.
But the other bigger issue is that at the end of the day, there's two kinds of AI.
As you know, there's the machine learning data side of AI.
Then there's the chat GPT stuff.
and on the generative AI things.
And really that machine learning data analytics,
that data side is the new oil
and you can't do anything unless you build refineries on top of it.
I think that's potentially a bigger opportunity
for some of these organizations.
But the problem there is that it's garbage in garbage out.
So really understanding their data is going to be really critical.
And if they can do that,
they'll understand how to start building the tools
to augment everything that's happening in the business
and then do the one thing money can do,
which is by time and just really build a culture,
on it and don't chase the shiny new tools that are coming out because the space is moving too fast.
All right. So, Syash, I want you to answer this question. We've got a ton of listener
questions on the topic of AI and job losses. So I'm going to read one about the disruption at
the managerial level. Aaron Hoffer in San Francisco wrote in to ask, why hasn't there been more
serious efforts at replacing leadership roles, even CEO with AI from a data perspective? There's so
many books and think pieces as well as case studies and decisions made by executives that could
and have been built into LLMs, if they were so great,
why wouldn't these AIs be replacing the mediocre leadership?
Is that realistic?
Could AI eventually fill some upper-level managerial roles
in potentially the C-level suite?
That's a very interesting question.
I mean, this reminds me of a case where someone tried to run with Chad GPD as the mayor.
The candidate's entire pitch was, you know, any policy-related question
would be put to Chad-GPD and, you know, I'll make sure I implement the outputs.
And I think, unfortunately, this sort of, this perspective sort of misunderstands what the role of leadership is often at these companies.
So, for example, the role of a CEO is not just to sort of take in all of the data and put out the optimal context, where it's also to build relationships to sort of figure out in sort of power struggles within the company, what the vision of the company looks like, holding people to that standard.
And I think all of those are things that, like, chatbots can't do.
On the other hand, we have seen a pressure on the managerial class on like mid and upper level managers as a result of AI2.
So for example, across the board, at least in tech industries, there has been a rapid sort of reduction in the number of people who are employed as mid-level managers because the expectations have risen.
And each manager instead of earlier managing like three or four or five employees is now managing on the order of 10 or more.
So to some extent, this has happened.
But at the very top, at the questions about, like, where the CEOs can be replaced with AI,
I think that's not really realistic for the longest time to come,
simply because the job of a CEO is not just to take optimal decisions.
It's to sort of balance various countervailing forces within the company
and decide what the vision for the company should be.
So presumably, you could use it to help instead of reading 90 leadership books,
it could posit things in, for example.
Absolutely.
I mean, I think it can be very helpful as a decision support tool,
just not as a tool to replace decision-making.
Do you remember Paul Graham,
like at some point somebody built a
what would Paul Graham do answer generator?
I'm sure I didn't use it, but go ahead.
No, no, but like it existed.
And Paul Graham was a co-founder of Wycombinator.
They took a corpus.
He had sort of kept a blog,
and somebody took the corpus of everything.
He had written and dumped it into a basic answer system
pre-machine learning that just sort of spit out answers.
And then General Catalyst at one point
had appointed their first female board member, who was, of course, an AI, which is all to say,
I think that these systems are useful as a way of cataloging and extracting institutional knowledge
about a place or a body of work from a person. And there's been pretty good examples since then
of this working out really well. At the end of the day, though, the most important technology is
people. And for at least the time being, people will keep working in organizations. And so this is
about relating to them versus replacing them with leaders with technology. That's a good point.
All right. Our next listener question, which Amy is for you, comes from Shelley Wilson in Providence,
Rhode Island. She sent us a voicemail in a moment. Let's hear it. I believe AI in business will impact
the younger generation more than any others. My question is, what advice would you give somebody early
in their careers, say maybe early 20s, mid-20s, about learning about AI in business and incorporating
it into their daily work routine. I ask this because I have two children in this age range,
both with good professional jobs. One is a chemical engineer. The other is a talent recruiter,
both at Fortune 500 companies, and they've learned nothing from their employers about how to
incorporate AI into their jobs. By comparison, I work for a large professional services firm in the
us. We've been trained on using AI, strongly encouraged to use it, and expected to use it every day.
So I worry about the younger generation, and I'm certain that my adult children would love to
hear what you have to say. Thank you. So, Amy, what advice would you have for Shelly, and how
should workers think about future-proofing their careers, whether they're executives or entry-level
employees? Sure. So it's a good question, Shelly, and one that I actually get a lot. And I just
want to highlight that oftentimes the fears that we have and that we have for our children are
actually just the fears that we have ourselves. And the same question that you asked was asked
at the dawn of social media, when people were confused about Twitter, and at the dawn of mobile
phones, and at the dawn of the internet, and I'm sure all of the technology that came before that.
So this is to say there's a lot of training that's happening and upskilling and re-skilling,
which are two words that I hate because it's sort of...
They're bad words. There are bad words. It sort of throws out all of the skills you've taken a
lifetime accumulating. So there's a lot of that happening right now for mid-career and above
professionals, but it's good to keep in mind that people are in their 20s. I've got a daughter
who's 15. These people aren't just digital natives. They are natives of a new world where there's
lots of different technology that all does different things. So the fears that you have about the
training they're receiving may be well-founded in a sense, but they're probably ambiently or on
their own making use of all of these tools. And in terms of future,
proofing a career. It's another word that I don't love future proof because it assumes that you
have total control over all of the uncertainties and like the math doesn't work out in all the
variables. So it's more about being flexible as you go. So Rajiv, you're also doing a lot of work in
the AI education days. Do you have anything you want to add for Schelling? Yeah, sure. It's interesting.
You know, it's actually really opportune. I just drove my son from L.A. to Dallas last week.
He started a new job on Monday working for an up-and-coming AI company. And the exciting thing for me
is exactly kind of what Amy said, right?
This is the most advanced technological generation we've ever seen,
and every generation is like that.
I have zero concern about where they're headed.
But I will say a little bit of a different viewpoint here.
I think critical thinking skills and really focusing on those
are going to be really, really important.
You know, I would say the vast majority of folks
don't really need to be concerned with how the sausage is made.
So learning how to clearly prompt,
how to really work with AI,
how to communicate with AI is going to be a critical skill,
fact checking, understanding how to deal with hallucinations,
and those kinds of things, making sure the data they're receiving is clear and accurate
and how to then deploy that across an organization.
But curiosity and judgment and critical thinking skills are where I would really encourage parents
today to really make sure they're spending time with their young adults and their children,
making sure that that is not a skill that they're just neglecting.
It's going to be more and more important going forward in the future.
So let's talk about how to actually use it for business, Syish.
We're going to ask you this one.
This is a question that's sent to us to be a blue sky.
I think it's De Jongo wrote,
What do AI developers think we do?
They spend trillions on an AI that will, quote,
create images, music, and video.
But when we ask it to report how many times an account number
shows up in a PDF of an invoice,
it cannot provide an accurate answer.
So, Zayash, that was admittedly a question wrapped in Iran,
and actually amount of money invested in A is probably not trillions of dollars yet.
But nonetheless, it's hitting on a bigger topic
that you wrote about in your book,
that is, what task can AI do well
and why does it sometimes hallucinate
and make up answers when asked a simple request?
Yeah, I mean, that's a great question.
And I think this goes back to how these language models
are trained and what they're trained to do.
So basically, the way chat GPD works
is given an input sentence,
maybe a query by a user,
all it is doing at any given point
is sort of predicting what the next most likely word is in its response.
So if you ask it, what is your name?
maybe the next most likely response is
my name is chat GPD and that's what it comes up with
and for some of these responses
the chat part can be really precise
some of these responses are data
that it has been trained on hundreds or thousands of times
for example if you ask it things
or information that belongs in Wikipedia
this is something language models have been trained
hundreds of times on and it is quite likely
that the chat part will get it right
that's what gives it the illusion of actually
sort of being all-knowing in some sense
or giving these answers correctly.
But unlike humans, if you, for example,
ask the same question to a human,
you would expect a human who can tell you anything on Wikipedia
to also be able to count up to 100.
But that's not the case for Chad GPD.
This has also been called the jagged frontier.
So the frontier of what language models can do is quite uneven.
They are very good at certain skills, but very bad at others.
And so we can't really extrapolate
based on how well a language model can answer questions
about Wikipedia, we can't really extrapolate that to how well it can do maths.
And to be clear, I think this is not just so that we shouldn't use AI at all for a number of
tasks, but just that we should have these really clear evaluations before we start to use them,
especially for critical tasks.
We need to understand how well it can do at that specific task.
Right.
So, Rajiv, what about vibe coding?
That's when people use an LLM like chat TPT to write a code for, say, an app that can do
XYZ.
So you can use plain language-prompted chat bat to write code rather than actually
program yourself. From a business perspective, what are the pros and cons in the rise of vibe
coding? And for those of us who aren't software engineers, is vibe coding a viable way for non-technical
people to use this? Yeah, that's a great question. Ultimately, I think what vibe coding is going to
enable is a whole level of entrepreneurship that we haven't seen in quite some time. I kind of like
to call it this dawn of a new enlightenment period. It's going to enable a whole new ways of
looking at art, music, sciences, all these kinds of things. It's what's going to lead to more of this
broader revolution from an industrial perspective that we're going to see, you know, across
country. So I think, you know, those are the pros. I think the cons are, you need to have a
very specific skill set. I think it goes back to the things that we talked about a few minutes ago
that Seish and Amy also touched on is that this idea of critical thinking, the idea of prompting,
the idea of really making sure that you that you have that skill set and you're developing
that skill set to be able to really ask the questions the right way and really to be able to put
into that prompt box because that's going to become probably one of the most critical
skill sets that anybody can develop. And if you're struggling with that, you could get frustrated,
bad data could come out of it, so a little bit of garbage and garbage out, frustration,
longer periods, lack of sticking with a particular project, some data chaos. But I think
one of the bigger challenges might be in organizations is you see this concept of shadow AI popping
up, right, where different groups, different departments are all doing kind of their own thing
and with the lack of any sort of control or process. Making, vibe coding this stuff.
Right. Yeah.
So speaking of critical thinking, our next listener has related queer.
It was sent to us via threads from at ScanMy Photos.
They wrote, as AI takes over the, quote, thinking tasks at work, how do we make sure humans don't lose the ability to think for themselves?
Amy, why did you tackle this one?
There was a recent example of this, a study published in the journal Lancet, gastroenterology, and hepatology, found that in just a few months of using AI, you interpret test results, doctors are being about 20% worse at doing it on their own.
And many people can't look at a map.
I mean, you could use examples from the past.
So, a listener asks, how do we make sure humans don't lose the ability to think for themselves?
Well, that's up to us, isn't it?
Right.
There's no question that with certain technologies that offer automation, the resulting impact,
if they're really good, is a sort of learned helplessness.
That's a very good way I'm putting it.
The more that you use a chat GPT, for example, to write your emails or to help writing your essays,
There's a little bit of a slippery slope between structuring that organization and doing the critical thinking yourself to start and then just like, please answer this on the style of a McKinsey senior exec or whatever it is that you're trying to do.
And that doesn't cause catastrophic long-term negative impacts necessarily, but it does put you on a path where you're not going to continue to learn and grow.
And learning and growing is important even in the absence of, you know, game-changing technology.
So, look, when I was in advanced calculus, I remember at one point there was a connected calculator.
And I had a professor who was very adamant that we did not use this advanced calculator technology.
You know, and I think everybody's been through some version of that.
A calculator is just a brick if you don't know basic math.
And in a way, AI is also kind of a brick if you don't know what to do with it.
but it is becoming easier to know less
and be able to sort of do more.
So this is really up to everybody.
And it kind of relates to the vibe coding conversation earlier.
You know, one of the key pieces of missing information
is not what do you put into the prompt window,
but as a company, how comfortable are we dumping
in a corpus of our own data
so that people can build stuff with it?
I'm much, much more concerned about the lack of depth
about what happens when you sell or license
your data or you allow employees to dump your data into a corpus or to make their own corpus
and then throw like a llama on top of it, an open source tool, with the potential consequences
of that in terms of lost revenue in the future or lost ability to make revenue or now
all of your stuff is out there for anybody to use. Those are the kind of fundamental thinking
questions that I think we should be spending more time with. Yeah, in business. You know,
Scott Galloway, who's my partner, Pivot, puts everything in and I put almost nothing. Yeah.
same. I actually was saying this to Sam
Alton, he goes, why not? And I go, I just don't trust you.
Yeah, it's not, that's not, it's not you. I said, no offense, but
no, no, I mean, look, there's not a lot of people stopping to have
thoughts about that, let alone a critical debate about it, and it's super
important. Well, he's not a lawyer and he's not a doctor, so there's no
expectation of privacy, right? No, we could have a whole separate
conversation about other ways that that company is breaching what I would
consider to be basic privacy. But again, like, sometimes when tools are
so easy, it's easier for us to use them than to stop and ask a question about what the potential
implications are of using them. And then you give up, like, I'll just tell them. Well, yeah. And there's a
lot of already instances, especially in the world of news and media, where archives have been sold
and nobody really thought about what that would mean going forward. And that's revenue,
it's going to be a revenue problem for companies in the future. So now is the time to be thinking
and making some decisions using information about the future.
We'll be back in a minute.
Support for On with Kara Swisher comes from SmartSheet.
Most people listening know what it means to be burned out.
When you feel overwhelmed by a new project, a complex spreadsheet, or a data-heavy report.
Sometimes the work keeps stacking up and doesn't stop, halting your creativity and blocking you up.
Smartsheet wants that work to flow.
Their AI-powered platform is designed for you to spend less time struggling, freeing you up for your most creative, impactful work.
How?
It all starts with a simple conversation.
Their AI-powered onboarding helps you set up a project in minutes, even if you're not an expert.
Need a formula?
Just describe what you want, and SmartSheet AI generates it for you.
Want to understand your data better?
SmartSheet AI analyzes it and creates dynamic charts.
you can add to dashboards to track your projects in real time.
It's personalized, it's intuitive, and it makes work more efficient for everyone.
So you can break through the distractions, find your focus, and start working with flow.
Go to smartsheet.com slash box to learn more.
Okay, our next listener question comes from Carrie Tolerico in Scranton, Pennsylvania.
She sent us this voice memo. Let's hear it.
Hi, Kara. I'm curious if your panel would agree with my opinion.
that the broad adoption of AI is going to be limited by the lack of tech savvy among average
employees at average companies. My company has been testing some of these tools, and I've been
struck by the amount of handholding our middle managers have needed during demos or training.
These aren't stereotypical tech illiterate boomers. They're mostly Gen Xers and older millennials like me.
I think there's a real skill gap between the people working at your average small or mid-sized company
and those working directly in tech or industries that are tech-adjacent or particularly tech-heavy.
and this gap will have a big effect on the broad adoption of these tools.
We'd love to hear everyone's thoughts. Thanks.
Syash, what do you think?
I absolutely agree.
I mean, I think we've seen this process, once again, as Amy said, with past technologies as well,
where technologies go through phases of invention and innovation on top of it,
but then the broad adoption takes much, much longer.
And so this is one reason why I think the true impacts of AI will actually not even be realized this decade.
It'll take us much more than the next five years to,
actually realize the impacts of AI to figure out what productive uses we can put it towards.
And in some cases, it might not even be the case that, you know, existing business structures
can make use of AI very well. I mean, I'm reminded of how after we invented electricity,
it took 40 years for us to figure out how to electrify factories. And, you know, just putting
in electricity in the existing factories was not enough. We needed to basically reorient the
entire factory layout around the process of electrification to make it work. And so in some cases,
the most dramatic transformations and the sort of most in-depth adoption of AI will only take place
as people learn how to use it as we figure out what new use cases there are, and in some cases
even reinvent how we run businesses around AI.
So to her a question, where do companies get it wrong when it comes to them rolling out
AI for their business?
One of the main things I've seen is not recognizing the gap between capability and reliability.
So chatbots are very good at taking you 80% of the way there.
If you ask for a first draft of something, you can get something that's about 80% good.
If you ask it to do something on the internet, maybe you'll get like an 80% version of it back.
But actually getting to like what is often called the five nines of reliability, that something is 99.999% reliable is much, much harder, especially with random systems or sarcastic systems like chatbots.
And so I think that's what many companies have failed to realize.
And we've seen catastrophic product failures as a result.
So I'm reminded of these two products, the Humane Tech PIN and the Rabbit R1, these were both general purpose AI assistants.
Both of these were pretty capable.
They could like order DoDash to your home address, but maybe 10% of the times they would order your food to the wrong address.
And, you know, that might be all right when you're using this product in a lab.
But from the perspective of a real world actual product that consumers can use, it's a catastrophic failure.
So both Amy and Rajiv, I want you give an example of a scenario that might unfold at a company that's encouraging its employees to use AI.
say you have an employee who's hired to be a subject matter expert on a given topic or
project, and they're put in a situation where someone higher up in them is used AI to solve
a problem.
They're working on the solution generated is wrong, and the employee knows it's wrong.
The hire-up doesn't know that, but keeps pointing to the AI's evidence that they're right.
For example, that's something that could easily happen.
How can companies get ahead of situations like that and what policies need to be in place?
Amy and then, Rashid.
Yeah, I mean, look, the real world scenario that is happening in just about every company
that I've been inside of for the past two years is there is very likely an enterprise level
AI, something or another going on, and almost exclusively that has to do with paper forms.
So financial services, healthcare insurance, it is automating the process of manually entering
those data or synthesizing data into one standard set, but that used to be a very capital,
like human capital intensive job. That's already happening and unifying those data sets.
So there is that kind of AI going on.
Most of what's happening is individuals are sort of going rogue and doing their own things on AI,
whether that's writing their reports or, I don't know, screwing around and trying to come up with brilliant, big ideas to bring into meetings,
outside the purview of the executive leadership or any management team.
And I also have anecdotal evidence because my father-in-law, who's like 82 years old and fat fingers every text message,
he's ever sent is now primarily using, I think it's chat GPT on his mobile phone and
replace of instead of search, right? Because it's just easier and gets them to where he needs to go
faster. Okay. That is the present day scenario. The problem with that is there are not yet
legal provisions in place in most companies. The data protection, I mean, look, in the United States
alone, there's a patchwork regulatory situation between states. So in some states, with some
insurance, for example, I think in the state of, it might be Connecticut, I'm going to get this
wrong possibly, but I don't think you can use AI in most circumstances there, totally different
in Texas. So that part is very real and very challenging. I think what we need to do is stop
going crazy and wild and thinking about all of these doomsday or totally utopian scenarios
for AI and just get very much into the weeds on what's actually happening right now and why and how to
put it on the right path in every company.
Actually, Rashid, I'm going to restate that.
You said earlier that data is oil,
and you've also said that business leaders need to understand
the difference between genera of AI and machine learning
in order for them to be successful,
need to focus on their machine learning and invest in their data.
Flesh that out for us,
because what both Amy and Saezha had been saying
is that a lot of people go rogue,
they're doing other things.
How do you become disciplined in what you're doing
and what's important for you versus what Amy was just talking
about, like, hey, jump on chat GPT and get a very unsafe answer.
Yeah, I mean, first of all, Amy's answer was dead on right.
And in terms of the issue of machine learning and data, look, here's the bottom line, right?
Which is companies create data every single day, every single minute.
But if you start looking at organizations, I've spoken to probably about 2,500 CEOs in that
SMB space in the last two years, Kara.
And I can count on two hands and my two feet.
How many of them feel like they have good data?
data. They just don't have good data. And the reason why is because when they think about data, they think spend, they don't understand. It was never a priority for them to understand and realize how data becomes so important to them. And so that's when, you know, you start to see some of the projects start to fail and you start to see some of the report from MIT about projects failing and this and that. And that's part of the reason why is because they just don't have good command of their data. They don't understand. As much as we live in this bubble of tech, there's a whole world out there that doesn't understand a data warehouse. That doesn't understand.
data management, doesn't understand data analytics, doesn't understand how to pull data and use
it to have it tell a story, whatever the case might be. So I argue with folks that don't chase
all the shiny new objects that are coming out every single day that all these companies are
putting out, which is great and it's awesome. But it's moving so fast. Just to focus on your core
execution of let's just get a data scientist on board. Just start with that. Just get somebody to
look at your data, start understanding and finding basic implementations of what you have
access to and just start there and use that as a win to start building off of. Use that as your
foundation. If I may. Because you can't do anything. Go ahead. Go ahead, Amy. Just get a data scientist
is like telling a fancy lady in Los Angeles to just go get a burkin. Right? Which is to say, like, very,
very scarce. So I would love for every small and medium business or like any business to be able to go
to the data scientist store and hire people. The reality is they aren't there. And this is
another big problem. We didn't really think through the future of the workforce.
we would need. So we don't have enough people in the pipeline. And I just, I just want to set
expectations for everybody because you do need to have some understanding of how these systems and
tools work. And in some cases, you do need a data scientist, but there's not a lot out there
to choose from right now. Right. So let me point to that. You touched on this right now, but we're
seeing more come to encourage your employees to use LLMs, which is even worse, right, just on their
own. But also more employees using without telling their employers. A few months ago, the security and
software company, Avanti, put out a report showing that a little more than 40% of office workers
are using generative AI tools like JetGBT, and the one in three said they're doing it in secret.
So if AI adaption is inevitable, I'd like each of you, what is the smartest way for leadership
to shape it rather than chase it, Amy, and then Saish and Raji?
I think the, this is going to sound a little ambiguous, but I think the bottom line is
to ask questions. There are too many leaders out there.
throwing cash, obscene amounts of money at every big professional services firm, and they're all
knocking down the door, looking for handouts to help them build huge, enormous AI systems that
are either going to underperform or going to need to change. So there is some amount of, like,
you have to be a little bit more level-headed and understand, first of all, what problem are you
trying to solve? That's the most important thing. What are you trying to do and why? And if you can't
answer that question, AI is not going to answer it for you.
It's just going to be a very, very complicated, very expensive, time-intensive solution.
So that's the first thing.
And I do worry about over-reliance on consultants.
I'll give you a quick case study, quick little example.
One of the smartest CEOs I've ever met had a very smart idea for how to advance, not just his company, but the industry.
And one of the big consulting houses came in and said, we got it.
you know, we'll build it for you.
And they overbuilt something to a degree that I just never seen before.
And it drastically underperformed.
And now they're nowhere closer to where they needed to be,
but they're out a significant amount of money.
And this was a publicly traded company.
So they took a hit on the balance sheet.
You first of all have to figure out what it is that you're trying to do
and then figure out who's in your value network that's going to help you get there.
And it may or may not be a consultant.
It might be partnering with some smaller group of people.
I don't know.
But this is not a one-stop shop thing anymore.
Right, Sej?
One of the things I'm quite optimistic about is actually running experiments, small pilots in-house.
And I've seen, like, many businesses have had successes where, you know, employees start out with the site project, like trying to use AI to sort of speed up or improve the productivity of a certain business process.
And that's how you figure out, like, you separate the wheat from the shaft, basically.
and, you know, to some extent, to the extent that it is possible, I think it's also good
for companies to lean into this type of thing. For example, I remember all of the big tech
companies giving people like a few hours of their week off to pursue side projects. And some
of them went on to become these really important, crucial business ideas that have now grown
to like hundreds of millions of users. And so this is the type of thing where, you know,
the experiments that are already happening organically within a company can actually be used to
sort of showed up the company's responsiveness to AI.
Rajiv?
Yeah, no, I mean, I agree.
Look, we, at my company, one of my divisions, what we do is we do a lot of data analytics,
education, and training programs.
And the thing that I'm encouraged by, and maybe a little bit of a different take on what
Amy said earlier, which is CEOs and executives are now coming with their IT people to these
events because they realize it's a partnership.
They're starting to understand and realize.
where they have gaps. So I'm encouraged because this hasn't happened before. I mean, we've been
doing this now for years and decades, and now for the first time, you're actually seeing the C-suite,
some of the members of the C-suite, actually coming with their CIO, their CTO. So that's really
exciting to see. So I'm encouraged for that for the future. Okay. So I've got a kind of two-and-one
question now about transparency around the use of AI. Both of these come from Blue Sky. Tav asks,
how are business leaders addressing or not the imperative of labeling AI-generated content in internal
workflows and client-facing outputs? And Shane Spicer asks, if you work in a field that involves
relationship and relation building, education, mental health, etc., how transparent should one be
with people that they're reading or seeing is generated by AI? I think this is sort of getting the idea
of standards. Why don't we start with you, Amy, and then Sayish on this one? I think this is,
is industry by industry. And what I've seen so far is that in the creative industry,
certainly news and journalism, I think there's a much bigger emphasis on being transparent.
And I think in the health and medicine industry, because again, very heavily regulated,
I don't know that I've seen the same in many other industries, whether that's like big engineering
and construction or retail. And I don't know that there are any current regulations around that in
most countries that's different in the EU, and it's different in, like, Japan. So really,
it really depends. But if the question you're trying to get at is, should there be transparency,
the answer is yes, but good luck figuring out mechanisms to enforce that, because right now
the financial incentive is not to, just use it. That's absolutely right. Use it as a tool and
hide it away. What do you, say, Ash, what do you think? I mean, I broadly agree. I would say,
at the end of the day, the solution that is most likely pragmatically to work is to just
enforce accountability on the final outcomes. So irrespective of what tools an employee or someone
in these industries uses to get to the sort of final point, they're the ones who have to stand
behind it. And I think we're seeing some good examples, I mean, like depending on how you look
at it, of lawyers actually being held to account for using chat GPT generated citation. So I think
over the last couple of years, we've seen over 100 cases where lawyers have introduced AI-generated
hallucinations into their legal briefings.
They've actually even presented it.
We'll be back in a minute.
Support for On with Kara Swisher comes from SmartSheet.
Work is taxing.
It's the reason it's called work.
But if you're struggling with endless new projects,
complex spreadsheets, and data-heavy reporting,
then work isn't even the right word.
It's become.
a problem, and it's stopping you in your tracks. That's why there's SmartSheet. Their AI-powered
platform is designed to get your work flowing so you can spend less time struggling and more time
doing the things you're actually excited about. It starts with a simple conversation. AI-powered
onboarding helps you set up a project in minutes even if you're not an expert. Need a formula? Just
describe what you want and SmartSheet AI generates it for you. Want to understand your data better?
SmartSheet AI analyzes it and creates dynamic charts and dashboards.
in real time. It's personalized, it's intuitive, and it makes work more efficient for everyone.
So you can break through distractions, find your focus, and start working with flow.
Go to smartsheet.com slash box to learn more.
So let's wrap up talking about standards ethics and the rules around it.
So you co-write a substack that you recently retitled AI as normal technology
where you try to find the middle ground between dystopian and utopian versions of what AI can do.
Take out your claim here. Does calling AI normal technology mean it's not really going to change how companies operate? And how does AI compare to, say, the Internet, which clearly has been a truly revolutionary technology? I mean, so the claim that AI is normal technology is in opposition to claims that we are inventing a new species or we might potentially read superintelligence by 2027 or 2030 or what have you. I think in the absence of this kind of common sense default way of thinking about the future impacts of the technology,
I think lots of people were starting to think about whether they should prepare for a world where the company has no employees in the next five years or whether superintelligence is actually going to take over.
And so a lot of our emphasis with the AI as normal technology project is to kind of push back against that to outline that AI actually might well be very similar to previous general purpose technologies like the Internet or perhaps even like electricity.
Now, these technologies were clearly transformative and the normal in the title is not to say that AI is not.
not transformative, but just that the impacts of AI will play out not over the next two years,
but perhaps over the next decade or two.
But now, Rajiv, you're more bullish on AI shifting things rather quickly.
Make the case for why it fundamentally transformed the economy and society, presumably,
and how government leaders should deal with the unintended consequences of a revolutionary technology.
And, again, should there be limits on how businesses can use AI?
Yeah, look, I mean, I guess a couple way to answer these questions.
Why am I bullish?
I'm bullish because I'm starting to see it really make a difference in a lot of people's lives.
I'm seeing it happen with individuals.
I'm seeing it provide major change in transformation and helping organizations and companies really think about their company going forward how to compete.
I'm bullish because I think we're going to enter into a new world of entrepreneurial expansion around the world.
And when you democratize these tools and people have access to these tools from around the world, I think you're going to see a lot more entrepreneurial ideas.
coming on board. Now, in terms of, should people just be able to run Hog Wild? No, I mean,
probably not. But look, at the end of the day, there's some regulation that probably should
happen here. And I realize that our word is really difficult for some people to embrace and understand.
And I don't think the big companies are going to want it. The government's here not going to
want unless everybody does it. But with that being said, I think there are some common grounds
to really address areas such as deepfakes and those kinds of things. So I think there's some common
ground there that we could do. But in terms of
why I'm bullish is because I'm seeing it.
Kara, I'm seeing it every day. And I think
this is a different type of revolution. So
look, I remember the e-commerce time, Carrie, you know
I used to work for Michael Dell. And I remember
walking into his office. I think I told you the story
when he said, hey guys, go figure out how
to sell computers on the internet. We all looked there and thought
he was crazy. Like, who's going to buy anything on the internet?
Right. Look where we are right now. And so
this idea of being curious, not judgmental,
I think it's going to be really important. And I really think that
this is a revolution that's happening more
at that individual, small, medium, business level,
I think that's going to really rise everybody up here,
and that's why I'm bullish about it.
Amy, now you've argued that privacy is dead,
and now with LLMs, it seems like intellectual property
is also on shaky ground.
What other rights will AI erode then,
and is the trade-off worth it,
both for businesses and citizens?
Yeah, so privacy is dead.
I think intellectual property,
if not already, having breathed its last breath,
is certainly on its way.
The problem is that we have to stop thinking,
thinking personally and in business of data as just things that are written or numbers that
are in spreadsheets. Data exist in many different ways. The way that you type on your keyboard,
the way that you walk, your unique gate, your heartbeat. And we are very quickly entering an era
of robotics, which I know, I know for 100 years everybody's been promising robots. I think
they're actually coming now, but not in the form that we all expected. And the reason for that is to
achieve the next iteration of AI, we need embodiment, which is to say AI systems need to pick up
more data than exists that's scrapable in physical form. So with that being said, your personal
data is not just a way that you look, but literally the gestures, the way that you move
your body around and the combinations of all of these different points, which is true for
employees, which is true of teams, which is true of businesses, and I could go on and on.
The challenge is regulation is inherently a reaction to something that's already happened.
It's not a pathway to the future, yeah.
To the future.
And the problem that that will eventually create is that as we start to see true advancements
and I have a whole separate conversation about AGI and where we're actually at and what all of that means.
But in the event that we're headed in that direction, because we're going to have more types of data, more contextual data,
What that implies is lots and lots of lawsuits ahead
because we're not currently preparing for a world in which we define data much more broadly.
We're scraping data in different ways,
and we're starting to use those data in ways to support different types of businesses
where you may not become a beneficiary on the other side personally.
So what are some of the AI trends you see on the horizon,
ones that we didn't get to cover,
but that our listeners should be thinking about.
Sayash, let's start with you, and then Rajiv, and then Amy,
So I think I continue to think that one of the biggest changes will happen organizationally.
Like I think the organizational structures that we've seen so far might need to be radically changed in response to how we can use AI as a technology.
And to give you an example, I think let's turn to software engineering.
So for the longest time, we've had companies that build software and companies that use software.
We have had like a few hundred companies that build the majority of the software that the rest of the world uses.
but with AI, as the cost of developing software keeps going down,
I think we might need this fundamental change in the sense that
rather than sort of us all relying on tech from a few companies,
it might turn out that it's more efficient for each company
to have this small software engineering team that can actually handle
all of the sort of business needs that this company has.
And so I think this is the sort of seismic shift that usually takes a decade or two to unfold.
Like to Rajeev's point, the reason it took the internet so long
to get to the point where it is today was because we needed shifts in entire business models.
We needed e-commerce companies like Amazon or perhaps like even others to sort of come about to
really make this move possible.
And so I think for people trying to look ahead, maybe the questions are at the level of
organizations and what does the future of the organization look like, whether it is because
of AI for developing software or for any other application, for marketing, for any other business
process.
Okay.
Rajee?
Yeah, that was a great point,
Sayish, and to build on kind of Amy's point about the robotics
and Sayish's point there,
like Carol, we are the last generation,
like right here, the four of us,
that's ever going to manage humans alone.
We're going to have to manage humans,
AI agents, or whatever that format takes,
and at some point, robotics.
So that's a whole different type of skill set.
So if you look back earlier in the pod where we were talking about...
I don't even want to manage humans, but go ahead.
Yeah.
So, but if you go back earlier in the podcast,
You're talking about leadership and CEOs being replaced.
You just can't because you can't replace that empathy.
And so, look, the bottom line here is the spark's been lit, the genies out of the bottle.
And the biggest challenge, quite frankly, it's moving so fast.
Like people I speak to, again, the CEOs I'm speaking to, is that they're getting whiplash
because it's just moving too fast.
Like, they just don't know which tool to use, which one to go to.
It's going to be interesting what the future unfold.
I'm excited and just, and I'm glad that there's folks like Amy and Saj and yourself
who can really help bring a lot light on this.
All right, Amy, last word.
Sure. So in my world, trends are not trendy.
They're long-term indications have changed.
I think it's more important to think about convergences when it comes to AI.
So I've got three.
The first is the convergence of AI and biology, but not for medicine, for other things like construction.
So there are metamaterials that are being generated, created.
So think of bricks that can move with seismic activity or things that are made out of wood,
a different type of wood that normally wouldn't be,
or packaging, also engineering soybeans and cane sugar,
all different types of things to grow in environments
where climate has presented a huge challenge.
So that's already something that we're seeing
that will have long-lasting impact on the economy and society.
AI and robotics, going back to that for a moment,
forget anthropomorphized humans.
This is stuff like, imagine instead of a person
having to put up scaffolding on a building,
which in New York is persistent.
and everywhere. A robot doing that instead, faster, more efficient, and definitely safer,
among other things. In China, they're already using to park cars. That's right. That's right.
And then the third area, because I guess we're at the end now, is AI in space. I think we're going to
see a faster advancement space exploration, and this is not actually an area that I research,
but my daughter wants to study aerospace engineering and become a lunar architect. So part of what she
who's learning how to do with AI is a totally different type of design and simulation so that she
can build habitats on the moon. And going back to that earlier question, she's 15. She's in ninth
grade. And this is the kind of stuff that she's already starting to think about how to advance
human civilization, not just to be off planet because Elon Musk thinks it's cool, but for the purpose
of learning new things about ourselves. Yeah, you know, I was just interested in astrobiologist because
I'm talking about this. And he said the worst day on Earth is better than the best day on Mars.
100%, 100%, but you can still have some cool days on the moon, you know?
The moon, yeah, yeah, the moon, I guess.
It's a little dusty for me.
It's like Burning Man.
Anyway, I really appreciate all this,
and I think it's really helpful for people to keep this dialogue up as things change over time.
One of the key messages I think all of you were saying is do not run away from this
because it's like running away from electricity or the internet or something.
It's inevitable, and if you're not part of it, you will be definitely left behind.
Anyway, we appreciate it.
Thank you so much.
Thank you.
Thank you so much for having us.
Thanks.
Today's show was produced by Christian Castro, Roussel, Katerioca, Michelle Alloy, Megan Bernie, and Kaelin Lynch.
Special thanks to Bradley Sylvester.
Our engineers are Fernando Arruda and Rick Kwan, and our theme music is by Trackademics.
If you're already following the show, your normal technology.
If not, robots are coming for you.
Go wherever you listen to a podcast, search for On with Kara Swisher and hit follow.
Thanks for listening to On With Caroushisher from Podium Media in New York Magazine,
the Vox Media Podcast Network, and us.
We'll be back on Thursday with more.
Thank you to SmartSheet for supporting this episode.
Technology is moving faster than ever.
To stay current with works changing landscape, you want to partner with a platform that adapts with it.
AI that can empower you to move faster.
think bigger and work with flow while the AI capabilities of smart sheet can help you cut
through complexity and focus on what matters most find out more at smartsheet.com
slash box that's smartsheet.com slash box.