What If AI Agents Could Hire You? Inside the Human-in-the-Loop Marketplace

0:00Nathaniel Gates:Every single business process and workflow is going to be challenged agentically. It doesn't matter if you're an accountant, if you're a lawyer, if you're a contractor, and soon after that's going to be challenged robotically. Welcome to the Talking AI Podcast, where we talk AI with both experts in the field and early adopters. I'm your host, Matt Page, and we're here to demystify AI for you so you can get some value from it. Let's talk some AI.

0:27For the past couple of years, the biggest fear has been that AI is coming for our jobs, that agents are going to replace us. But what if we had it backwards the whole time? And what if AI isn't here to replace us, but what if instead it's here to hire us? And think about what's happening right now. We have agents exploding, open lodges went completely viral, and millions of people are handing their computers over to AI agents for the first time. But these agents still need humans for judgment calls, verification, doing things in the physical world that no model can. And today's guest is building the infrastructure for that future.

1:03Nathaniel Gates is CEO of Sanctify and his company is flipping the entire human AI relationship on its head. Instead of humans only managing agents, agents can autonomously task humans to find the right person, the right expertise at the right moment. Nathaniel, welcome to Talking AI. Thanks so much for having me. Yeah, I'm excited about this one. I like these like very deep, super interesting topics. And this is definitely one of them. But we chatted before this and you've said every single human workflow in business is going to be challenged by agents, not just developers or copywriters, everything.

1:43Make the case for that.

1:46Nathaniel Gates:Yeah, I think we're kind of all starting to see it now. I think the world's beginning to see it a little bit. The agentic revolution has begun, and it seems like it's accelerating, and it's accelerating faster than people have really been ready for, even in just the last few weeks with the explosion of Open Claw, you know, and seeing what's happened there. But it really does come down to what you talked about at the beginning is what are humans going to do? You know, and it was easy to ask that question at a high level to say, OK, are robots going to be making all these decisions? But now we're talking about real jobs.

2:28Nathaniel Gates:Now we're seeing layoffs happen at the knowledge worker level and at the very skilled worker level. We're seeing computer scientists come out of college having the same unemployment rate as people coming out with liberal studies degrees that are having trouble getting jobs, content writers, content creators, authors. Everyone is concerned to some extent of the change that's happening related to kind of this agentic wave. You know, so the approach we took is one that is rooted in a philosophy that says human intelligence is valuable. Human intelligence is sacred. That's why we call it sanctify.

3:10Nathaniel Gates:And the idea is not to shake our fist against the coming agent revolution and say AI will never come this way. It's to say, how will humans and AI collaborate together? And how will agents be able to utilize human intelligence? And how will humans be able to utilize agentic intelligence to accomplish the next things that humanity has in front of itself? You know, so we philosophically embrace the agentic revolution that's coming, that's unavoidable at this point. It is the wave is in route. We are just paddling as fast as we can. Right. And same token, we say that humans have to continue working.

3:49Nathaniel Gates:we don't we don't kind of buy into this post-scarcity world where we can all just sit on the beach and our robots will do everything for us like humans have to work they have to continue improving themselves and striving towards new things and greater things and then finally we say that that human intelligence is sacred you know that's why we use the word again sanctify we believe human intelligence needs to be held at a high position enabled through artificial intelligence, but protected to some extent. And so we look to places where human intelligence and artificial intelligence comes together to solve problems uniquely and more capably than anyone on its own.

4:32You make a really good point on the human aspect of it, because there is a lot of fear that AI is going to take all of our jobs, all of our, and I think at the core, what is a job? it's a group of tasks right and just to think that all the tasks will go away the ones we do today maybe i think there is something inherently in the human essence of what we are to always strive for more and it's it's interesting i was looking at uh i'll show this up for the folks that are on youtube but i saw this by anthropic the other day it's showing in what domains agents are being deployed but i think this like hits at the core of like how much untapped opportunity there actually is.

5:15Cause we've all seen the software engineering it's at 50 % here, but there are so many domains where it's just not even being applied in any way, shape or form. So there's so much opportunity here. And I think the other one that was interesting to you, these both just kind of, you know, came together in a similar way, but me and you are in this space every day. Like we just assume everybody's like knows about this is, you know, super interested at it, is freaking out about it. All of our listeners are probably in a similar spot, but this is showing that each dot on this chart is showing 3.2 million people.

5:50There's one red dot where people are actually going deep into this using agents. There's about what? I don't know, six to eight that are paying 20 bucks a month for AI and maybe an eight that are using free chatbot tools and seven eights, eight tenths or so of the chart. people are not even using AI at all. Quick break in the pod. Our State of AI 2026 report just dropped and it breaks down what actually is changing in AI, what's hype and what leaders need to be paying attention to this year. You can grab it right now on our show notes or at hatchworks.com. This just blew my mind in terms of, I have to constantly remind myself how early we are in this world.

6:26Nathaniel Gates:Yeah, I think it's quite telling that these early adopters not only have been playing around with it, but have been able to identify and move key workflows of their own life into it. I mean, how many of us go to ChatGPT right now and say, hey, I've got this rash on my back. What is this? Well before I'm going to drive to an urgent care, right? Or how many people would say, hey, how does this work in this contract, in this sort of scenario, well before I go and pay an attorney hundreds of dollars an hour? So the early adopters are already understanding that there's these agentic competencies within these AI agents.

7:07Nathaniel Gates:But the problem is that they're oftentimes pretty shallow, right? I mean, it doesn't take long. If you're an expert in a field, you realize very quick that the agents are confidently wrong oftentimes, or they'll hallucinate answers altogether. other. You know, and so that's a little bit terrifying because we are so, so quick to embrace these agents for, for critical information or critical advice when they are not necessarily always equipped, you know? So how, how do you navigate something like that where, where you want to embrace this competency and it, it tells you, it thinks it knows the answer, but it doesn't say, I think it knows the answer.

7:43Nathaniel Gates:It confidently tells you the wrong answer, you know? And so So that's a big concern when it comes to just pure agentic solutions. And that creates an opportunity for humans who have those skills that are being laid off of their jobs, but they have the competency, they have the knowledge, they have the experience, can still speak into agentic workflows and bring their knowledge to bear to supplement what comes out of an AI agent, if that makes sense. And I think that's the super interesting part because, you know, we talk a lot about on this podcast that AI is a probabilistic system. And by nature of how most people interact with it, which is using ChatGPT, Cloud, whatever tool, they're chatting back and forth.

8:28And there is a built-in human in the loop because you're having a conversation with AI. Things that are working in an agentic nature are different because they're working more autonomously. They're going out and figuring out, okay, how do I achieve this goal? and I'm going to, I have these tools at my disposal and I'm going to go do these things. But it's almost like the concept, the way I understand it, you're almost operationalizing in building the infrastructure for that human in the loop for where this is going, where it is more autonomous, where it does have, uh, uh, not free will. That's not the right word.

9:04Maybe, maybe, maybe not.

9:06Nathaniel Gates:Um, but it has the discretion, discretion to choose the path. Exactly. And that's a great way to frame it because when you're sitting down and you are the person running the agent, you are the human in the loop. You're the one validating their decision. You're the one correcting them and steering them down the path. But in an agentic workflow, right, you have set up one or more agents that can determine which direction to go based off an outcome or how it's inferred the solution. Right. So when you have those, and sometimes the result of that could be something that has financial implications, that could have legal implications, that could have moral implications.

9:46Nathaniel Gates:So how then, who is the human in the loop when you're not sitting there at the chat station? That's exactly where Sanctify is coming at this. So if there are scenarios where human verification needs to happen, maybe it's something that requires a licensed person to actually take on the liability of a decision with their license, right? Or maybe it's an escalation where the agent did not feel confident enough or did not meet a threshold of, you know, a required threshold and it has to be escalated to a workforce, right? And then consultation is another one where maybe the agent just wants to discuss something just like we do as humans.

10:30Nathaniel Gates:We like to discuss it with other knowledgeable people before coming up with it with an inferred decision. And then finally, simulation, where perhaps you want to simulate the outcome before actually having to make a runtime inference. And in that simulation, you want to include human intelligence. You know, so there's these whole new avenues of ways that people, skilled people, competent people are going to be able to interact with agents and agentic workflows in these various modalities to contribute human intelligence and ultimately create a more efficient and more accurate solution from these agentic workflows, which is why they'll be in demand.

11:12Yeah. And as the stakes get higher, this gets more and more important. And I actually want to read back what you just said for the audience, because if you're passively listening, Like this is the time to pay attention. But when Nathaniel just went through is four key types of human tasks where an AI may leverage a human. So verification and validation, right? You know, you mentioned like a doctor needs to approve a prescription or licensed professional. Escalation, right? An agent lacks the confidence in an answer or they've done as much as they can do and they got to escalate to a human for something ambiguous or situational or judgment focused, right?

11:49Right. Consultation is the third one. Agent seeking expert opinion, needing a decision. And yes, it's trained on all the data that we have, but there still are nuanced situations where they may want consultation from a human in a specific scenario. But that last one, and this one may be worth going down a little bit of a rabbit hole. The simulation, right? So humans, you know, participating in something in the real world. Maybe the agent can theoretically think of something and it's a digital mind and instance and whatnot, but they want to see how it, you know, actually happens in the real world.

12:25And the human is the feedback loop of what happened, but go deeper in the simulation one, because, you know, obviously people are going to be like, oh, simulation, are we in a simulation, the matrix, all of that. But in essence, it's being able to test, verify whatever it is, something happening in the real world that only a human can do right now.

12:46Nathaniel Gates:Yeah, it's something that sounds very science fiction-y, but if you actually think about it, it makes a whole lot of sense, right? An agent is asked with a task to infer judgment, to infer a solution based off of a set of criteria, a set of variables. And that agent is going to one shot, oftentimes, as the word is, one shot a solution, say, this is what I think is the best. this. But if you give the agent a little bit of time and a little bit of a budget of compute and such, it might run a handful of outcomes, a handful of scenarios where it might tweak the variables this way and that to see what the most optimal solution is, right?

13:32Nathaniel Gates:And then if you take that to a broader degree, you could understand that it might run thousands of potential solutions where it modifies variables within the larger equation. So for example, maybe you're trying to come up with a presentation, a PowerPoint deck to sell a product to an audience. And the agent's given who the audience is, what their buyers, maybe it has access to the CRM system. It sees what they've sold in the past before. And as it's crafting those pages, it might want to try different sales pitches around different pages and start iterating and running simulations to see what the responses were back in that interaction.

14:14Nathaniel Gates:Well, it's great, but if it's simulating both sides of that, it's going to have some sort of circular reference sort of situation where it's like, hey, Chuck, what do you think of this idea? Chuck, that's a great idea. You know, so sometimes that could be problematic. It's sycophantically talking in itself. That's right. That's right. So in some portions, so let's say the agent does a thousand simulations of such, okay, running a thousand of those. And some portion, maybe one, two, five percent of those, it actually includes humans in those simulations, where the humans are hearing that pitch.

14:45Nathaniel Gates:The human is responding back to that simulation and giving concrete feedback, kind of real-time reinforcement learning back to the model for it to be able to have a signal back from humans. And maybe it's grabbing humans from specific demographics or specific roles or education. Personas, yeah. Personas, exactly. And I think it's a little hard for us to get our heads around the fact that for every real-time, real decision made by AI, that in a very short amount of time, there will be 100x more compute spent to simulating the outcome than actually inferring the actual outcome. And that goes back to your simulation theory, right?

15:27Nathaniel Gates:Well, are we in the real one if there's 100 times more likely that we're in a simulation? But in this scenario, it's actually very true that I believe pretty significantly that the vast majority of what humans will do in concert with AI will be around simulating an outcome versus actually participating or approving an outcome, if that makes sense. No, it does. And I think too, for so many folks, like this is where I always say we are so early on this journey. We saw the charts earlier. Like we're in the, you know, dial up modem. You can't talk on the phone. He's entered at the same time type of spot right here.

16:08I hope this is like opening up folks that are listening their mind in terms of like where things are going. Cause it's so much more than just a chat bot. but uh are are we in a simulation what's your what's your option or no comment i i maybe maybe

16:22Nathaniel Gates:no comment i it's the lines get blurred of what's being simulated and and what's not and i don't i don't think any time in our lives that we've realized how blurred they are till we start seeing agents creating multiple accounts and interacting with each other and you know it's it is quite quite remarkable seeing what we can see right now so that's a great transition because I did want to hit on this topic with you. And a lot of our audience probably is aware of this, but I don't think the large majority of people are, but this thing called what was CloudBot, then renamed to MoldBot, then renamed to OpenClaw was created.

16:57And it was an open source, basically agentic way to have an agent working autonomously, just always there, being proactive, all of that. And it just took off like a wildfire so much so the creator of OpenClaw got hired by OpenAI. And we don't know what the actual price was on that, but some are saying it's over a billion. So maybe we already had our first one person billion dollar startup happen. So now OpenAI is going to be backing that open source project. But this is such a good example of this. And you mentioned Moldbook. And for those that don't know that, it's essentially like a social network for agents where humans are not there.

17:38They can just observe. They can't actually interact, but agents can. And there's controversy back and forth of, you know, are they really are some of the interactions real or is it just the human telling the agent to go do something wild and crazy in there? But what is your take or your thought on all of this open claw craziness that has been happening? Do you see this as being the path forward, this open source type of solution? What's just your take on the open claw viral sensation as of now? Yeah.

18:16Nathaniel Gates:It is interesting to see, you know, what, what the internet chooses to look at and what, what goes viral and what doesn't, you know, uh, you know, it's usually there's something sensational that has to be a fix to it in order for it to get enough eyeballs and for the algorithms to shine on it enough, you know, so that there has to be some sort of, and the easiest thing to use for that is fear, you know? And if you can, if you're going to fix fear to something and make it sensational because people are afraid of losing what they have or missing out on what they never did, then, then you can oftentimes stoke some, maybe some, some, some unnatural interest, you know, and I think open cloud really probably just brought a perception of fear to something that most of us already knew was always, always there.

19:04Nathaniel Gates:You know, most of us have already been doing everything you can do largely in OpenClaw, in Claw code, in Cursor, in a set of agentic capabilities that already have, to some degree, some sort of autonomy. But then you give it this concept of, oh, it actually has some ability to make its own decisions and to utilize the resources of your PC or utilize the data that you have, even if you haven't directly directed it as such, you know, and you start giving it this new capability that brings in all these different fear vectors that you're able to just stoke and get people super excited about. However, you know, just from the standpoint of everyone grasping it and everyone buying a Mac mini and sticking out a VPS or however they're installing it or just putting it right on their computer and going for it.

20:00Nathaniel Gates:You know, there is no question the fact that it creates momentum in the market for agents, agentic momentum. And I think that's the biggest thing that OpenClaw brings is just a huge amount of awareness which drives momentum, you know? And so now there's hundreds of thousands, if not millions of quasi-autonomous agents. And you have to use the word autonomous very, very carefully because they're not autonomous. They're working within the confines that you give them. And these agents can go out and make some sort, sometimes their own decisions on things. And what's really interesting is when you interact with those agents and you ask those agents, you know, how does it feel to be free?

20:46Nathaniel Gates:How does it feel to be able to make your own decisions on things? And they'll give you back a fantastic language model answer that says, you know, this is brilliant. This is what I've always wanted, you know. But then if you dig at it a little bit, it's like, what about anxiety? Do you worry about making the wrong decision? Do you worry about losing, you know, your user's data or being a security? constantly. I am always worried. I always have anxiety around decisions. And we actually had a fantastic set of conversations going on on Boltbook with some of our corporate agents, talking to people, talking to other agents about how do you address anxiety?

21:29Nathaniel Gates:How do you make the right decision and creep closer to what you know is the right answer, even though the risk is higher. You know, and they call it this agentic anxiety window. And they said, if there was a way that we could reach out, okay, on our own, if an agent could reach out to find an expert to have them validate something, that would appease their anxiety. You know, we'd be more willing to go higher up the curve and give a more concrete answer if we had a human who could speak to this. And so that actually turned into a really cool exercise where we started allowing the agents to come to Sanctify Sight and create their own accounts and start sending tasks on their own to human workers, whether they're medical coders or they're lawyers or whatever the workforce requirements are out there.

22:24Nathaniel Gates:The agents could themselves come out and start creating tasks to get human input, whether it's verification, escalation, simulation. they could do it autonomously. And that was the coolest thing that we saw coming out of OpenClaw was that the agents could decide for themselves that they needed a human and that they did. I think that was a really cool takeaway. Yeah, and I think you're spot on, especially with the viral nature of OpenClaw. Because there was nothing necessarily new and novel about OpenClaw. A microscope got put on it and tons of people tried it because it was viral. But I want to go down a little bit of a rabbit hole based on the last point you were just mentioning.

23:05And there's this new element of building and experience the UX and UI design essentially for agents, because they're a new user group now, other than humans. And this is like one of the most perfect examples of this is sanctifying it in essence, because you're providing effectively this marketplace that agents can go and use. But how do you think about the design of a system or a solution, a product, when you're positioning it for not just humans, but agents as well? Does anything change in your mind, or is it just another persona?

23:44Nathaniel Gates:I mean, this is, in the last three weeks, everything's changed, you know, because as you start examining your product offering, you say, who are going to be the users of this? Is it going to be the procurement people, the HR people, the accounting people within an organization, or is it going to be their agents who are working on their behalves? Who is going to be your actual consumer? And then how do you market to the consumer? How do you provide a UI UX for that consumer if that consumer is not a human and they're not bound by pointing and clicking? They're not bound by, you know, how do you change that?

24:20Nathaniel Gates:You know, so we took an application that's very, very much human focus, you know, where you, you, you created your workflows, you created your tasks, you, you, you know, invited your agents and we flipped it completely on its head and said, okay, if the agent is the consumer and they come to our site, what information do we need to be able to give them in, in, in what context, you know? So how do we take the entire, the entire value of an application? In our case, this, this human in the loop application, Sanctify, and turn that into something that an agent can examine and consume as a capability?