1 Apr 2026 · 37m · Maithra Raghu, Rana el Kaliouby, Bob Safian · 18 quoted facts
Maithra Raghu, CEO of Samaya AI, argues that AI will grow the labor market rather than shrink it. As AI agents become more capable and run around the clock, she says more people will be needed to give them the right context, frameworks and judgment. In high-stakes fields like investing, reliable results come from carefully built systems with humans guiding them, not from one general-purpose model.
The central tension is between disruption and opportunity. Raghu admits some roles will change heavily and some tasks will be fully automated, yet she says the bigger story is new kinds of work, such as people who oversee AI agents. She also warns that strong benchmark scores don't reflect the messy, open-ended questions real investors ask, so success depends on human input.
Raghu says finance benchmarks use one precise question with one numeric answer, while real investors ask vague, complicated questions with many acceptable answers. Samaya built a rubric-based evaluation called Criteria eval that grades how many important points an answer covers.
Raghu argues that agents running overnight produce huge amounts of output and need people to supply judgment frameworks and context. She compares it to machine learning researchers who carefully set up overnight training runs, because a poor handoff wastes the work.
Raghu says 2024 and 2025 mostly produced workflows, which follow a fixed set of steps. Real agents, which she dates to late last year with tools like Claude Code, decide their own steps as they work toward a goal.
Raghu says Samaya builds systems with smaller, domain-trained language models because larger models repeatedly stumble on certain accuracy and precision problems in finance. She compares this to self-driving cars, which are systems of many components rather than a single model.
Because markets exist when people disagree, Raghu says the AI must give different users opposite answers based on their own views and firm context. Samaya's causal world models trace cause and effect from big themes down to specific companies so users can see why a prediction was made.
Raghu says some roles will be transformed and some tasks fully automated, but she expects new jobs, such as people who oversee AI agents, and says finance and economics degrees are not obsolete because human judgment and taste still matter.
Raghu says general chatbots give similar answers to similar prompts, but markets need AI personalized so that different investors get different, even opposite, answers based on their own views and firm context. She also says larger models often trip up on finance-specific accuracy and precision issues.
Raghu says a workflow follows a hard-coded set of steps with no real decision-making, while an agent is given information and a goal and decides its own steps as it goes. She says true agents only arrived in late last year.
Raghu says common benchmarks use one question with one number as the answer, which doesn't resemble real work. Samaya's Criteria eval uses a rubric of things a good answer should cover and grades answers on how many they touch.
Samaya builds AI agents for investment decision-making that research an investor's thesis, check it against the firm's frameworks and portfolio, and report back takeaways. Raghu says the agents work alongside people and don't execute trades by themselves yet.
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
Maithra Raghu began collaborating with researchers at Google Brain in 2015, before OpenAI existed.
“I started collaborating, working with folks at Google Brain all the way back in 2015, and to put that in perspective for where the the AI space is right now, this was before OpenAI even existed.”
Facebook had just started Facebook AI Research with Yann LeCun joining, when Raghu began working with Google Brain.
“Facebook had just started the the Facebook AI research effort with with Yann LeCun joining them, and and that was that was basically it.”
Raghu says international math Olympiad competitors are drawn to Trinity College, Cambridge.
“if you compete internationally in an Olympiad, that just takes you straight to Trinity College, um, in Cambridge, uh, that just pulls everybody across Europe who's competed in these these Olympiads.”
Maithra Raghu had about ten years of experience as an AI researcher before founding Samaya AI.
“I'd had ten years as an AI researcher at that point. I had this beautiful body of of work.”
Raghu says that from Google Brain she saw large language model capabilities coming together around 2019 and 2020.
“sitting where I was in Google Brain, you started seeing the pieces come together in 2019, 2020 or so.”
Raghu says the wider world noticed large language models when ChatGPT was released in late 2022.
“I think the world saw it in late 2022 when ChatGPT came out and everyone started paying attention.”
The name Samaya means time or a moment in time and has roots in Sanskrit.
“Samaya actually means time or or moment in time. So I I really like the name. It, um, has a history in in, um, Sanskrit”
Rana El Kaliouby lists Eric Schmidt, Yann LeCun and Mark Cuban among Samaya AI's investors.
“that includes former Google CEO Eric Schmidt, Jan LeCun, Mark Cuban, who we've had on the show.”
Samaya AI recently announced new investment from Nvidia's venture arm NVentures and from Databricks Ventures.
“you also announced recently a new investment from N Ventures, that's Nvidia's, uh, venture arm, as well as Data Bricks Ventures.”
Samaya AI created Criteria eval, which grades answers against a rubric by how many rubric items they address, instead of checking one numerical answer.
“is we actually created an evaluation rubric that you would use to evaluate different types of answers and then grade on sort of how many different things did they did they touch in the the rubric.”
Samaya AI has an internal project called causal world models, with some results shared as a research preview.
“One thing we've been working on internally, so it's a it's a project we have called, um, causal world models. Um, we posted some things about it in research preview”
Raghu says the term 'agent' became popular around 2024 but real agents did not appear until late 2025.
“if the the term agent became very popular in 2024 or so, I don't think we had real agents until about late last year.”
Raghu defines an AI workflow as a hard-coded, specific set of steps, unlike an agent that makes its own decisions.
“a workflow is very hard coded. So it might be multiple steps, but it's a very specific set of steps.”
Raghu says Claude Code, released in late 2025, was an early example of real AI agents.
“late last year, you know, I think Claude Code was like an early beginning of this.”
Raghu cites Waymo's self-driving cars as an example of an AI system made of many components rather than a single model.
“Actually took a a Waymo in today, and, uh, and but that's a that's a system, that's not a single model by itself”
Samaya AI focuses on investment decision-making within financial services.
“Samaya is laser focused on investment decision making, which we think is at the heart of the the industry.”
Samaya AI's stated mission is taking people from information to conviction.
“the company's mission is taking people from information to conviction.”
An ad in the episode states that 86% of the Forbes AI 50 run on Stripe.
“86% of the Forbes AI fifty run on Stripe, along with industry leaders like Salesforce and Pepsi.”
Raghu introduces her central claim that overnight AI agents need good human handoffs and judgment frameworks, much like machine learning research.
“we'll need more and more people in the loop so that we can make sure that they all have the right frameworks”
Raghu describes her math Olympiad path to Trinity College and early research at Google Brain, where unwritten know-how mattered.
“this was before OpenAI even existed”
She explains why she became a founder once large language models made AI general-purpose enough for real-world use, and what the name means.
“Samaya actually means time or or moment in time.”
Raghu explains Criteria eval and why investors need personalized, differing answers rather than one-size-fits-all responses.
“progress on benchmarks does not translate to to progress in the real world.”
She describes tracing cause and effect from macro themes to specific companies, and why Samaya uses systems of smaller specialized models.
“we're all about systems over models.”
Raghu weighs disruption against new jobs like agent shepherds and new investment opportunities.
“there's a whole whole set of, like, jobs around these, like, AI agent shepherds”
She separates workflows from true agents and shares her mixed experience building a digital version of herself.
“I think this year is gonna be the year of real personal, um, of of real personal agents.”
Raghu calls human-in-the-loop slowdown a myth and says thriving takes energy and experimentation.
“Energy and experimentation.”
Stripe is built for whatever comes next because AI isn't just changing business, it's evolving fraud. Stripe Radar detects fraud with the fast scale of Stripe's 1.9 trillion dollar annual transaction volume. In fact, when a card is used, there's 92% chance the Stripe network has seen it before. That means emerging fraud patterns are detected sooner, blocking bad actors, welcoming good customers. Learn more about fighting fraud with Stripe at stripe dot com slash radar.
On Masters of Scale, iconic leaders reveal how they've beaten the odds.
Asking really strong questions is a superpower.
You want to show up with something radically different and how they've grown companies to incredible heights.
The greatest rewards always come from the greatest risks.
Let's hit the gas.
Airbnb, Zillow, Microsoft, Liquid Death, and more. Hear from the founders who've changed the game. It's anything but business as usual. Find Masters of Scale on Apple Podcasts, Spotify, YouTube, or wherever else you get podcasts.
The business world is moving faster than ever, and when change hits, we need to learn in real time. On Rapid Response, you'll hear candid conversations with CEOs and leaders making tough calls about AI, human talent, responsibility, and the bottom line. How they navigate uncertainty, pressure, and high-stakes moments. I'm Bob Safian, former editor-in-chief of Fast Company, and I'll be your host as each episode breaks down what you need to know right now. You can find Rapid Response wherever you get your podcasts.
I think actually, as the AIs get better, paradoxically, and the thing no one's talking about is that we'll need more and more people in the loop so that we can make sure that they all have the right frameworks for judgment and all of that.
It's kind of a big mental and emotional shift to move from research to entrepreneurship, to leave big tech, go start from scratch. What was that transition like?
I always say the motivation for doing a company is really, really important. I think it has to be quite pure in some ways. I think you have to leave not caring too much about, you know, the exact outcome or where it goes.
If you could wave a magic wand and have a personal agent that can do anything for you, what would you want it to do?
My goodness, uh, probably clone myself, you know, be me in in all of these places.
Maithra Raghu is the co-founder and CEO of Samaya AI. They're building AI systems for the financial sector. Think hedge funds and investment firms like Morgan Stanley. Their AI agents tackle complex analysis that financial professionals do every day, which raises some big questions, even if you don't work in finance. What happens when an AI agent can do your work? And when billions of dollars are on the line, how do we ensure that these AI systems are accurate?
Plus, how close are we really to seeing personal agents at scale? Today we're digging into those questions and a lot more, so let's get into it. I'm Rana El Kaliouby, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution. Hi Maithra, welcome to Pioneers of AI. I'm so excited to have you on the show.
So excited to be here, Rana.
So, first things first, we both went to Cambridge. I guess you did your undergraduate there in mathematics and Trinity College.
That's right.
How how is that experience?
How is that experience? Um, extremely intense is the is the truth of it. Um, Trinity attracts, uh, all these folks who have competed in these international math competitions, these math Olympiads, which are now in the news again because AI is now making progress on these these math competitions. So that's been an especially exciting thing for me, seeing parts of my past and parts of my present come together in this way. But my my story of getting into Trinity is just I, uh, was really into math around high school.
Did you grow up in the US? You grew up all around, right?
I grew up I grew up all around. The the bulk of my life was split between the the US. and, um, the the UK. So when I was in London, uh, I discovered, you know, these these math competitions pretty pretty early on, uh, got really excited about, uh, these Olympiads, the the reasoning and the complexity, the abstraction involved in them, the problem solving. And, um, if you compete internationally in an Olympiad, that just takes you straight to Trinity College, um, in Cambridge, uh, that just pulls everybody across Europe who's competed in these these Olympiads.
And some people go on to pursue, you know, mathematics research of various kinds, but of course, many people end up in London working in in financial services and especially, you know, in hedge funds, lots of different types of of investment firms.
So before you started Samaya AI, you were a researcher at Google Brain, and I am just kinda curious about your experience there. And also, what was it like to do research in industry versus research in academia?
You know, I started collaborating, working with folks at Google Brain all the way back in 2015, and to put that in perspective for where the the AI space is right now, this was before OpenAI even existed. So there was Google Brain, um, Facebook had just started the the Facebook AI research effort with with Yann LeCun joining them, and and that was that was basically it. O- one exciting piece about being in industry at that time, and maybe a difference to academia is even back then, you were seeing the formation of these slightly larger teams that were going after, uh, specific capabilities or specific infrastructure or specific foundations for AI.
That was maybe a little bit different to the to the academic approach. So so that's what really pulled me into Google Brain. I mean, there are all these people, and they have all of this, um, all of this secret know-how, sort of all these like intuitions about how to build and train some of these systems. And, you know, now that I mention it, maybe that's one thing that's actually stayed the same between back then and now, where working with this group of people and getting these insights that are not easily written down anywhere, but it's just lived experience is something that's sort of stayed through, um, the same in the the AI field.
I actually think people don't talk about that, right? And it's it's almost like a bit of an oxymoron that in this age of AI where we're thinking about automation and everything's kind of right, like productivity focus and automation focus, there is still a bit of intuition. And how do you train these models? Talk about that a little bit more, actually.
Yeah, I think people are vastly underestimating how much of the human perspectives and human inputs are required to to really enable our glorious AI future, you know? Um, and and it's it it's gonna come up in all kinds of different ways. Uh, one thing that's been top of mind for me recently, as I've seen some of these AI agents develop, is that pretty soon we're going to have AI agents running through the night. Uh, you know, in some sense, nighttime might not be dead time in the same way that it is right now.
Um, and you can imagine that, you know, you as a human are going to do a handoff process with some AI agents that are going to go do things for you and then come back to you the next morning being like, Hey, here's all of, like, the things I found, all of that. Actually, this this future that we're heading towards is is not that different to the, um, to the experience of being a machine learning researcher over, you know, even the past decade.
Because you'd as a machine learning researcher, you know, you wouldn't want to to leave those GPUs not being used overnight. So you'd be setting stuff up for your models to go train and then the next day you'd look at it, look at like the the results that are, um, you know, that that you're going to get. And and then now, you know, that's just gonna be more accessible to everybody. So we don't have to think about the models anymore.
People are gonna have these agents, and then these agents are gonna go do things for them, um, o- overnight. But coming coming back to your question to for those agents to do well, just like with the machine learning researcher, you had to make sure that that handoff had to be really, really good. You had to go in there, like, get all those details right, make sure that the model was well set to train. Similarly, people are going to have to make sure their agents are well enabled to succeed, and that's something that we're really, really underestimating.
So, in a world of AI agents, I I don't think you have that that productivity, that that, you know, positive impact unless you also have people guiding them on on how to succeed.
Yeah, I remember those days. A, you don't want to, like, run the training kind of y- y- you know, the the training iteration while during the day, cause then you're just sitting kind of idle, right? So overnight runs were were super smart. But to your point about setting it up for success, you don't want to, like, go home, you know, be asleep and then, you know, an hour in, the code runs into something and it just stalls or stops, right?
yes, exactly. It's gonna be similar for the agents if they have this amount of time. Um, yes, we're going to see the AIs get better at decision making, better get better at, you know, planning, reasoning, all of those things that we're seeing. But putting that within the right framework, the the right context for that decision making, that's still going to be heavily driven by by humans. And I think actually, as the AIs get better, p- paradoxically, and the thing no one's talking about is that we'll need more and more humans, more and more people in the loop, so that we can make sure that they all have the right frameworks for judgment and all of that.
There's going to be an element of that that's not easily scalable, and there is going to be a heavy human touch in those pieces. And, um, yeah, that's that that's really like the future that I see us I see us heading to.
Hold that thought because I want to come back to it. I think it's a very important one and a very relevant one, and I think it's top of mind for a lot of people. But you left Google Brain to co-found Samaya AI, and as I think of my own experience, kind of leaving research and starting a company, it's kind of a big mental and emotional shift to move from research to entrepreneurship, to leave big tech, go start from scratch.
What was that transition like for you, and also what was the impetus for the transition?
Yeah, so I I always say the motivation for doing a company is really, really important. I think it has to be quite pure in some ways. I think you have to leave not caring too much about, you know, the exact outcome or where it goes, but driven by this desire to bring something to the world and hopefully be driven by that positive impact that you see in bringing that that thing to life. So, so that was definitely the driving force for me.
You know, I'd had ten years as an AI researcher at that point. I had this beautiful body of of work. I'd I'd, you know, been fortunate to collaborate with num of, um, number of the the leading, um, figures in the field, some of whom, uh, went on to become some of Samaya's first angel investors. And, um, it was a it was a very satisfying period. And I think the thing that really drove me was exactly that desire to to go zero to one to bring something to life that that didn't exist yet.
And what I saw is over this past decade of AI research, AI had not really been ready to to be used in the real world the way it is today. And people had tried in various ways, but it was either too brittle or it was, um, you had to make it very, very focused on one specific thing, or it was better within, you know, a larger ecosystem of some kind. It wasn't in a place where it could really stand on its own.
And, um, finally, with the, you know, emergence of large large language models, which which we saw early. So, I think the world saw it in late 2022 when ChatGPT came out and everyone started paying attention. But sitting where I was in Google Brain, you started seeing the pieces come together in 2019, 2020 or so. I think from that point on, um, I could see the potential of finally having this AI that was general purpose enough, that was generalizable enough, that that you could put it out into the real world and have it drive impact for for people.
Very cool. And then what what's behind the name?
Oh, so Samaya actually means time or or moment in time. So I I really like the name. It, um, has a history in in, um, Sanskrit, which is ancient language in in India, and I'm I'm Indian. And, um, the, I mean, most of all, the the goal with starting Samaya, that that impact that we wanted to see in the world is giving people back that their time and then, you know, driving them to that moment in time, which is which is sort of that moment of insight as they go about their their their day-to-day, um, which is that which is, in our case, um, you know, a lot of a lot of these investment decision-making use cases.
In a minute, why ChatGPT won't cut it when it comes to making big investment decisions. Plus, we go behind the scenes of Samaya and see what it takes to build an AI system that can model real-world messiness. The companies winning right now aren't just building great AI, they're building the infrastructure to actually run a business on top of it, which is where Stripe comes in. Stripe is built for whatever comes next, with planetary scale financial infrastructure to help you power any business model.
As a matter of fact, 86% of the Forbes AI fifty run on Stripe, along with industry leaders like Salesforce and Pepsi. Last year, Stripe users processed $1.9 1 trillion combined. See how Stripe can power what's next for your business at stripe dot com. So Samaya has garnered a lot of attention, so I'll just name some of your awesome investors. Um, that includes former Google CEO Eric Schmidt, Jan LeCun, Mark Cuban, who we've had on the show.
Um, you also announced recently a new investment from N Ventures, that's Nvidia's, uh, venture arm, as well as Data Bricks Ventures. So, yeah, congratulations, that's a huge accomplishment. Um, how did that all come about?
Okay, wow. So, um, the unifying theme between the investors was our clear North Star of s like, you know, seeing both these AI capabilities, but then, you know, our our, um, our vision on how we would take that and really translate that into meaningful value in the investment decision-making space. How that influenced, like, what we wanted to focus on in terms of, um, our our, um, AI work and, you know, the use cases we were able to support, uh, was something that I think that was very inspiring for for investors.
So, from the inception of the company, we saw both the, you know, the development of these fundamental AI capabilities, but we also saw the work it would take to take some of those capabilities and translate those into to practice. You know, you see all these benchmarks, but these benchmarks are often quite stylized evaluation settings, and, um, progress on benchmarks does not translate to to progress in the real world.
Can you give us an example? Cause I I think again, the average person is looking at how AI is passing all these benchmarks and not realizing that oftentimes it does not translate to the messy real world.
Let me give you an example just from Samaya. So, Samaya, we work on, you know, investment decision making. Um, and, you know, there are some benchmarks out there to try and test out the the capabilities of AI in the financial domain. And what these benchmarks look like is the following. There's a very specific question that has, um, a very specific numerical answer, and you can look at that question and the numerical answer the AI gives you and decide whether it's right or wrong.
That's that's the benchmark. That is not at all representative representative of the real world. So in the real world, what it looks like is somebody is asking a very complicated, ill-formed question. Um, there are many variations for what right would look like. Some things are more right than others, and so it's not about just a like a, you know, a specific number, yes or no. Uh, you so what we did, and this is a project that we call Criteria eval, is we actually created an evaluation rubric that you would use to evaluate different types of answers and then grade on sort of how many different things did they did they touch in the the rubric.
So that's that's one example of the the the messiness, um, seeping in. And and and there are many, many more examples examples like that.
Can you give us an example of what this looks like at one of your, you know, customers or users?
Take the financial market. Um, it's a market because people have opposite views of what might happen. Now, imagine two users with something like ChatGPT, and two users give ChatGPT, you know, a relatively similar prompt. Then ChatGPT is gonna give these two users a, um, you know, a relatively similar response. Um, but what we need to do is we need to be able to personalize the AI so much that they're able to give opposite responses to different people based on the context and frameworks and, um, you know, views that the people are providing to the AIs.
Like, what's an example of an instruction? Are you asking the agent to go do research on a specific investment, or are you asking it to actually go execute on the in like, like, how agentic? What what does agentic mean in this context?
I'd say it's adjacent to the to the decision making. It's not, it's not doing the end, like, execution all by its, um, its itself just yet. And I think having, um, people in the loop is still very important. So, I mean, one example is on, uh, maybe evaluating a thesis. So, um, you know, one of our, um, users, an investor might come in, they might have a particular thesis, a view of the world, um, um, of of the market.
Um, that thesis might also be tied into, um, you know, specific, like, firm-wide context frameworks, as well as their exact portfolio. And so they might want the agent to go do research on their thesis, understand the implications for their portfolio, be able to, you know, cross-reference that with, like, the broader frameworks, firm-wide context, and, you know, come back to them with, um, you know, kinda key takeaways, like things that, you know, they should change a position on, um, potentially, like, new opportunities that they could they could pursue that's in line with everything, um, everything they've shared.
Yeah, you know, the analogy kind of I'm I'm creating in my mind is in the same way that vibe coding democratized access to building apps and websites and whatnot, are you in a way democratizing access to financial information? So, uh, y- y- you know, or or databases or manipulating and visualizing data?
Um, so yeah, so vibe coding, I'd say it it didn't democratize access, um, access to code bases, it democratized access to coding itself. And so, um, you could say that. So similarly, I wouldn't say Samaya democratizes access to information necessarily. I mean, we certainly make it easier, we put it all together. So so, like within within a firm's context, putting it all together. But I think some of that reasoning and analysis on that information does become more accessible.
And and um, th- this has also been like a a kind of a guiding piece for us, which is we really wanna see AI that up levels humans, that, like, lets us do things that, you know, that that are innovative. So it's not just, you know, just like, oh, this is like a productivity game, but you can approach things completely differently. Um, and so that's what we're seeing a little bit too. So now that people can put all of these, like, um, content sources, this data in that wasn't even possible before.
Maybe you're putting in some structured data with, like, broader views of the market, plus, like, very specialized, like, views, then then that becomes something innovative, that becomes something you couldn't do before.
Yeah, yeah, I I I love that point of view. Now, one of the considerations when you're making investments is also, like, the geopolitics of the world, right? So how do you incorporate this into your models?
That's actually one place where I think Samaya has been especially valuable to all of our our users and our clients. Because, like I said before, when when you look at financial services, often you have this choice between zooming out and zooming in, and and but that's still not perfect for the decision making. So it's a it's a mechanism that we have s as humans have put together just because we need some way to to navigate, um, what we're, what what we have to do.
Um, and then Samaya really lets you, like, you know, connect that thread between the zoomed out and zoomed in views, and that's so, so important, especially in a a a situation where there's so much, you know, geopolitical change. There's there's also this wave of AI and how how disruptive it's being. And there are those, like, there are these, like, broad themes going on, and they're affecting everything at the, um, at the very zoomed in level too.
And you need to be able to understand the connection between both of those. One thing we've been working on internally, so it's a it's a project we have called, um, causal world models. Um, we posted some things about it in research preview, is being able to get the AI to to do some of this cause and effect reasoning, taking people from macro to micro, uh, to trace through, you know, millions of sources, but do that in, like, in a way that's very sensitive to the the cause and effect pieces.
Do that in a fully attributable way and give people that ability to connect the the macro and the micro.
So can you give an example of, say, I'm a hedge fund manager and I'm trying to make an investment decision. What kind of questions can I ask Samaya's AI, and and how does that causal world model come into play?
You know, there are so many open questions right now on on AI and software, for example. So you might ask this broad question, you might say, Okay, we have this huge disruptive wave of AI, we have the SaaS apocalypse on the other side. Like, help me work through and understand how, you know, where are the places like that that, you know, we we see a huge disruption, a huge change, and where are places where things are just much more reactionary in terms of the market.
And where are maybe new categories that are, you know, coming that are coming up that we should sort of pay attention to? That's an incredibly hard question because what you're doing is you're taking this, like, high-level, like, theme on AI, and then you have to translate it down to, like, the very specific entities that are being impacted in this and then kinda categorize those in different ways. And do that cause and effect reasoning because you need to be able to show your work.
When you when you come back, you need to be able to say, this is the prediction, but this is also why. And here's my chain of reasoning, my cause and effect reasoning that connects me all the way back up to that that macro theme.
So one of the, um, it sounds like one of y- y- your approach is very focused on kind of these smaller models as opposed to a general purpose model that can do everything. And and you've built these models specifically for the financial services sector. Can you talk about that decision? Like, why not use a general purpose model?
Yeah, absolutely. So, firstly, we're all about systems over models. If you look at every AI advance that's happened that's out there in the real world, it's always systems of of some kind. Um, I always like to to bring up self-driving cars as a as a meaningful reference point. So, you know, computer vision predates large language models, self-driving started even earlier. Now we have them out in the real world. Actually took a a Waymo in today, and, uh, and but that's a that's a system, that's not a single model by itself, so it's all these components.
So always systems over models. And then as part of that system, when you're when you're trying to bring, you know, go from the library to the office, you're trying to embed in people's day-to-day work. There are just all of these other AI components that need to be built out to do that translation very effectively. Um, and so that's where some of these smaller language models that we've trained specifically for the domain, specifically for various different types of tasks, for certain types of accuracy and, like, precision issues that we see, um, these larger models trip up again and again that are very domain specific.
It's very powerful to have all of that.
Yeah, I love the approach of thinking about it as a system of technologies and kind of components that all need to work together effectively and productively. If you're listening to all of this and thinking, Wow, Samaya's AI system does a lot of the same things as a junior financial analyst. Well, you're not wrong. But after a break, Maithra shares why she thinks AI is actually good for job growth. So I wanna zoom out a bit and talk about what does this mean for jobs.
I'll make it quite personal in in in two ways. One is we've been using this chief of staff AI agent, right? And we have a couple of junior analysts on our team. And I'm already noticing how this is causing us to kind of rethink, like, what our junior team members ought to be focused on, right? And then the second kind of personal anecdote is my son is a junior in high school and he is quite interested, or he's exploring potentially, like, studying economics slash finance for college.
But I don't know, will will these jobs exist anymore? Will it will it look very different? So I am curious about your thoughts on how this is changing the jobs landscape.
Right, I'll share some maybe bad news and good news in in that order. Um, I I think there's more good news than people talk about, so I'd like to spend a bit more time on that. You know, a bit of bad news is it it's true this is going to be disruptive. You know, it is going to have us rethink a lot of like, b- both some roles are going to be transformed quite substantially.
Um, we're gonna have to rethink, like, what we do. Some tasks may end up fully automated, like all of that. So there is a wave of disruption that's that's coming towards us and and, you know, we have to we have to acknowledge that. Um, but maybe the the good news, again, that I see fewer people thinking about is, a, I deeply believe that, especially as these agents get more capable, you are going to have to have more humans working with them just to deal with, like, how much output they're gonna be producing.
Like, the AI doesn't need to sleep, the AI is constantly gonna be working. That's great, but then I think you need to have humans in the loop in places again to put in some perspectives of, like, either those, like, frameworks, like the broader, um, you know, universe in which, like, whatever decision making is happening, all of that, like, the long horizon agents, like, running overnight, that kind of thing, like, who is going to be, like, sh- like, shepherding them, you know, like, so so there's a whole whole set of, like, jobs around these, like, AI agent shepherds, maybe that we don't we don't even, um, see yet.
And there's there's gonna be a lot of that, uh, to to come. Uh, and then secondly, I'll say that sometimes when you have this wave of tech technology, um, disruption, it's y- you see the places where it's just, like, you know, clear productivity gains or clear cost optimizations, much earlier than you see the things that are deeply innovative and are going to, like, uplevel all of us, you know, make things that we wouldn't have even done before, like suddenly possible.
Do you have favorite examples?
I'll take one from the investment landscape. So financial services is a huge industry, of course. And within that industry, Samaya is laser focused on investment decision making, which we think is at the heart of the the industry. Um, that the company's mission is taking people from information to conviction. So super, super focused on that decision making piece, but partially focused on that because, um, w- we think that that's like a place where you can drive innovation, where you can really kinda enable things that, um, that weren't possible before.
So, so maybe super simple example, publics and like privates. So, like, you know, there's a whole ecosystem around, like, p- um, um, public companies. Um, there's an ecosystem around private companies. Those ecosystems have changed and, um, you know, um, transformed in various ways. They now influence each other way more. Oftentimes, they were, like, you know, different strategies that were used for both of these. Like, is there a way to, like, bring some of these together and in that prospect, like, create new opportunities for in for people to invest in, like, mixes of these?
And that's like a simple one. It would be very meaningful. It would be meaningful for maybe more people to have, like, access on the private side, for professional investors to gain confidence that they have, like, a really good understanding of both of these, like, areas and, like, the factors that influence them. And, you know, new types of, like, investment mechanisms, products, all of that suddenly becoming, um, available to people, um, that could sort of touch on on both of these.
Yeah, very cool. You also talk about personal agents. When do you think those will become mainstream?
I think this year, I think this year is gonna be the year of real personal, um, of of real personal agents. Um, you know, maybe some things happening on the model side, some stuff where you're seeing, like, more involved, like, translation, especially probably on, um, some of these enterprise use cases. But, um, if I if I trace back actually the history of, like, agents to put it in context a little bit, so li- like, if the the term agent became very popular in 2024 or so, I don't think we had real agents until about late last year.
So 2024 through 2025, what we had were more workflows, not agents. And the difference between a workflow and an agent is a workflow is very hard coded. So it might be multiple steps, but it's a very specific set of steps. There isn't actually any agency that's that's, um, that's happening in that that execution process. Um, whereas, like, late last year, you know, I think Claude Code was like an early beginning of this.
And then since then, I think the capabilities of these systems have also developed. Like, now we have a like, you know, a little bit more of real agency. Like, you give the, um, you know, you give the AI, like, the the relevant information, the relevant inputs, and it is able to go away and, um, you know, make decisions basically as it goes through that execution loop.
It knows the desired end result and it can backtrack what it needs to do, right?
Exactly, it's it's real, it's real agency for the for the first time.
If you could, n- wave a magic wand and have a personal agent that can do anything for you, what would you want it to do?
My goodness, uh, prob- probably, um, probably clone myself, you know, be me in in all of these places.
Oh, I love that.
You know, just come in as me with, like, and and be able to assess the situation the way that I would do. And then and then be able to, you know, take the appropriate actions or or produce the appropriate outputs.
Yeah. Have you experimented with creating a digital twin of yourself? I've tried, like, kind of doing a video digital twin, but also feeding a model with all of my blog posts and my book and my interviews. It's not there yet at all. I would not send my digital twin to speak on my behalf anywhere. I don't trust it at all. But I'm curious, have you have experimented with some of these?
I I have done some experimentations, and I have, I have, yeah, I have some. So so, I mean, like, pr- probably closer to what you've done. Like, I've tried to give the, you know, I've connected it up to, like, things I've written. I like audio a lot, so I like speaking to it because I think if you can really speak to it instead of typing, you can be much more descriptive. You can give it, again, more information of where you're coming from.
So that's been, that's been really powerful. But but I've noticed two things. Like, if the if the thing I'm trying to do is, like, scoped enough and it has very good inputs, like, I kind of tell it, like, th- this is, you know, this is, um, I don't know, something I'm trying to put together. Like, these are, like, a few different draft versions. Here's what I think is missing, like, help me put this together. Like, if I give it very specific inputs, like, it can do a reasonable job.
But what I find when I don't give it as precise inputs and I and I try to get it to produce an output, it produces this output. That output is not like me, and I I need to correct it, like, in a bunch of ways. So what I've learned is, yeah, very scoped things, it's fine, but something more open-ended, you're you're gonna, you can't hill climb from it.
Okay, so I'd like to do a quick rapid fire. I'm gonna throw out an assumption about AI, and you tell me if it's a, uh, myth or reality. Humans in the loop slow down progress.
Myth. The feedback is really helpful. How's the AI gonna get better if if humans can't can't give it any feedback or steer it or correct it in any way?
That's great. I think I agree with that one. Large models always win.
For some things, um, m myth, I would say myth, because, um, like, I think I think some things are they're very good at, so I I don't wanna take that away, but but, um, you know, o- other places that we we've sort of lived and seen as we've built out some of these agents and stuff, like, I think the smaller focused models can really, really drive value.
Chat interfaces are dated.
Uh, reality, I think that's coming.
Interesting, cool. And then for our younger audiences, um, economics slash, uh, finance majors are obsolete.
Myth. There's a there's a lot of human, like, d like, yeah, just human judgment, human taste, that'll be a key piece of that that field still in the future.
That's awesome. Okay, last question, um, what does it mean to thrive in the age of AI?
Energy and experimentation. We're in a we're in a, like, a, we're in a, uh, s uh, you know, time of of disruptive change, that's true. So let's let's face that with energy. Let's, you know, face that with also experimentation, like, go out there, sort of see what's possible and, uh, you know, see see see what we can create, see what's out there. And I I think that will 100%, like, like, like, leave behind some of those, like, old inhibitions, all of that, like, embrace the new, embrace it with, um, energy, don't be afraid to experiment, and I think it'll lead you to exciting things.
I love that. That is awesome. What a great way to end our conversation. Maithra, thank you for joining us on the show. This was great.
Oh, thank you so much, Rana. This was fantastic.
It was great talking to Maithra about how to build an AI system that can make sense of our messy world. An AI agent for the financial industry is incredibly complex. Think about all of the contextual information that Samaya's agents need to digest and then pull together to make a recommendation. But at the end of the day, there are judgment calls that need to be made, and this is where humans will stay in the loop. I've said it before on the show, AI is disrupting the labor market, but it won't destroy it.
Yes, some jobs will become obsolete because they will be automated, but there will be a whole new class of jobs to manage AI output, and we're already seeing those kinds of jobs grow. That's it for this week. Thank you for joining us. We'll be back in your feeds with a new episode next week. Pioneers of AI is a WaitWhat original production. Our executive producer is Eve Trow. Our producer is Rachel Ishikawa. Our senior talent executive is Stephanie Stern.
Mixing and mastering by Brian Pew. Video editing by Eric Purcell. Original music by Ryan Holliday. Our head of podcasts is Lital Malad. You can join the conversation across social media platforms. Just look for us at Pioneers of AI. Thanks so much for listening.