# AI Transforms Healthcare: Strategy, Efficiency, and Human Connection

**Podcast:** a16z Podcast
**Published:** 2026-06-10

## Transcript

The way that we make sure this technology benefits all of society is we make important things cheap.
And the most important thing is healthcare.
This is one of the most beautiful humanistic technologies we've ever invented.
We've spent 40 years building technology that really extended our intellect.
And now we have this technology that can actually connect with humans on an emotional level.
And that's a very beautiful, powerful thing that we should not be skeptical of, we should lean into.
The shape of this new technology is so fascinating because it actually can do work on our behalf.
1% of the world has seen God and 99% is walking around oblivious.
And how do we actually connect the dots?
Healthcare is one of the most complex industries in the economy.
It's also one of the most administrative.
Today, roughly 45% of healthcare spending is tied to administration rather than direct patient care.
AI has the potential to change that, not just by improving productivity.
but by fundamentally reshaping how healthcare organizations operate and how patients experience the system.
In this conversation, ScanHealthPlan CEO Sachin Jain sits down with A16Z general partner Anish Acharya to discuss AI adoption, organizational transformation, and why the biggest opportunity may be making healthcare dramatically more efficient, accessible, and human-centered.
Thank you guys for having me.
Thank you so much for joining us.
It's an incredible honor to have you here.
And you've had this meteoric career and you're now at the center of so much that's happening.
I was hoping you might just take a few minutes to actually set the broader stage.
Tell us about you.
Tell us about Andreessen, what your view is of the world, whatever framing perspective you're able to offer.
Yeah.
Okay.
So I'm just going to yap for a few minutes.
So, and like, I want to have a real dialogue here.
So please, you know, during Q&A, let's have real substantial questions.
I'm happy to tell you what I know, what I think, what I'm not sure about, how Andreessen's using AI.
Let's just cover everything.
So that's the sort of goal to make sure you guys get as much value as possible.
Let me frame up what Andreessen is for a moment.
So we have $50 billion of assets under management.
We're early stage investors.
That means we're like two guys in a garage, first check, second check.
We're in companies like Coinbase, Airbnb, OpenAI, Mistral, Databricks.
So the goal of the firm, we're founder-led by Marc Andreessen and Ben Horowitz, is to be a part of every important technology story that happens.
And what's happening right now with artificial intelligence is the most important technology story, certainly in my lifetime, and maybe since, I don't know, like the wheel.
Like that's how big of a deal it is.
It's a very, very, very big deal.
I think the two things that are missed, and I got a little bit of background at where all of you guys are right now in your journey.
So one, I heard that you've had a tremendous year of growth up to 450,000 members.
That's incredible.
You know, they always say in my business, the reward for solving problems is harder problems.
So hopefully we can help you solve some of those harder problems.
And I know that a big focus for SCAN is member experience.
I think one of the things that gets missed in the AI story is that this is one of the most beautiful humanistic technologies we've ever invented.
You know, we've spent 40 years building technology that really extended our intellect, you know, spreadsheets and better spreadsheets and more spreadsheets.
And now we have this technology that can actually connect with humans on an emotional level.
And that's a very beautiful, powerful thing that we should not be skeptical of.
We should lean into.
So I'm going to talk through kind of where we are in the cycle.
I'm going to talk through what are the things that both individual employees are doing as well as the best companies are doing.
Let's talk then about the sort of economic implications and what it means for the structure of how we all work.
Because I know that's on our minds.
And then let's do Q&A.
Does that work?
Perfect.
Perfect.
Okay, cool.
So just to kind of frame up.
what I believe anyway, the two most powerful forces for human flourishing are market economies and technology, right?
We've seen that over and over and over again.
If you look at GDP growth rates pre-market economy, the GDP of the world was essentially flat.
Post-market economy, it grew from flat to the industrial revolution when we started to see 2, 3, 4, 5, even 10% GDP growth rates.
So market economies have been a huge part of human flourishing.
And then technology.
Every time we've had one of these big technology cycles, forget about even the internet for a moment, just look at electricity.
Electricity took 40 years to diffuse across the industrial world.
It drove huge GDP growth.
But the number one thing about electricity is that 80% of the surplus, the sort of incremental value that was created out of thin air, was actually captured by individual humans and consumers.
So this is the story of every consumer where luxuries turn into commodities.
Look at something like therapy.
40 or 50 years ago, therapy was this incredible luxury that most people didn't have access to.
And now, you know, I think most Gen Zs are texting with their therapists.
So there's this story of technology turning luxuries into commodities.
It's played out over and over again.
And the thing that has catalyzed that transition has been technology.
So what are the recent technology transitions we've seen?
I mentioned electricity.
We've got internet, of course, we've got mobile.
Those were all really, really important technologies.
But if you look at the shape of those technologies, what they mostly did was, at least from a professional perspective, help us be more productive, right?
Help us do work more efficiently.
The shape of this new technology is so fascinating because it actually can do work on our behalf.
And I understand why there are some scary implications of that.
We should talk through it.
But it's just a fundamentally different technology in terms of its attributes than anything we've seen before.
Okay, so there's two sort of views of AI, I'd say, right now that seem to be prevailing.
I'll like pejoratively call it the Instagram view and the Twitter view.
So I think the Instagram view of a lot of what's happening is people use ChatGPT maybe 12 or 18 months ago.
They, you know, generated a joke or a poem or an essay or an email.
And they're sort of like scratching their heads saying, I don't really know what the hype's about.
Maybe some new house music.
Maybe some new house music, exactly.
Yes, exactly.
Or some Cyndi Lauper.
They're sort of scratching their heads asking like, hey, what is the hype really about?
Because like this is cool and interesting and I can chat with it and it can write these things for me, but I haven't had the aha moment.
And then there's a small set of people, maybe programmers and technically oriented people that are having conversations on Twitter where there's a term called AI psychosis, which is people have just...
They believe things like money will cease to be a relevant concept in two years.
Now, I think that's too extreme of a view, but it's like 1% of the world has seen God and 99% is walking around oblivious.
And how do we actually connect the dots?
And I think we're going through that transition right now, but the truth is somewhere in between.
And I think if you're, you know, and I'm sure nobody wants to admit that they're in the former camp, but if you're in the former camp, you probably just have not been curious enough or have not revisited the tools recently enough.
Things are changing dramatically from the initial JAT GPT release to what is called reasoning models, which was the DeepSeek release, if you've heard of DeepSeek, to a lot of the creative tools models that have happened this year.
It's just extraordinary.
So anytime you have assumptions about AI, you have to revisit them every three or six months because everything is really, really changing.
How do you, just to ask a question about that, how do you think about the right time to actually invest?
if it's changing as quickly as it's changing, right?
When do you start?
In some ways, the market may punish an organization that invests too early in a particular set of tools or technologies.
How do you think about when we should kind of lean into this journey?
So I think that we have to lean in now.
And I think the thing for us to expect is that things that are not working today.
So let's say we tried customer support and we felt like they weren't up to our standards for member experience.
Customer support agents have improved exponentially.
So you should sort of squint and take bets that things that are not quite working today will work tomorrow.
Because even though the technology is changing, it's sort of compounding on top of itself.
It's not an assumption you made today will be incorrect tomorrow.
So I sort of think we've got to invest now.
And that actually leads to maybe two other quick things.
And sorry to yap so much, guys.
I'll go through it.
Number one, I think, like, what is the shape of the employee or the individual that's going to be successful in this sort of new economy?
So I think it's a lot less about seniority because in a world we're moving from totem pole management to flat structures.
We're seeing that across the board.
I don't know if there's a role for pure, pure management in the way there has been traditionally.
But in this world where almost everyone is an IC, what wins is curiosity.
So the people that are kind of going to the edge and obsessed with the tools and also having fun, it's really hard to compete with people who are having fun.
And because these technologies make the cost of trying things so low, like whatever your idea, however arbitrary it might be, you should try it or find a way to try it with AI.
And those are the people that we're seeing be successful.
As a kind of proxy for this, I was in a board meeting yesterday and I asked the founders, hey, Are you guys using AI?
Of course we're using AI.
Everyone's using AI.
All right, let's pull up your Claude dashboard and let's look at on a per person basis what their token spend has been.
You know, I think the whole company spent $625 last month on AI.
I spent $625 last month on AI.
Like that's not enough AI consumption.
So a proxy for this is sort of token usage and things like this.
But I guess the direction I would point you is curiosity and having fun because these things are very low cost to explore.
Does that make sense?
Okay, so let's now talk through specifically what's working in industry.
So there's essentially three things that are working, chat, code, and what I'll call customer support, but I think is much more broad than that.
So chat, of course, we've all experienced chat and, you know, ChatGPT and all these other models.
I think the way to think about this product category is a thinking partner.
OK, so you really want to be using I use Claude a lot.
ChatGPT is also excellent.
Everybody hopefully has the desktop app or maybe will.
Thanks to your IT team.
I would use it to help you sort of think through your work and generate all the artifacts that you need.
So it's not just sort of, you know, creating emails.
It's creating decks.
It's ingesting decks.
It's trying to reconcile conflicting views in the organization.
It's doing market research.
Every single part of your job should actually be intermediated by one of these models, which is probably going to be a chat window for most of us in the near term.
So I think that's very, very important.
The people that are using chat as a thinking partner are seeing a lot of success.
The ones that are sort of using it as a way to draft emails, less success.
Okay, that's good versus great.
Now let's quickly talk about code.
So how many people here are programmers?
Anyone?
A few?
Okay, great.
Everybody in this room, I'm going to ask this question in 12 months.
Hopefully I get invited back.
I don't know you have too much.
Everybody in this room should be raising their hand because the thing to know about programming and code is it went from this scarce, precious resource to this very abundant, cheap resource.
And you don't have to be technical.
You don't have to understand syntax to produce code anymore.
Every team in the organization should be a software team.
And that can mean many things, but I would encourage everybody to use Lovable, Replit, Wabi, Cursor.
Cloud code, like you can decide which one has the right level of sort of complexity versus, you know, expressiveness for you.
But everybody should be using code and writing code as a part of their job and producing code, even if you don't understand the code that is produced.
That may sound daunting, but I promise there's a way for everyone in here to actually be a software person.
And I'd love to see all the hands go up in 12 months.
All right.
Maybe the last thing I'll hit on is- Can I ask you a question about that?
How do you think about governance, right?
So a lot of times when, We had an instance that happened a couple of months ago, a couple of weeks ago, rather, where an enterprising leader in the company started building his own code for a particular use case.
Then the question just becomes like, how do we as an organization manage a thousand flowers blooming for specific use cases without having enterprise visibility into what's happening, how the data is being used?
And, you know, how do we create scalable and replicable process around that?
Should we be thinking about things now?
Is that an old world?
Is that an old view of it?
So, look, I think that equilibrium is a point that you pass through, right?
So we've got to take risk in some direction.
We're either going to take the risk that we underexplore the technology.
or that we overexplore it and we, you know, potentially we have a security issue.
We potentially even have a regulatory issue.
Hopefully not that, but we're going to have to take risk in some direction.
And I think the risk that we fail to adopt this technology is much more existential.
Fair enough.
That's my view.
Yeah.
Okay.
Maybe the last thing I'll touch on is customer support.
So, you know, I think customer support is a misunderstood part of artificial intelligence.
So I think that the naive view of 18 months ago was...
hey, we're going to take our member support, member experience, we're going to replace it with customer support.
The jobs are going to go away.
The experience is not going to get better.
It's going to be a sort of cost optimization function for the business.
And that's just not the shape of the technology and the sort of deployment that we've seen at all.
Instead, we're seeing a bunch of things.
So first of all, for most member support, member experience people, They're finding that the technology and the customer support agents do all of their low value queries.
So when someone calls in with a very basic question, all those things tend to get routed to these support agents.
The highest value, most complex conversations end up going to humans.
I think the other thing to know is that support is merging with sales, with operations, with collections.
So now you've got people that have relationships, you people.
And you're able to actually fluidly move across sales, support operations collections because you know that the agent can do all of the road stuff.
We're seeing this in places like freight and logistics.
You know, these are, I mean, it's shipping.
It's companies like DHL and FedEx.
These are not super sophisticated technology companies.
And what they're seeing is that because the customer support agents are able to do a lot of the administrative work, they're instead focused on, you know, taking the customer out for a steak dinner and understanding their problems and being able to problem solve in ways that are enterprise value creating and sort of top line growth drivers instead of simply efficiency drivers.
So support is a really interesting and important technology.
The agents have also gotten awesome.
Like they're so much better than the phone trees that you're all used to.
And I think for senior citizens in particular, like we're seeing this with one of our companies, Hippocratic, which I know you're familiar with.
It's an AI nurse that calls senior citizens and reminds them to take their medicine and sort of everything before they have a procedure the next day.
But there's these beautiful sort of emergent experiences where the patients will then speak with the AI, even knowing it's an AI, and start to sort of chat with it and connect with it.
And of course, the models are infinitely patient and infinitely interested.
So you start to see this sort of moment of human connection, even though it's not a human on the other side.
And you think about sort of scaling that, both for us as humans and for all the people who can't access someone in the room.
I think there's a really beautiful picture on the other side.
I will stop yapping now.
Where should we get started?
I mean, we have some amazing skunk works going on in the company.
We saw some use cases yesterday.
But not knowing our business at all, where do you think we should be leaning in to get started?
I think it's the three things.
And what kind of leadership should we be demonstrating as leaders of the enterprise to actually really get started on this journey?
Well, so I guess...
A couple of things.
So I think we should start in the three areas that I've discussed.
So let's make sure everybody has access to the ChatGPT desktop app and the Cloud desktop app.
You can sort of pick your personality.
Let's get a hackathon going where we get every team actually, you know, let's pick a platform and let's build an actual workflow that we might use in our day-to-day.
And third, I think for the customer support thing, I'll help with this, but let's talk to a bunch of vendors.
Instead of saying, hey, how do we actually handle the lowest value work?
Let's figure out how we actually drive.
top line from a strategy perspective with these support agents.
The number one thing is let's just get everybody using the technologies.
I think we should maybe have a leaderboard of token consumption, which your CFO is going to hate, I'm sure.
But things like that, I think, are really important.
It's just about the technology is, you know, it feels alien if you don't use it.
If you use it, it actually feels like, hey, I get it, you know?
Are there examples of legacy enterprises that you think are inspirational in their kind of...
in their AI transformation journeys that we can look to as an example.
Because we are, you know, we were talking about this before we got started.
We're a company that was founded in 1977.
We have a lot of legacy process.
We were not born out of Silicon Valley.
We're a not-for-profit.
At the same time, you know, part of what we're doing right now is refounding the company, right?
They're going to talk about it.
Like I said, 50 years from now, they're not going to be talking about the 12 Angry Seniors.
They're going to talk about the 200.
100 angry executives?
I don't know.
We'll figure it out.
But the point is, we have an opportunity to truly re-found the company right now.
But we have a lot of legacy process, mindsets.
What are some inspirational organizations that we can look to for an AI transformation journey?
So I'll give you an unusual one that's not in your industry.
And maybe we can talk about some of the health care, what we call the white pill, the pro case, in a moment.
So C.H.
Robinson.
It's a freight brokerage company.
It's public.
If you look at their earnings reports for the last year, they have not meaningfully reduced their workforce.
And they're doing it because they're taking this technology.
One, they're driving technology literacy into the organization, asking people, in fact, demanding that employees use the technology every day.
And the second thing is that they're thinking a lot about, hey, how do we maximize our ambition instead of use this as purely an efficiency driver?
I actually think that's the mistake that most incumbents are making today inside and outside of technology, which is, They're insufficiently ambitious around what's possible.
I would start with what is the most incredible A-plus member experience we could deliver?
Imaginable, with no restrictions.
And then let's sort of work backwards from that, given the technology we have.
So how do you overcome a curmudgeonly, like, CFO who's just thinking about it?
I mean, brother, if you know the answer to that question, I'd love to know.
Yeah.
Look, it's...
This is not IT.
This is not a cost center.
This is a value driver.
We've just got to think about it that way.
Again, equilibrium, a point that we pass through.
We're going to take some risks.
We're going to waste some money.
Sorry, CFO.
But on the other side of it is just a much better shape of a business.
And by the way, also a much better shape of employee experience.
You look at the 20% productivity increase.
Jobs are not tasks.
AI is good at doing tasks, but not entire jobs.
And I think it probably shows up.
closer, maybe you'll hate this, closer to a four-day work week than 20% less jobs.
So I think that on the other side of this transition is not just a better sort of business model configuration, but a better employee experience for all of you.
No, I mean, look, one thing that happened this past year with all the growth that you referenced, we added 127,000 new members.
Does anyone know how many staff we added to serve the 127,000 new members?
About 600 plus some temps, plus some contractors, right?
So the question I think really coming out of this is if this coming AEP, we add another 127,000 members, maybe we had 150,000 members.
How do we do that with zero additional staff, right?
I think that's a great challenge for us to embrace as a company.
As you guys think about it such in like, what are you worried about?
What are you most ambitious about?
I mean, give me the sort of shape of the discussion you've had for the last two days.
You know, I'm worried about dollars and cents thinking, honestly.
not being ambitious enough, not being visionary enough, you know, really kind of getting, we're a company with a lot of pressure on us right now.
We have got reserve capital that we need to maintain.
You know, we don't have unlimited budgets to invest in this.
We don't have venture capitalists that we can go to for help when we need it.
Can't raise a series EFG, H-I-J-K-L-M-N-O-P.
We can't, we can't.
Everything we invest, we actually have to make.
And so, I think we do have to demonstrate an ROI quickly with AI to build some of the momentum around it.
But then I worry about also just not being aspirational.
I worry about being left behind with our competitors who do have bigger bank accounts, bigger budgets, making bigger leaps than us.
Yeah, that makes sense.
Questions from the group.
Benish, thanks for being here.
Of course.
Thank you.
Wanted to ask you, When you're thinking about investing in, you know, that are, you know, earlier in the revenue trajectory, and you think about the end state of where they're going to go, and you think about how their end revenue markets are going to, you know, potentially change dramatically.
How are you thinking about that as an investor right now?
And what's under my question a little bit is SaaS companies have taken, you know, quite a drawdown year to date.
And, you know, the prevailing theory is they sell seats.
Legacy companies who buy those seats are going to need less seats.
Hence, they'll sell fewer seats.
How do you think about that, you know, from where you sit?
Yeah, so maybe just to frame it up, all the sort of pure software names, you know, everyone from ServiceNow to CrowdStrike to Salesforce to Oracle have taken a beating in public markets this year.
They've been dramatically down.
And the genesis of that is Anthropic released a model called Opus 4-6 in December, which is a very, very powerful coding model.
This is sort of what drove some of the psychosis I mentioned, people feeling like they've seen God in this new model.
It's just a great coding model.
It's very, very good at producing code.
And then the sort of public market reaction was, well, maybe it's trivial to reproduce any of these sort of enterprise software products, and therefore corporations won't be buying them, they'll be coding them, and therefore all these software stocks are worth zero.
So maybe a few notes on that.
So one, I think the market's hugely oversold software.
If you look at a lot of these systems, they're regulated.
you know, imply enormous liability if you get things even a little wrong, like payroll.
So like, sure, you could vibe code your payroll, but like if you're a penny off, somebody's going to jail, which is a problem, right?
So like one, I think we're underestimating how many of the system touch regulatory guardrails.
Then number two, if you actually take a look at this incredible weapon that we have, this innovation weapon we have, are we scan going to point it at replacing payroll and saving?
I mean, what, $5 a payslip per month?
Or we can instead point it at driving our sort of ambition, our member experience, our top line.
So just the reality of we have these incredible coding agents.
They're not going to be applied to taking what is an 8% cost center out of the enterprise down to 5%.
They're instead going to be pointed at driving top line.
I think the market's oversold.
I do think the one thing that's happening that's maybe not discussed is that, you know, SpaceX, OpenAI, Anthropic.
are reported to be public in the next 12 months, and they're going to be in the index right away.
So I think a lot of public market investors are sort of selling other names to create capacity for these names they'll have to buy.
It's all financial market idiosyncratic behavior.
Yeah, I mean, that's not financial advice, but I do think that sort of there's an overcorrection.
And look, like, it's funny because it happens every time, you know, like the internet was supposed to kill all of these companies.
Google was supposed to kill Microsoft.
You know, Microsoft was supposed to kill IBM.
IBM is a $250 billion company.
Microsoft's bigger than ever, right?
Multi-trillion dollar company.
Adobe has survived, I don't know, nine product cycles.
So the shape of disruption is often that the incumbents sort of stay leaders in their existing categories and new categories are created where they're irrelevant.
And I think that's some of what we're seeing now.
Yeah.
You do.
Depends.
How good at yelling are you?
So one question, just at the way SCAN, we've organized AI and IT are slightly in different departments.
Of course, we all collaborate with each other.
But any advice about, so I run a business unit that does homeless healthcare, and it's a portfolio company of SCAN.
I kind of have like two directions in the organization to go for AI versus like our EMR and our IT team.
Advice on how those worlds kind of come together within an organizational perspective and how we can make that more seamless.
Sorry, what are the two?
The AI, sort of Amin coming on board, will kind of have a team and Corinne and other people at Scanner, the AI folks, and then the...
To give you some context, we like have a strong conviction that the AI transformation that needs to take place is actually a human capital transformation.
So we actually have positioned our AI department...
inside of our people organization.
Close collaborators with our IT organization, but not synonymous with our IT organization.
So I think the spirit of the question is, how do you think about organizing AI efforts in a company to really amplify their success?
So I think there's a bunch of models for it.
I think the most important role that the IT team can play is just creating infrastructure for people for curiosity.
so that every team can kind of pull the threads they want to pull.
I think the most important thing that management can do can encourage the curiosity.
And also just say, like, look, we're going to make some mistakes.
We're going to waste some money.
That's okay.
I think it's got to be both centralized and decentralized because ultimately it's going to come down to every member and individual on your team knowing exactly how they want to use the technology, IT creating the infrastructure, and then Amon and the centralized AI team, you know, distributing best practices.
Like, you know, one of the models I've heard is that many divisions would have a sort of AI lead.
who would liaison with a sort of centralized AI team just so they can discuss what's working, what's not, help sort of, you know, drive best practices through the organization.
But it's going to be a little messy.
I think the number one thing is we can't be like sipping the innovation through a straw, right?
Everybody's going to have to have their own, I don't know, milkshake or whatever and have their own, like, be using it directly themselves.
You know, we couldn't have imagined where the world was going to go, you know, kind of pre-ChatGPT.
You know, maybe I'll ask you an unfair question.
Five, 10 years from now, What does the world look like?
I mean, it's hard for us mere mortals to really kind of think about that.
So given that you're on the bleeding edge and you're seeing the most interesting companies before the rest of us are seeing them, what does our life look like in five or 10 years?
I mean, it's, it's, so maybe I'll tell you the things that seem obviously true.
And then let's talk about maybe some of the controversial things I believe.
So look, I think what's obviously true is that we're going to dramatically increase the NPS of the human experience and of the employee experience.
You know, let's like track the human experience, the NPS of it over time, you know, in the prehistoric days, it's like.
Didn't get eaten by a lion, like pretty good day.
You know, it was like the average day was not that great.
You know, all the way through to the industrial revolution where you're like, all right, didn't get like killed by a machine at the factory.
Like, great.
It was a good day today, you know?
And look, even today, if you think of our day-to-day experience, look, there's just like a lot of things that are a drag.
Like think about the parts of your work where you do your best work and you show up as your best self and the parts of your work that are administrative, that are overhead, that are coordination costs.
I mean, this is why people like working at smaller companies.
If all of that overhead goes away, if we're all able to find a way to be even more productive and perhaps work a little bit less, I mean, even think about politics at work, right?
I've, you know, worked in many organizations as all of you have.
And I found one source of politics is that there's this misattribution of professional judgment to things that feel personal.
You know, this decision went against me because it was personal, right?
Or this person doesn't like me or whatever, whatever.
And when you actually have an AI making decisions and tie-breaking, you know, look, you may not like the judgment, but it'll never feel personal.
In my view, people feeling that process fairness is there is more important than outcome fairness.
So it's possible we reduce corporate politics.
It's possible that we work less.
It's possible that we just have more sort of abundance around us.
Also think that there's no ceiling on human desire.
You know, people say, like, where will all the jobs go?
Like, we arguably already have fake jobs.
And there's going to be all kinds of new jobs, right?
We're going to want vacation homes on Mars.
We're going to have all kinds of new luxuries that like we need to have.
So I just think that the idea that there will be no more jobs because we'll make the existing work too efficient is probably overstated.
Janice.
Hi.
Hi.
So I'm going to go back to a theme I think we're hearing.
Yes.
So what I've seen in the last few months is the opportunity.
Thank you.
Or the cycle from going from business expertise in a problem you need to solve to an actual solution has shrunk dramatically.
How does a legacy organization take that and actually implement versus just generating a lot of ideas that don't quite fit into how it works today?
I think there are small changes.
For example, no more presentations.
We're only going to show prototypes.
You know, and look, they can be throwaway prototypes.
They can be sort of illustrative.
You know, no more asking the design team and waiting on them to create marketing artifacts.
We're just going to generate our own.
So it looked like I don't think we refounded the company.
We don't need to think of ourselves as a legacy or an incumbent or old guard.
Like we're none of that.
Right.
We refounded.
We're going to be native in this technology cycle.
So that's where my head goes.
One question.
A lot of what we do is we sell and we distribute.
How do you see sales?
changing i mean we we're a very retail business we work through brokers we have our own field sales force we have a telesales force yeah um and you know a lot of our customers buy our product because they trust the person that is selling it to them yeah how do you see distribution changing in other industries and how do you think we should be thinking about distribution?
Yeah, so I think there's two, this is a really interesting topic.
So there's two important points here.
So one, I think that there is a set of sales that are so high stakes, they have to be human led, but we as human sales leaders have to make all of these trade-offs.
We have administrative overhead and we have other accounts.
So I think we're going to have this barbelling where our most important accounts we're going to be able to dedicate almost unlimited time to and close these very, very high value in-person sales.
So that's part one.
I think on the lower value sales side, this combination of sales support operations collections, like if you think the archetype of someone who's good at support versus sales, right?
Support is like you're patient, you're a listener, you know the product surface, you really like people.
The archetype of a good salesperson, you're kind of a yapper.
You know, you're always in a good mood, a lot of momentum.
Do we have any yappers at this company?
Stanton?
That's right.
That's right.
Sorry, man, you earned that one.
I may be a bit of a yapper myself.
But these, look, so these are like two, the reason we've always organized the functions this way is you got two different human archetypes and there's very few people that are both good yappers and good listeners.
And now actually because of AI, for the lower value sales, you can have AI do things like cross-sell on a support call.
So even these historical silos that we built, I think we can break down with the technology and we pair that with a lot more high value human led sales.
Kate.
I have a thousand questions I could ask you, so I'm going to try and sneak into.
The first one is, and you've kind of touched on this a bit, but are there specific KPIs you use to make sure that the member experience as we evolve it is really driving true enterprise value and not just shiny UX?
And then the second question specifically is, do you think voice is kind of the leading channel that you'd be focusing on, especially in healthcare as we think about AI innovation?
Yeah.
So, I mean, CSAT is the obvious one to me, but I'd also encourage you to like listen in on the calls.
It's really, really amazing.
There's a strange thing that happens when even a human knows that they're talking to an AI.
We're so, it's like our animal lizard brain is so trained to start connecting when we hear something that.
has human shape on the other side, that you see people start to have these really deep patient conversations and, you know, especially our audience.
So I think, I guess I'd encourage you, of course, we're going to look at CSAT at the global level, but I just like, you know, do some sort of, pull some threads and see what the conversations feel like.
And usually whenever there's like a disconnect between, you know, sort of empirical experience and data, I always trust empirical experience.
So I would do those two things and just see if you can reconcile them.
Yes, I think voice is the most important thing that's happening in healthcare.
It's the most important thing that's happening in enterprise AI.
Healthcare is the number one buyer of tokens, at least at OpenAI.
So it's like, we have to be very front-footed on this trend.
Yeah.
So we're not naive.
We know that down the street, one of our competitors, either next week or last month, is having a meeting just like this one.
Yeah.
They're talking about all the same things we're talking about.
What do we need to do to stay ahead of that?
Yeah, I mean, look, I think a lot of it is what you've been talking about already, Sachin, right?
What's the sort of shape of the risk we want to take?
I think that's something.
I think another thing that can hold management teams back and companies back is the need to be consistent, right?
This is a moment where we're in wartime.
And I think there's a human tendency to want to be consistent with past policy and management decision and culture.
And I think we're going to have to be a little inconsistent to put our arms around this.
The good news is a lot of it kind of comes down to courage and curiosity.
And I don't know that all of our sort of competitors will have.
it anything any advice on how we can foster more courage and curiosity i think a lot of it is how do we make it high status at the company right how do we not just measure it but how do we make sure that you know the most important executives all of the people in this room are spending real time on it You know, are each of you, when you go back, let's do like a one hour meeting at the end of the week where you have key sort of people that are the most curious in your team, just demoing what they built or what they've tried or what they've learned or talking about it.
I think it's important for IT to get everyone the technology.
So it's not just the ChatGPT and ClawDapp.
One thing we've done at Andreessen is we're giving every employee an OpenClaw instance, right?
We're experimenting with how we can do that safely.
I'm sure that's scary for IT and finance, but we've got to take experiments in this direction.
And then we also have to create a culture.
We're working on these things as high status.
Hey, thank you for being here.
One of the questions I have is on integrated technology.
And, you know, for example, CRM, which is Salesforce and ServiceNow.
Where are you seeing industry moving in creating integrated applications all through, starting from, you know, let's say sales through?
end of data warehousing, for example, or data analytics.
Is that a trend that you're seeing where big players are able to not just do pointed, you know, focus solutions or they now they are able to, you know, with the speed of innovation, being able to build a lot more things faster?
Yeah.
Where is that you're seeing?
For sure.
I mean, It's like software teams are ripping through their roadmaps.
You know, I talked to a friend at Google and he said, while they're not, they're not laying anyone off zero people.
They're just, they have almost no backlog anymore.
There's no prioritization because they're just getting everything done.
So that's one trend.
Like they're, you know, just that the scope is insane.
The second is that as always, people try to commoditize their compliments.
So yes, people are trying to vertically integrate and make the thing, you know, their entire software supply chain underneath them less have less leverage.
But I think that.
And you want to focus your tokens on areas that are sort of highly accretive to your business strategy and your sector.
So then the important thing is for all the areas like payroll, again, I don't need to keep picking on payroll, but it's not something where you can get a kind of 10x on the other side, even if you sort of put a 10x into it.
So we're not seeing companies invest there.
I think the whole thing is how can our company be the point of economic diffusion for model progress to our members, right?
Just, I know it's a complicated question, but for a moment, like if the models got 3x better in the next six months, how do our members get 3x more value?
To me, that's a sort of measure of whether we're staying on trend or not.
Hey, Anish.
Thanks for being here.
My question is more about, you know, your past when you have talked about how it is important to have local processing for coding agents.
And I just wanted to understand your thesis on why.
we would do edge versus cloud, kind of, if you can talk about that.
Sorry, can you elaborate?
So, you know, you talked about how coding agents need to be on the local processing instead of cloud.
Oh.
So, you know, if you can talk about that, that'll be great.
Oh, yeah, no, I don't know that they need to be, most of the models can be run locally anyway, so I think they will be cloud.
I think that there are different ways that, like...
Coding intelligence and this AI intelligence, it's a primitive, right?
It's just like compute.
And many people will want to consume the primitive in different ways.
So if you're a developer, maybe you want to consume it with, you know, quad code or codex at the command line.
If you're somebody who's maybe technically literate, but not a programmer, you want to consume it through quad co-work.
If you're somebody who's completely non-technical, you maybe want to consume it through Replit.
And if you're a pure consumer, you'd consume it through Wabi.
So I think there are different product interfaces for the same intelligence primitive.
I think our customers should think about this in a similar way, the least to the most sophisticated.
So it's sort of a product packaging pricing question versus should it run locally versus the cloud.
Yeah.
All right.
Thank you for being here, Tim Ibrahim.
I'm asking about, so you talked about the 99% and the 1% who have seen God.
I'm thinking about rolling out AI interventions within our organization and how it's sort of been asymmetrically adopted across service lines.
And so when you have a service line that's adopted, you know, Claude or ChatGPT or, you know, a chat instance of an AI tool and you have a marketeer in your, you know, on your desktop or you have an analyst in your desktop.
And then when you're able to do those things without, you know, you know, pinging another team.
And there's a sense of stepping on toes where you've said, oh, here, look what I was able to produce using XYZ.
What lessons have you learned in sort of implementing AI with this probably near-term problem of asymmetric adoption and sort of managing those sort of culture issues?
You're right.
I think it's a trend, you know, I think it's Mark called it the kind of Mexican standoff between product design and engineering right now.
where every PM is driving the engineers crazy by generating their own PRs, and then all the designers are pulling their hair out.
I guess it's mostly a PM problem because the PMs are doing designs.
And so, look, there's some convergence of all the functions because we're going from specialists to generalists.
But I think that the reframing is that now you need to rely on your sister teams less, which means they can do higher value work and you can be more self-directed.
But look, I don't have a great answer other than it's going to be a tension for us to all navigate.
And probably the best way to start is by acknowledging that, you know, the shape of our jobs are changing.
Thank you.
Andy.
Hi.
Thanks a lot for being here.
We're doing a lot of technology modernization here at Scan right now.
And so we can only do so much at once.
I'm curious, your thoughts.
What do you think we do first?
Like in terms of foundational capabilities that will allow us to do everything else better, like where would you start now?
Is it, like let's say, platforms?
Is it data?
Is it skill sets?
Like where do we get started?
I think the most important thing is to just get the new tools in every employee's hand.
Let's just start there.
And I'd probably pair that with the customer support, the most ambitious version of customer support that I outlined.
I think let's just start simple.
Otherwise, it's overwhelming.
And also the language, it's like the data, the systems, the inference, it's too much, I think.
So I would just start very simple.
Let's get ChatGPT and Claude on every person's desktop.
Let's use customer support, redefine it to be much more ambitious on behalf of our members.
And then let's go back and collectively study our efforts in three months.
Simple as that.
Hey, Anish.
How do you see the regulatory environment developing here?
And more importantly, how do you advise companies on getting into some good trouble while they're waiting for the regulations to catch up?
Compliance team?
Yep.
Yeah.
Great question.
Yeah, look, I mean, OK, so without commenting on maybe AI regulation itself, because I just don't know, I think that agents can't be liable for things.
Humans have to be liable.
So, I mean, that's an obvious thing, but worth explicitly stating.
I think we have to very carefully define what our third rails are.
And of course, we have to protect against those third rails.
So I think there's sort of business mistakes and regulatory mistakes.
Now, regulatory mistakes are much more expensive, but I also understand that regulation is, you know, sort of shades of gray.
I'm guessing that you too will probably have to take more risk than you're comfortable with.
And, you know, we're all going to have to trust you to tell us what are the true third rails.
But it is important to like, like the models can't be liable for things.
Now, actually, the good news is that that's a moat for us against labs and model providers.
They don't want to touch regulated businesses in a direct way.
So I think if we're thoughtful about how we organize the intelligence primitive around these regulated industries.
then that can be a real point of offense for us.
But, you know, it's going to be a tension for us to navigate.
Yeah.
Thank you.
How do you get people to be ambitious enough and not just stuck in incrementalism?
I mean, I don't know.
How do I get you to work more hours?
I mean, it's, you know, it's like it's all going to come down to us.
I think part of it is not having a fear of the technology.
Like, here's an example.
So there's some nomenclature, hopefully, that everyone's familiar with, but I'll outline it anyway.
So everyone knows what a prompt is.
Then there's something called a skill.
A skill is basically a prompt that you've written that you can reuse.
So instead of rewriting, like I use Claude to triage my email and I have a very long prompt about, you know, find all the messages from my partners, all the messages from Mark and Ben, all the messages from portfolio companies, find promises I made that I haven't kept yet, find, you know, interesting social things that people, like all those, it's a very long prompt.
So now I have a skill called slash email triage.
So I use that skill.
And then you've got plugins, which are collections of skills that you can share with people.
I've seen two views from employees and organizations.
One is like, wait a sec, my prompt is my job.
So I got to protect it.
Last thing I'm doing is writing my prompt down and putting in a skill.
And the other is like, well, look, if my job can be defined in a prompt, then that's not really a job anymore.
So I should probably just be the one to put it in a skill, distribute it to the organization, and then figure out what else I should work on.
And a lot of that is just mindset, a sort of like fear versus abundance.
So, you know, I think that's probably the one thing that we can do, which is get each other comfortable that like, hey, like, you know, if the parts of your job that can be defined or prompt, you probably don't want to be doing anyway.
So let's just write that stuff down, distribute the best practices.
And like, you know, all of us uplevel, spend more time on strategy.
Yeah.
I have a question.
Where is it coming from?
Oh, hi.
Hey, Ida.
Thanks.
Sorry.
I'm going to ask the money question.
Uh-oh.
Because, you know, finance.
So I would love to get Claude and ChatGPT and all of these tools out to everybody.
But with the way that the pricing is being done in the industry, they're still trying to figure out how to charge for it and how to use it.
What kind of guardrails do you put around some of the use of it so that it doesn't balloon out of control unexpectedly?
Yeah.
Because it could.
And it is.
Some companies are getting flat.
They're falling down because they did not expect the cost to exceed.
Are we a public company?
No, no, not for public.
Thank God.
All right.
Great.
Nothing market moving.
All right, great.
No one's going to trade up.
That's good news.
Look, obviously, we should not spend so much money that we bankrupt the company.
But look, I think we just, we're going to need to have a budget for this.
And I think the very difficult thing to do right now is to say which are high value tokens versus wasted tokens.
You could argue that my email triage skill is a massive waste of intelligence.
You know, it's like $5 every time I check my email.
That's not great.
But I think we're going to just have to pay some of that cost while we figure out how we use these things efficiently.
And there's always this sort of exploration and convergence phase, right?
And we're just in the exploration phase.
It's going to be a little bit expensive.
There's going to be wasted tokens.
You know, you will have controls at the organizational level of like, what is the max spend per person per team?
So we can institute those.
I guess I just encourage you to like...
consider it a cost center in support of future strategy.
And, you know, we will, it'll be a little painful.
Thank God we're not public market investors and we will reconcile it and see it in future upside.
If we spend two years and we kind of blow the budget out on tokens and we see no upside from that, then like it's a good time to reassess.
But right now I just, we've got to take risk in some direction and the shape of that direction is probably unpleasant to the finance department.
Anish, you talked about, you know, it's a wartime mode and you've also talked about the messiness, the moment and the tension.
Can you just take us a little bit into the mindset that you have or Andreessen has to get through those obstacles because those are real things, but sometimes acknowledging we're in a messy moment, what helps you get through.
So just kind of take us through the mindset of that because these are hard.
tensions we have to figure out to know what our third rails are and then in order to invest.
So give us a view into your mindset of that at Andreessen and how you're handling any of these messy conversations, number one.
And number two, can you tell us a little bit about what you think the future of work is going to look like, you know, between departments?
But you started, you touched on it in five years.
How should we be thinking about that?
Yeah, so I think the thing about wartime, everyone should read Hard Thing About Hard Things, which I know is one of your favorites.
You have read it.
It got, you know, some of the salty language got received pretty badly.
I re-recommended it to folks.
This was four or five years ago.
It's wartime, you know.
You'll forgive the salty language.
Sometimes we curse in wartime.
Yeah, look, it is a messy moment.
And I think the thing to just internalize is that everything we knew is upside down or at least zeroed out.
So let's just start from no assumptions.
Number one, let's not feel the need, as I mentioned, to be consistent with sort of past approaches and past thinking, even past culture.
I think number two, all of us are going to feel so much more empowered if we use these tools and technologies and start to familiarize ourselves with them and start to, you know, put ourselves out of a job by creating the plugin or the skill or whatever that represents the work that we don't want to do.
You're right, there's going to be new points of sort of cultural tension, as you mentioned, around people doing work that other functions had historically been protective of.
But there's also going to be new points of leverage because all of a sudden you don't have to wait for the software team to assign you an engineer to like explore.
You can just try it on the weekend and go and show such in or whoever.
And, you know, so like there will be these puts and takes across the organization.
And as always, we're going to have to be very, very patient with each other.
At least I'm paid to be an optimist.
So you can, you know, you can discount this as you see fit.
But look, I do think on the other side is, you know, potentially 10% GDP growth, potentially a four day work week.
And by the way, the way that we make sure this technology benefits all of society is we make important things cheap.
And the most important thing is healthcare.
45% of healthcare costs in industry level is administrative.
If we can make that more efficient, you can actually start to see deflation, not just disinflation in healthcare costs.
And that is my greatest white pill for society in the country.
So let's go do that.
And I think we'll all be happy.
Awesome.
We'll end on that note.
Thank you.
Thanks for having me.
Cheers.
Thanks for listening to this episode of the A16Z podcast.
If you liked this episode, be sure to like, comment, subscribe, leave us a rating or review.
and share it with your friends and family.
For more episodes, go to YouTube, Apple Podcasts, and Spotify.
Follow us on X at A16Z and subscribe to our Substack at a16z.substack.com.
Thanks again for listening and I'll see you in the next episode.
This information is for educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product.
This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z.
Such advertisements, companies, and individuals are not endorsed by AH Capital Management LLC, A16Z, or any of its affiliates.
Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.