1. Library
  2. Podcasts
  3. Platform Builders
  4. Ep. #9, Manny Medina Explains: What the F Is an AI Agent?
Platform Builders
47 MIN

Ep. #9, Manny Medina Explains: What the F Is an AI Agent?

light mode
about the episode

In episode 9 of Platform Builders, Christine and Isaac sit down with Manny Medina, founder and CEO of Paid, to unpack the tectonic shift happening in software. From the rise of AI agents that autonomously perform tasks, to the death of seat-based pricing, and why ARR and SaaS metrics just don’t make sense anymore—Manny makes a compelling case that we’re entering a new era. If you're building, investing in, or working with AI-native companies, this is essential listening.

Manny Medina is the founder and CEO of Paid, a platform helping AI-native companies monetize through outcome-based pricing. He previously co-founded Outreach, the leading sales engagement platform. Manny brings deep experience in building, scaling, and rethinking how modern software should operate.

transcript

Christine Spang: Hey, everybody. Welcome to Platform Builders, the absolute best place to hear about the titans building the tools you use every day. I'm Christine Spang.

Isaac Nassimi: And I'm Isaac Nassimi.

Christine: And today, we're here with Manny Medina. Manny is the founder and CEO of Paid.

And I thought to kick things off, we should have Manny tell us what the fuck is an AI agent, because I'm not sure everybody really knows what they are and people have a lot of different definitions. So, Manny, what's an AI agent?

Manny Medina: So, an AI agent is a piece of code that uses foundational models or LLMs and runs without the help of a human in the loop, or at least as agentic as possible.

And when I mean agentic is that the software has agency to make decisions on its behalf. And it will involve the human seldomly to complete a task.

So, it excludes a lot of things like Copilots. It excludes a lot of things like SaaS software that you need a human being to go execute the tasks. The agents run the task for you.

So, the easiest test is to see if the agent is completing a task that replaces a human being completing the same task. So, at that point, and it can do it on its own. It can do it autonomously.

If it can do it autonomously, then it is an agent. If it needs a human being to intervene every single step, then it's not. That's the definition of it.

Isaac: Well, how is that different than, like, a Zapier flow or a Cron job? Those don't need humans.

Manny: That's true. But then you can say that Zapier is introducing agents into the same thing. So like, if there are forks in the road and the Zapier decides which fork to take, then it's more agentic than not.

So, I guess that's a good distinction in that the flow has to be long enough to complete a full task that a human would compose without the human intervention. So, would a Zapier flow be agentic?

Yeah, if it's long enough and completes the task. Then why not?

Isaac: Fair enough. So, you think that agents may have already existed in some format in very limited ways?

Manny: Of course. Even right now, the majority of the agents that we see out there are sort of like, you know, a little bit like glorified RPAs (Robotic Process Automation).

The new breed of agents that I'm seeing are way more non-deterministic, in which you don't know how they're going to execute the task, but you just give 'em the task and they will go figure out how to do it.

That has both incredible opportunities and some scary risks in terms of like, what are they going to do, how much tokens are they going to consume?

But they are slowly but surely superseding fully human tasks that requires some judgment. And as long as the agent continues to develop with judgment, they're going to continue to take over some of the stuff that we used to do.

Isaac: That makes a lot of sense. I think that when I talk to a lot of people about, like, the AI agent stuff, like, "Hey, we should probably try out using the software. We can do this thing that we do at the company with agents."

Sometimes you get some eye rolls. Sometimes you get some people saying, "Aah, I don't know." What do you say to that?

Manny: I think that in most every function, agents are early. The one that I get asked a lot of questions because of my lineage is AI SDRs.

And they will come to me like, oh, they all suck. Or like, you know, it's just spam cannons or they're just creating noise.

And in reality, no technology started bang on working. Like, everything took a few iterations until it got to a working condition. But the fundamentals for the existence of an AI SDR is still strong. Meaning, no AE likes to prospect. And as long as no AE likes to prospect, there will be AI SDRs and they will keep coming until somebody figures it out.

So, that's the beauty of it. It's that gravity only pulls in one direction. So, this AI SDRs will continue to develop great, you know, better judgment.

They will continue to do better higher relevance, and eventually one of them will work out.

Like, I was talking to somebody here in my office that is building agents for accountants. And I was telling them that like, it's not a question of if, it's a question of when.

And if all of them, if all the AI SDR providers just move to the O3 model for research and do a big research step before sending an email so that the email is hyper-researched, it's relevant, they will start killing it.

The problem is that the input tokens for the O3 model are incredibly expensive. So nobody could have, and like, the AI SDR would be more expensive than a human SDR.

You see what I mean? But it will actually do better work, but nobody will pay for it because everybody's looking for the cost advantage.

So yeah, no, it's not going to work until it works. But the direction of this is one way, is that there is no going back. There's no putting the toothpaste back in the tube.

Isaac: I mean, I believe that you are, I would consider the authority on the subject, right?

You've built, you build Outreach, which is the best I would seriously say, the best product for outreach out there as far as if you're an SDR or a BDR and you want to do your work.

I remember I was shown it, like, there were like these little things, like showing the local weather of the prospect. Like, just for personalization, things like that.

I was like, man, whoever built this product, who I now know this person, really understands the life of an SDR, right? That they just have to figure out something personalized to say.

So, if you say it eventually can become fully agentic, that's like, John Carmack saying that, you know, programming will be agentic soon to me.

Manny: Oh my god, you just made my day. I've never been compared to John Carmack. He is a genius, right?

He belongs to a different category of human beings. So, I appreciate you saying that.

Isaac: Okay, I'll retract half of the compliment.

Manny: No, no, no. It's already been said. You can't unsay it. No.

Isaac: Okay.

Christine: Next thing you know, we're going to hear about the Leet video game that Manny wrote last weekend.

Manny: Yes, yes. Yeah, he is way too famous.

Christine: Spit it out, Manny. What were you vibe coding?

Manny: Yeah, I wasn't vibe coding that. I was by coding some billing stuff.

We are in the middle of this agentic revolution. I feel like I'm the tallest of the children here because everyone is growing and growing so fast. I just happen to see a little bit ahead of everybody else. And so my job is to make sure that as my customers come in, building these AI agent companies, that I have a good answer how they're going to monetize.

And it's been the Wild West for a while as to exactly, you know, what is outcome-based pricing, how do you price workflows, what are the parameters to think about it?

And it's finally becoming kind of clear that there, you know, the models that you can apply. There's not that many of them.

And as long as you apply one of those models to your customer base, and as long as you don't get too married to a particular way of pricing because your customer base may change.

But it's going to be, you know, everyone is going to do good work. So, my job now is sort of to take this knowledge and take it out to market.

It's a little bit like, you know, it maybe too much of an overstatement, but it's a little bit like bringing vaccines into a country that needs it.

Like, your job is not to convince people to take it and they may not want to take 'em because they don't know you. But you know that it's going to be better for them.

And so, as a doctor, you have the oath to make sure that you save as many as you can. So, that's what I'm trying to do. I'm trying to save as many AI agent companies as I can.

Christine: So, the traditional way of pricing for all of us who've been building B2B SaaS for the past 10 years is to charge by the seat, and a seat is an end user. But it seems like seats don't really make sense in the agent world.

Manny: Exactly. They don't.

And that was the genesis of the company is that, you know, we're building these agents at Outreach and we will go to a CRO and/or a CEO and they were asking me, "Manny, what's my head count plan for next year? 'Cause I surely don't need as many people to hit the same number, right? And if I want to increase the number, like, what is the agent contribution to my efficiency?"

So, people are already banking on this. So, if you're banking on not having seats or not having more seats, then the seat model just breaks.

So like, you know, I came back and I'm like, "We got to change all this. And then we have to find something that aligns to what the customer is expecting in terms of value."

Not in terms of my cost, but in terms of their value.

Isaac: Wait, this kind of makes sense to me. So, if you're selling a platform for, God, let's say, posting your LinkedIn influencer posts on LinkedIn and someone's using that platform in the seat.

There's like an implicit cap to how much they can use it, which is how fast their little fingers can type, right? So, your margins can only suffer so much, right?

Manny: Right.

Isaac: In an all you can eat buffet, someone can only eat so much food so quickly per month, right? But these agents can eat food a hundred thousand times faster than a human being.

And yeah, I see a big risk there if you're charging per seat for the product, which is the industry standard today.

Manny: Right.

Isaac: That makes perfect sense to me.

Manny: Yeah. And not only that, but like, the agent is doing all the work. So, not only you're doing a lot more work, but the work is fully completed, right?

So, the agent is not delivering. Let's just say there is an agent that is posting bangers for you on LinkedIn and raising your follower count.

But at the end of the day, you're paying the agent to do a job for which you count, right?

So, you're either asking the agent to engage with more people or to create more followers, or to get more likes per post, or to raise your engagement.

There is all these things that you can train an agent to do. And the compensation is just part of the optimization model. And if you're not putting the right compensation in front of this builder, then you're both going to miss out.

The agent builder is not going to make all the, it's going to capture, they're not going to capture the value that they're creating.

And you are going to be paying for something that you don't want. So, it works for everybody.

Christine: How do people measure the quality of the outcome that it actually did what they want?

Manny: That's a great point. One of the questions is, you know, how do you know that you got your outcome?

So, the outcome definition has to be to some degree known or tight, but you can put the bar whatever you want, right?

So, back to the AI SDR problem, right? So, you can pay the AI SDR per email sent, right? But that's not a real outcome. And it doesn't really, that doesn't really drive anything for you.

Okay, so you can say, "Well, I'm going to charge per positive reply." It's still not there, right? It's closer, but it's still not there.

And you can measure the positive reply. You can just put up a sentiment classifier on top of it and figure out which ones are good and which ones are bad.

It's not hard to judge. As you go down into the final outcome that you're really on, which is on opportunity clause one, then the specificity of the outcome, it starts getting a little fuzzier, right?

Because like, you got a meeting. Is that a good meeting or a bad meeting? Or you got an opportunity to create. Is that a real opportunity created or a bad opportunity created?

But at some point you have to agree on what is it that you want. And if you don't know what you want, then I strongly suggest that you don't buy agents to do something that you don't know that you want.

If you know what you want, then you will know the quality of it. And then, the question is then, you know, is there a dispute resolution flow where you can contest some of these outcomes?

And that is all part of outcome-based pricing, right? Like, you can't be selling outcomes if there's no way to contest it.

And that's one of the things that we think about here at Paid is like, what are all the things that the new industry needs to be able to charge for things that don't exist before?

Isaac: It makes a lot of sense. I do just have to say though, what you're doing now versus what you did before, they're really different businesses.

How much of that like, an adjustment has that been for you to jump feet first into starting a completely new business in a completely new sector, one that didn't exist two years ago?

Manny: So for me, it was more, what are the abstractions that I can take from my previous business that I can apply here?

And pricing and packaging was something that always drove me crazy. So as we grew, we had consultants come into pricing and packaging.

And I found out that they were not, they were not particularly good. The pricing and packaging was not something that you drop in a consultant and you do it once and they leave, and you're a million dollars poorer.

It's something that you need to own. It's something you need to understand. The pricing and the packaging is something that your customer is the recipient of it. So, you need to go talk to your customers about what value you're delivering, how they want it packaged so that you can both strike value.

You know, it's kind of one of those interesting things that, you know, you don't want to do pricing and packaging in absence of your customers.

You always want to do it with your customer to make sure that you're understanding how is it, you know, where are you delivering value, how are you delivering the value?

How much do they value the stuff that you're delivering? Because if you don't have that conversation, they're going to turn anyway. You see what I mean?

Or they're going to be talked, you know, and different providers going to come to them and align closer to what they need, and they're going to go to them.

So, that hasn't changed. I'm just doing that all the time and I'm doing it for other people.

So, in a sense, I get to sort of like take the good parts of my last job and not the bad parts, and double down to the stuff that I found really interesting and mission critical.

And I get to educate everybody else about what I know that they don't.

Isaac: So, beyond pricing and package, which by the way, I completely agree.

Pricing and packaging is something that people want to touch once and then they say, "Man, that was so painful to change it. We're not going to touch it again anytime soon."

But it has to evolve with the business of, with your customers, I agree. Beyond that, what are some mistakes that you're not going to make again a second time?

Manny: Oh my God, I have a catalog of them. Like, where to start listing?

Because we are living in a new era, you know, we were growing so fast at Outreach at the very beginning that, we were building products as our customers were demanding it, but we were not building products to make sure that the operations run more efficiently.

And we would throw people at the problem, right? So like if, you know, for the longest time people want to know what kind of value they got out of Outreach and I would say, "Well, you know, I can't do that and build AB testing or build deliverability charts or build this sync that you want into this platform, or build X, Y and Z, or you know, be SOC2 compliant or blah blah blah, blah blah. Or put a data center in your..." Like, all these things I couldn't do.

Here, like, it's just a hard prioritization. I'm not going to be okay with bringing more people for something that software can do.

So the first question is, you know, is it a priority, A? If it's not, if it's a something that can wait, then you just don't do it.

The next question is can AI do it? And if AI can do it, then, done. Just buy the AI and done.

And then, the third question is, "Okay, so if AI cannot do it, you actually need a person. Then, can you bring a contractor to do for this, for like, the very narrow serious activity or outcome that you need? And just get that done so you don't have to, you know, go through the effort of increasing headcount."

And I'm sort of like in the train, I'm not going to put on, I'm not going to say like I'm going to build a company that is a million in dollars per employee because I don't know that that is true.

But I can tell you that keeping headcount very, very small and keeping them in the same location and keeping us all very pointed and accountable to what we're really getting done is invaluable. And it's the secret of running a high growth company that is very impactful.

So, I will not fall in the trap of like, oh, you know, you need to scale to see some regular results. That's how big companies do it.

I think that paradigm is broken. And we're just going to go first principle on this one.

Christine: And that's a really interesting point in that like, you know, I think that every first time founder underestimates the complexity that you add to an organization by adding people because of the fact that communications points between all those different people multiplies in N times N fashion as you add more people.

And it's an interesting point around, you know, designing a company ground up in this new world of, well, we're going to have people doing some things and then we're going to have AI doing other things.

And what if that allows us to almost think of a bit as like, operating the business in a more human friendly fashion because we are designed as a species to operate in small groups, and the sort of like, traditional corporate trappings are really something that was designed in sort of like, industrial America to deal with the fact that you have these huge groups of people that are working together to accomplish things.

And personally, I think it's really cool that hopefully there is going to be this new crop of organizations that keep things lean and have just a really tight team that's working really closely together that doesn't have to add that level of complexity.

Manny: Right.

And what I see is that there is some myths that you have to break from the SaaS world that don't apply to specifically agentic companies.

So for instance, in the SaaS world, there is a ratio of how much revenue you need to carry per CSM, right?

And that number is around 6 million and the majority of people run at 4. And in the agentic world, like, there is no training per se because there's no seats.

So, I see a lot of agent companies as they're hitting some scale is like, "Oh, I talked to somebody who used to run a company like mine and they tell me to go get, hire a bunch of CSMs."

I was like, "But you don't have any seats." So like, where are you going to go success? Like, either your agent is delivering the goods or it's not.

If your agent is not delivering the outcomes that they bought you for, then go fix that problem. Don't hire CSMs. Like, a CSM may be helpful for something else.

Maybe for the initial deployment, maybe for, you know, maybe you need fewer of them for the amount of revenue.

But there is this set of rubrics that we used to operate under that just don't apply anymore. And you have to think about, you know, what is the next generation of that?

So, in the world of agents, agents is performing all the tasks. So, what you want to show your customer is A, all the outcomes that you have generated for them and why they matter.

And B, all the other work that you have done for them that you may not be charging for. But that's a problem of product or code, it's not a problem of people.

We have to rethink how we build companies going forward. And we mustn't apply the previous frameworks to the forward-looking companies. They're not the same.

Isaac: I guess in general, you seem very cowboy-esque, right? I imagine you break a lot of frameworks or a lot of traditional way of doing things.

Do you have any other things that you do abnormally?

Manny: I think that in the startup world, you have to live with a set of principles that are not commonly agreed on. Otherwise, you don't have a startup.

Like, a startup is you found a, this location in the market that nobody else saw and hence you build a company around it. That's what makes your startup viable, right?

So for us, we found that everybody was building agents and there was a ton of frameworks for building agents.

There was a ton of models and ways on how to build agents, but nobody was worried about how these agents are going to turn into viable, ongoing concerns and companies, and eventually how will these agent builders get paid?

And like, literally nobody was talking about it. And I found it really strange that you would spend so much time on the building phase and zero on the getting paid phase.

And I'm like, "This is America. We're all here getting paid. And like, nobody's talking about it."

Isaac: I completely agree.

Manny: Yeah, it's... Being in a startup world requires you to like, not pay attention to established norms.

Isaac: Well, I think a lot of startups, and it's really weird, are weirdly trying to emulate Fortune 500 companies in regards to how they run, in regards to how they do things 'cause they think like, "Oh, if we do things the way AWS does things, maybe we're going to be like AWS."

Like, no, no, no, no. AWS or whoever, I'm using an example, they're at that level despite the way they do things, not necessarily because of it a lot of times.

Manny: Yeah, and you know, it only takes is one singular insight that is a winner to work, right? And to stick with it long enough.

So, there's a lot of innovations that have come out of Amazon that haven't worked and a few that just took a long time to pay out.

So for instance, I was at Amazon in 2003 when AWS was started. And AWS didn't work out for like, 15 years.

Like, I remember the first conference for AWS for users of our services. And there was like, 10 people in the room that used it, and half of them were related to Jeff. Like, there were not that many users.

And you know, they stuck with it for so long until it turned into something meaningful in a real business, but there's some others that did not that were like, the Fire Phone didn't turn into anything.

And the Kindle is, it's a little bit of something, but it's not the banger that we all thought it was going to be. So yeah, you don't have to be excellent operationally if the idea is good enough.

Isaac: Completely agreed. Man, I was thinking more along the lines of like, things like PR FAQs, stuff like that where, you know, like, oh, the way teams operated, these different companies do this thing.

Quarterly business reviews, OKRs. All of these industry standards for startups, they came out of big companies and we can't do the exercise of like, "Okay, how big would Google be if they didn't do OKRs?" Because they were already big when they started doing them.

Manny: That's a great question. What do you think it would be?

Isaac: No effect, frankly. I really think no effect.

Manny: I think it would just, it would be just as fucking big.

Like, it had nothing to like, yes. You maybe on the margin, you know, something will change. Maybe some stupid ideas.

Like Google Glasses wouldn't have invented. But yeah, no. The cash cow of the ad business will continue to deliver the goods whether you had OKRs or not.

Isaac: Absolutely. I think that also comes back to if you just have a hyper-viable business with product market fit, the rest can kind of come out in the wash.

You have a lot of wiggle room to do a lot of dumb stuff. I'm not calling OKRs dumb, but in general, as a startup or as a company, you can do a lot of really silly things if you have a phenomenal core business, period.

Manny: Yeah, a hundred percent. And to be honest, you know, now that I'm starting again, there's a lot of things that I don't use and a lot of things that I do use.

So, for instance... So, the running of the company is a daily thing. And I'm not saying, when I say running of the company, I'm saying, "Okay, so what we agreed to do in terms of a milestone, so our next milestone in two months is to, you know, we have this margin tracking product that we're launching in two months, which is an improvement over our V1 and we're working towards that."

And the working towards that has weekly milestones. And the weekly milestones have daily achievable things that you want to go do.

And then you check in on the weekly milestones, and then you check on the next weekly milestones. But it's a full contact sport. It's not a set it and forget it.

It's not, you put it over here and you go look somewhere else. And the reason for that is that building software is hard still, even after all the things.

And number two is that if you don't stay on top of the minutia of the details of the craft, you're not going to get over the line the stuff that you need to get done in the time that you need to get done so you can move faster.

And OKRs don't do any of that. This is more of a fundamental human behavior of like, are you drawn to the problem and is that something that you want to be excited about every day? And then you go and execute it.

There's nothing that OKRs can or cannot do that this other thing won't replace it.

Isaac: I think that OKRs are funny to me because really, I would call it, lazy operators or new operators, the way they do things is they give the inputs. They say, "Here's what we're going to do."

And the ICs, the employees are responsible for the outcomes. So the person managing says, "Here's what we're going to do."

And if it doesn't work out, the ICs are the ones who are in trouble. And I think good managers do the exact opposite of that where they calibrate the employees, they let the ICs come up with the inputs and the managers are responsible for the outcomes.

So, you have to, of course, check to make sure that the ICs are coming up with the right things that they believe will steer the outcomes.

But OKRs are kind of that wrong way of doing things where management sets these Os, the objectives, the inputs and the ICs have to increase X by Y percent.

And I mean, I talk about how if you create a metric as a goal, it stops being a valuable metric all the time.

I mean, Spang hears me talking about it at least twice or maybe five times a day. And I think it really just leans into that. Doesn't work in startups, in my opinion.

Manny: Yeah, look, I think that some metrics are valuable, and of course, you have to track it. Like, you can't deliver a number if you don't have pipeline, for instance, when it comes to revenue.

Or, you know, you iterate towards a product, but if you don't begin with the iteration, if you don't make it more complex or like, you know, test it in front of users, you just don't get to a good product that has good product market fit.

So, there is activities that need to happen. I just don't know... You know, a lot of times we get into this like, "Oh, we're going to set up a metric and hit the metric."

But if you're building, if you're terraforming a completely new era or a completely new category, I don't know that those metrics are going to be useful. So, metrics may be good for running a join, not for building a join.

So, you have to have some other principles that are not outcome-based metrics to figure out how you're going to build something.

And then once you start running it, yes, of course, you're going to put some metrics around it and then you're going to optimize for us that.

Isaac: Wait, so you think like North Star Metrics will be different in this world?

Manny: I mean, that's a broad enough question. Of course, like, yes. Of course it will be different. Of course it will be subject to the use case.

But I do think that if you're building something net new in a completely different way, the metrics that were used in the past don't apply to you anymore because those are different standards.

So for instance, in the past, a $200,000 revenue per employee company was a fairly well run company, or maybe 500,000. I don't think that we need that metric anymore.

I think that the metric needs to be north of 500,000 because you don't need as many employees to do the same work that you used to do before.

And I'm just giving you an example of a rough metric, and nobody optimizes for this to be honest. But things like daily active users.

How do you even think about that in a world of hyper-automation? That you're actually reducing the number of people touching your thing because you want your agents to run more autonomously.

Like, what is DAO in that world? What MAL in that world? Like, how's that even work? So, this engagement stuff, like, I don't know how to think about it in a world where there is no human engagement.

Isaac: Yeah, I don't know if there'll be an industry standard because even base hits or base-like actions that an agent is performing, I don't even know how I'd measure those in a uniform way across the board.

Manny: Yeah, like the ones that we track all the time when we have our customers is margin, right?

And mostly everyone comes from this 80, 70% world margin and you know, of running Software as a Service companies.

In the agentic world? I mean, what I'm seeing is something in the 50s. And if you are in a high, you know, in a multi-modal type of environment, like a customer service agent, you're in the 40s.

You see what I mean? And is that good or bad? I have no idea.

Because of course you can use a cheaper model, you can use a cheaper everything. But then your experience goes to shit and you don't know which one did it.

So, I don't really know what the new norm is, but I can tell you that it's definitely not backwards looking, it's forward looking, and we need to decide what is it going to be for the business that you're trying to build.

Christine: So, it seems like a core part of figuring out this new world is almost like coming up with a definition for what is useful work in a business?

Like, it's not people using your tool every day. It's that tool driving, like you mentioned outcomes.

Manny: Right.

Christine: And what does that look like? What does that mean?

And how do you make sure that that outcome is productive motion rather than just motion.

Manny: Right. It's funny because it reminded me, I don't know who came up with the whole, like, " jobs to be done" framing for everything product-related.

But it's roughly the same thing, right? Like, you're not, you'd have to assume, like, we all were told in SaaS, right?

Like, oh, you have to assume that your product is being hired to do this thing. And like, how close can we get to the thing that we're trying to do?

But we also, we always have this pesky human in the middle executing our task, right? Because you needed them to push the buttons.

Now you don't, right? You're truly hiring the agent to do a job. And if you don't have clarity about what job are you're trying to hire, then that's on you.

If a gardener shows up at your house and you say, "You called me, you wanted me to be at eight o'clock," and you have no idea what you wanted to do, he's just going to do a bunch of stuff.

And it may be what you want to do or you not, but you're going to get a bill either way.

So I actually love this fact that agents have autonomy and you have to be clear about what is that you want or not because it creates more accountability across the board.

Isaac: I think that's a really good insight that people seem to either go one of two horrible directions with agents right now as far as how they internalize and how they think about it.

Either they're thinking about it in terms of like, this very, very, very specific thing that may not need an agent at this point, right?

That can just be, I mean, you brought up a robotic process animation, right? That can just have that. And then, I hear people hand wave a lot.

Like, yeah, we're going to have AI agent programmers where you just kind of say like, "Hey, just build this feature in the app."

There's a reason that there's, let's say, product managers in the mix. And I mean, I'm biased, of course, having been one.

But there's a thousand questions that need to be answered by going to all these different people in the company, interviewing customers and just thinking about the future, where things are going.

You can never put that on autopilot. I don't see a future where you could. There's a middle ground.

Manny: Yeah. I don't know the answer to that. Like, I don't know, where does it stop?

But the current status quo is definitely what you say. Like, there is, you know, I hate when this big, I see this big mandates of like, "Oh, you should go use agents for everything." And like, that's a mandate.

I think everybody should be curious and I think everyone should explore, and I think everybody should sort of like, gamify their own job in that, you know, is there something that could be doing this repetitive task that is being done by me over and over again that is not me?

And I think that's the beauty of tech, right? Like, curiosity drives some of this insights and eventually you get better at it.

But there's a lot of tech stacks. Like, I know we're going to get to this question. But there's a lot of things that I use that I love using and my team loves using that have very little to do with AI.

For instance here, we're like, the team is addicted to Granola. And like, it does a great job to capture all sorts of information.

And the tool just keeps getting better and easier to use. And it's just like, you know, it's a joy to vibe with.

I'm sure it has AI capabilities, but we don't use many of them, you know what I mean?

Like, we just love the fact that it jumps in to your calls and just takes great summaries and clearly that's AI. But it's just very easy to use.

It's just a very good slick, good old fashioned UI. That's it.

Isaac: I think that there's a lot of catch-up that needs to happen first, right?

COVID was a big one, I think, because the way spreadsheets just were commonplace in the workplace.

And there was a time before that though, where people put like, "I know how to use Excel" on the resume, right? COVID pushed all of the general workforce like, "Hey, all of you need to know how to use Zoom now."

And that sounds to us like some triviality, but to a lot of people that's like a huge learning curve. And I'm seeing that in a couple different ways.

Like, you're talking about Granola. Like, there's AI transcription and summary and feeling comfortable with AI note taking and whatnot, which is an area at Nylas we forayed into recently because there just needs to be a lot of catch-up there before you can stack, I would call it, like, hardcore AI stuff on top of it.

There's debt from a usage standpoint almost.

Manny: Yeah, a little bit. But I also think that we can supersede the whole thing.

Like, Granola was the first application that completely helped me not to take notes. Like, I was a big note taker in meetings until Granola just made it super easy, super clean in a very intuitive experience.

You know, the ability for Claude and ChatGPT to be multimodal allows me to now write post that I just sort of like vibe with the AI and they give me a graphic too.

So like, my LinkedIn posts now have higher performance and they feel more complete, et cetera.

So, I didn't have to go through an evolution of like, become a better writer and become a better designer and become a better whatever, whatever.

I just like, look at the end state. What do I want to do? I want to write more frequently around stuff that I know about that has resonance in the market and I want to put a graphic into every single one of those things.

AI allowed me to just get to the end state without having to get better at it. You see what I mean?

So, I feel like there is a step function in productivity and the fact that I don't have to hire a bunch of ghostwriters to do the same work.

Like, I just do it myself and it doesn't take me as long, and I am 10x better at, the output stand is better than what it was before.

So if you're smart, you just get to sidestep all the learning curve and all the depth, and just move on to the next paradigm.

Isaac: I like that. That for me, that's the framework of rather than solving problems, just trying to nullify them. They just don't exist anymore.

Manny: A hundred percent. I think that's a better, that's my favorite framing. It's like that, you used to have this problem, now the project doesn't exist.

Like, there's so many things that don't exist anymore. So for instance, like, I was obsessed with content when I was at Outreach and I could never dial it right in my...

The content team was, you know, I had a CMO and the CMO had a product marketer or like a content marketer. Under the content marketer there was like a team, so by the time I got to them, I was like, "Jesus Christ, this is so complex. Like, why do I have a process that is so hard? You know, when in reality I just want beautiful content."

Like now, you know, and I didn't want to go through like, I mean frankly, I just don't want to deal with that.

Like, I want to be able to like, you know, if I want to build some, if I'm inspired, you know, I have another person here that runs marketing, we just get together and vibe together and come up with something or just do it myself.

You know what I mean? And it has more impact even than I had when I had a big content team. Why?

Because like, you know, I can prompt three things at the same time and it will come back with a beautifully written article that I can just edit a little bit and a graphic that allowed me to get out to market. And boom. Done.

Isaac: You know what this reminds me of? And this is weird historical analogy.

London had this, like, horse manure crisis in the industrial revolution where there were too many horses in London and the city was literally drowning in horse manure and they couldn't figure it out, and they had no plans of what to do.

And then, cars were invented and like, in a couple of years, the problem just dissipated.

Manny: A hundred percent. Yeah.

Isaac: Right? It was nullified. But it brought on new longer term problems, right?

You have things like global warming coming out of that. All kinds of stuff that are just more subtle, pernicious, and they're a little harder to see coming.

And on that, you talked a little bit about margin. I think it's probably a pretty good example.

Like, Spang, I know you have a lot of questions about margin in the AI agent world, right?

Christine: Yeah, you know, I was thinking about this just a bunch more while you guys were talking.

Obviously, like, investors have gotten used to just like, "We have these benchmarks and this is what the expectation for good is like a company."

And this also links back to what we were talking about earlier around sort of established best practices.

You know, everybody has these established best practices around B2B SaaS and what does good look like for this industry?

And like, where should you start caring about certain problems and what should the end state look like?

I'm curious if you have a take in the AI world around, like, obviously, you know, what you're saying is that margins are going to be lower, at least for now.

Two, that underlying tech is something that's evolving super fast and sometimes you can kind of work on, kind of the use case and the problem that you're solving on top of the models.

And at some point, either your quality is going to get better or your costs are going to weigh drop even without you doing anything.

If I'm starting a business building, an agent for some use case, when should I even start to care about margin?

Manny: So, it's really interesting because I've had this conversation.

You know, we're early, we're starting, we're talking to all these agent companies and I feel caught that we're in the middle, like, fundraising.

And you know, we're talking about monetization. And they were like, "I don't want to hear monetization."

My investors, like, all the investors I'm talking to want to know ARR. And there is no ARR unless I have some kind of subscription revenue.

And if I don't have subscription revenue, then I'm putting, I feel like I've got to put in the company in jeopardy compared to my investors.

And I'm like, but your customers don't give a flying F about your investors. They care about their outcomes and how are they aligning to you.

So, what is the agent delivering for them in terms of the outcomes? Number one and number two. You know, sure you can get a subscription.

You can get a subscription on many things. But your agent is not delivering a subscription-based service.

Their value is not a subscribable thing. The value is you deliver the thing until you don't. You see what I mean?

So, your investors ask you for subscription revenue. I would just look for other investors that are not asking that dumb question.

And I would look for investors that are asking you, "So, tell me what your agent does and tell me how many of those have you delivered? And what are your customers saying when you deliver that stuff? And what is that worth to your customers?"

So, I feel like investors to some degree are holding, were holding, I haven't seen that in a while. But were holding agent builders back by not being with the times.

On the margin side, it's a little bit of the same, right? Like, so they're expecting big margins.

But if you're starting an agent company right now. Should you worry about margins? Yes. The reason you need to worry about margins is margin is the result of your cost and the problem that you're solving.

And what margin does is it gives you a lagging indicator of the size of the problem you're solving and the applicability of technology you bring into market.

So, you cannot say blanket, which I think that's where you were going, Christine. You're sort of like inferring that it's okay for you to lose money at the very beginning because the models will get cheaper over time. Maybe. Maybe.

But maybe better models come out later that are just as expensive. So, that's what I was saying about AI SDRs.

I feel like if they're all using o3 to do research, the relevance of the communication will go way higher and then the quality of their product will go way higher.

But the cost will go way higher to the degree that I believe that they will be more expensive than a normal human being. I saw that in insurance.

So, there was this guy who was building agents for insurance companies and he needed to do RPA-like.

Meaning, the agent needed to run on a browser to actually perform the task that the agent, the insurance agent was doing. Like, the insurance rep was doing.

And the quality of the agent running on the browser wasn't quite there. And if you were to buy the top of the line agent running on your browser, you're going to cost more than the BPO, in which case economics weren't going to make sense.

So, the guy was a little stuck, right? Like, I have to, he's like, "I have to use a shitty model that sort of performs okay, but then I'm below the cost line that are BPO or but if I use the top of the line stuff, the BPO's going to beat me."

So, the question is, is the stuff that you're doing worth the top of the line model? Because if not, then go pick something cheaper and then make your margin.

But if you don't have a sense for the worth of the work that you're doing earlier on and you cannot squeeze a margin with the current models, then your problem is not big enough and you should go look for a big enough problem. That's why margin is important. It's a good rubric to figure out whether you're barking up the right tree.

Isaac: I feel like I've started thinking about margin very differently after just kind of learning a little bit more about like, different businesses that are really successful.

I can't, I know it's weird, I can't get Costco out of my head, right? 14% margin is their absolute max on anything.

There's that old Bezos quote, which I really like, which is, "Your margin is my opportunity."

And I think that a lot of these old, and I will call them old standards for margin, come from a world where there weren't a lot of options, right?

You're signing up for whatever you've got. You've got your one option and these 90% margins are kind of like implied in there.

But I mean, now we live in a world where, seriously, one of my product managers can just spin up Lovable, create an internal app that does 70% of what a subscription-based app does and he does it in a day, and we're all using it.

I think margin tolerances are also going to be different from users in the future.

Manny: A hundred percent. We need to examine margins as a lagging indicator of something else.

You see what I mean? You could build a great business. Dude, Amazon was built on about what? 20% margin business.

Matter of fact, at Amazon, everybody knew the contribution margin of each of the products.

Why? Because you didn't want to sell a lot of electronics. That thing was 5% margin. And we were buying traffic from the web into the electronic store. That was your cap.

You see what I mean? And whereas books, 20% margin all day long. You know, toys, 70% margin all day long. Jewelry, 25% margin. You can do that all day long.

But if you didn't understand your margin components and you couldn't really run the business, and mind you, like even, like the policy at AWS was to continuously drop price or whatever the input metric was.

Every time there was a drop in hardware or like a, you know, or some kind of new model came out.

So, they have been running this business on the full assumption that the market is going to be capped at some percentage, and then the price will drop.

You know why? Because they were trained by Walmart. The competition at AWS for the longest time was not like, you know, whatever, some other online store. It was Walmart.

Whatever they did, Walmart would do it cheaper. Why? Because Walmart had a bigger operation that was running more efficiently for a long time than Amazon.

So, they were trained to look at margins at all times and they became part of the culture. And I feel like we lost that in the companies that we're building now because we're used to these massive margins in SaaS that are just not the norm going forward.

Christine: Yeah, it's a really interesting perspective. I think also kind of the zero interest rate era really kind of took people's eye off the ball around margin.

Manny: You know, Christine, like, I'm a little tired. And I'm saying this with all the love in the world. I'm a little tired of blaming the interest rates.

You know, I'm a little tired of like, oh, it was a free money that made us to do stupid shit.

No, let's take ownership. We did stupid shit because we did stupid shit. And now, we're not doing stupid, or we're trying not to do stupid.

I'm loving this new paradigm of like this, you know, very small company founders that are very obsessed about growth and cash, and efficiency and keeping things small.

What a time to be alive. Like, I prefer this decade coming up than the one that we just went through.

Christine: I think it's going to be super interesting. I feel like there's this list of like, things that were best practices before that are now dead for the AI era.

What are those things? ARR is dead.

Manny: SaaS pricing. SaaS pricing. Seat pricing is dead.

Christine: Seat-based pricing

Manny: Done. Dead as a nail.

Christine: What else?

Manny: SEO. You heard it here first.

Isaac: Oh.

Christine: Oh, that's true. SEO, dead. I mean, it is kind of amazing that we're like, having to reinvent all these things that were just laws of the land for a decade.

Manny: But it's to the point that no, there's no reinventing. You just don't have to worry about it. Like, just imagine that you use one less fucking thing. Isn't that great?

Isaac: I mean, there's that Justin Kan quote I really like. "First time founders are obsessed with products. Second time founders are obsessed with distribution."

Manny: Yeah?

Isaac: How do you distribute in a world where there's one fewer thing.

Manny: You just build a really good product. Word of mouth. I mean, that still works. You know, build audiences in places that matter to you.

You know, I started reading Seth Godin lately and I don't know whether the stuff that he says works or not, but it's new. I just find him inspiring.

You know, every business, I think his rubric is every business just needs a thousand fans. That's it. With a thousand fans, you can be a solopreneur.

Now if you was, yeah, I don't know how many people you are, but let's say you have a hundred mouths to feed, then that's a hundred times thousand.

So, you need to figure out how to get to a hundred thousand people that love you to some degree because they will buy your stuff, they'll tell people to buy your stuff, or like, look forward to buying your stuff, and eventually we'll grow to something.

So, you just need a minimum viable audience to run your business, and then you can grow from there. And there is many ways to acquire that audience that will still work, you know?

Physical mail still works. Swag still works. Hackathon still works. Events still work. Dinners still work. LinkedIn still works.

Isaac: I like the dinners. I like the swag too.

Manny: And like for instance, like, yeah, it's one of the things that you're asking me, like, what am I stop doing?

So like for instance, I'm not, I am against printing swag because I want my swag to be super expensive.

I don't want to see Paid stuff on everybody. No, no, no, no, no, no. We're going to be making really high quality stuff that is going to be earned.

Isaac: You're going to be the Hermes of swag.

Manny: Why not?

Isaac: Or something, right?

Manny: Yeah, I mean, of course you can get a sticker or whatever, but if you want to have a Paid shirt, you're special.

Isaac: All right, well, I hope to get one someday.

Manny, thank you so much for joining us on the podcast. And for everyone listening, please check out Paid.

It is fantastic. I really think this is the future of this type of stuff.

Manny: Appreciate it.

Isaac: Thanks so much.

Manny: See you guys.