Lindy AI founder on the agent moment: B2B is ready, consumers aren't, and p(doom) is terrifying

working on it. We heard him for a second. Okay. Can we talk now? Let's hear it. Can you hear me? Yep, we got you. All right. Uh yeah, Flo here, founder of Lindy. We are building a no code platform to build AI agents to automate various workflows.

So like sales, customer support, operations, you name it, we automate it using AI agents. Yeah. Um I mean the question we've been asking everyone is like when can a agent book my flight or something like that? What's your timeline and what milestones are we expected to hit? I guess we're at a big like big picture level.

We're all just waiting for the Studio Gibli moment of agents. I think we passed the touring test. Everyone used AI to to do their homework. Everyone had that magical experience of like, wow, I'm talking to a computer and then it faded out and people used it as a answer engine.

And then same thing with Studio Gibli where everyone's like, "This is incredible. " Uh, but now it's like a tool and it's a filter and people use it tastefully and they'll be creative with it and sometimes they won't.

But uh when when do you think and what milestones do we need to hit to get this kind of breakout societal level meme around agents? A year a year. I I I I broadly buy the meme that 2025 is the year of agents. It's like a yearish. I think there's going to be a pretty serious inflection point.

And and you know I I think like when you hear this kind of prediction like a year is qualitatively different from like five years because like five years is like h you know a year means uh you see it you see it in the pipeline. I can can I share my screen? Yeah you can.

Yeah should come up but whatever you share is live. So if you have tabs open with passwords we're going to see it. All my passwords everything is is open. Be be careful.

This is this is a demo we had of actually this is really cool because it's it's uh marrying computer use which we haven't released yet with uh agent swarms which we released uh last week and you can see here I'm asking an agent like hey help me plan a trip to like raises in Colorado need to book a flight find anb and get a rental car.

Okay. And it's deploying a swarm and check this out. So it's deploying a swarm of three agents one for each. So, it's like booking a flight, finding an Airbnb and res resolving a car. Yeah. Decomposing the problem and it works. It just it just works. No interaction. No, no intervention needed. It just works.

It's doing all of these things at the same time. So, is this live? Uh oh. No, no, no, no. I mean, this is screen recording, right? Recording. Oh, yeah. Okay. This is like really fast. God damn. It works really well. No, no, no. This is This is This is the downside to this. Like, it's pretty slow.

You can see this is like a six minute recording, but in the end, so yeah, I think like this is going to be in production pretty pretty rapidly. Okay. Yeah.

Follow up like uh part of what made the Studio Giblly moment so powerful in my opinion was the fact that um it didn't require a lot of creativity on the part of the user. And so everyone has a profile photo. Everyone has a photo of their dog or their family.

They could just click upload Studio Giblly style and they get something amazing. And and I think that that's almost as important as the underlying technology to make something go mega mega viral.

Do you think that we're using the right example with book a flight or do you think are there any other use cases you've seen where it's like oh I could see there's a day when every single person in America or everyone who's online is going to use an agent to do this because it'll be fun and viral and they'll want to share the results or it'll make their day better.

That's a great question. I am actually quite bearish for consumer agents. I think like consumers are time rich and money poor. Yeah. Right. It's like they they don't want to trade time for money like that. Which is why I think like Google Assistant and Alexa like they've been relative failures.

And I don't think it's because of the technology. I think it's just like one your time and money pool. And two, I think the life of a consumer is very convenient because on the other side of every transaction as a consumer, you've got a business which interest is to make the transaction as painless as possible.

Y I actually think like the flight booking experience is fine. I think the restaurant booking experience is fine. I think all of these experiences are kind of fine.

I think like there is going to be a transformational moment for businesses where you have like the holy grail of AI agency like what we call the the the drop in replacement for a human worker. And I think that's coming in like 12 to 18 months. Yeah.

I I I almost I'm just kind of riffing on this, but I almost wonder if the agent moment will look like the Studio Giblly moment, but instead of a single picture, you'll say, "Uh, go and and do some research about my life or my kids' life or put in I'll talk to it for a bit.

" And it just produces an entire animated book or something like that. And it's something that requires using multiple systems and writing and iterating and generating images. And so, it's still highly personal.

And so anyone in America or anyone in the world can get joy out of that experience, but it's something that wouldn't otherwise be be created. Like you might stop on a boardwalk and have an illustrator draw you at a carnival. Um but most people weren't hiring illustrators on Fiverr to turn them into anime.

Uh but they could when it was just one button. And so maybe the maybe the agent use case will be like for for consumers will be denovo. It won't be optimizing something that they're already doing. It'll be something completely new that they would never have paid for. I don't know. What do you think? We'll see.

I mean, we had like one competitor that went viral like nine months ago that did something like that. It was like use an agent to roast my Twitter viral and then it fizzled out because like the market is that really I think even the the Gibli thing I think is going to eventually fizzle out. It's all fizzled out.

It's like I I don't know that consumers need all that many agents. I think I think businesses need agents. Yeah. Well, do you think it's a good uh like marketing strategy at least?

Like have you ever thought about like how can I get uh even if it's even if it's a meme product or meme use of Lindy uh and it's not going to be an enduring or it's going to be extremely high churn but I know that everyone will sign up for that day do one task and yeah I'll keep 2% of them but that's a great trade because I figured out how to go viral with my product.

That's a that's a good point. I think the downsides though is that boy it's so expensive.

AI sounds so expensive and so that's gonna be that's gonna be a $10 million bill like actually $10 million build and like Open AI is burning money right now on the Giblly thing and they don't care because they raise like $40 million have infinite money if you're a startup it's you know it's it's a different move. Yeah.

Uh what what was your reaction to Toby's leaked memo this morning around you know being AI native? Did you did you see it? I I skimmed it and I I forwarded it to the team and I said the cosign before I even read it. I was like whatever he says I I agree. I love it. That's good management.

But but do you think do you think CEOs in general have been wary around you know being too pro pro AI in their businesses to not almost like basically scare employees because you know some it's just about not scaring employees.

I think there is a culture in tech of avoiding to be too prescriptive on exactly the how and really leading on outcomes and like letting engineers like do whatever they want and use any IDE they want.

And I think that's the graph the gap has grown so much now between the frontier of the level of productivity that you can achieve as an engineer and the median practice of the median engineer especially in big companies that now employers have no choice but to just lay the law and be like hey this is the tooling we use like you don't get to use tooling from like the 1980s anymore.

Yeah. Um, can you talk a little bit about uh Lindy being sort of multi-product and and I I I can see a lot of reasons why, you know, for phone calls as an example. You know, there's companies that do just do that, right?

And I could see the reason for Lindy to to do multiple things is like, you know, a customer does a phone call and then they need an email and then they want a recording and like there's a bunch of, you know, sales, customer support, all all these different areas end up, you know, connecting in.

So, I can see I can see why uh you would want to be sort of multi-product, but uh I'm curious if there's any more uh I'm sure you've thought very deeply about that. um and and curious your being less opinionated about the actual use case for the customer.

Yeah, I actually think of us I actually think it's a single product.

It's just a lot of use cases and I think that's an important distinction because the beauty of building it this way is that you build a a set of actually few low-level primitives and then this it's like Lego bricks you know and then you can combine these Lego bricks however you want and they combine super linearly and then they explode into this set of of use cases.

the demo I just gave with like agent swarms and computer use. We didn't have to build this. We just built agent swarms and now we're building computer use and now these two things just like this demo just like happens like spontaneously basically.

So it's the same thing for computer use you know like when you build a computer use uh startup I summize that maybe 5% of your bandwidth is the actual not sorry not computers phone agents 5% of your work is actually phone like phone code and phone whatnot right the rest is like the scaffolding around the product right like the the SAS features around like team collaboration and like the the way to like orchestrate your agent and then the go to market overhead and all of that stuff and so I'm like man if we it actually didn't take us that long to build an agent because we all of that scaffold scaffolding around it.

And once we have that building block, it's just that much more powerful because it's it's it's plugged into all our other primitives. And so you you realize a ton of economies of scale from like a good market standpoint and R&D standpoint.

I think it's a better experience very often for the end user because you end up having all of your AI agents in one place. So you have like one clean pane of glass and you have all of your AI agents in one place. And I actually think it's going to be much more important even once AI agents can actually work together.

uh you don't want like having all of these different platforms for all of your different AI agents is maybe the same thing as having like five headquarters. You want all your AI employees under the same roof able to work together very very seamlessly. Can you talk a little bit more about cost?

You said that AI agents are very expensive to run. Uh what are you what was your reaction to DeepSeek? What's your reaction to Llama 4, some of the open source models that might be a little bit cheaper? Uh what are you monitoring on the inference side?

Yeah, I mean we are seeing costs fall by roughly 40 to 100x every year. Okay, so today you know we don't have negative unit economics but our unit economics are actually pretty thin. The margins are pretty thin and we're totally fine with it and so are investors because we're like it doesn't matter.

Like we have actually seen the the prices go down by 40x. So it's expensive today but it's it's actually still the ROI is is tremendous and and soon it's just going to be an absolute no-brainer.

How much do you think that's driven by um new new algorithms and optimizations versus just new hardware and uh put the transformer on silicon or bake the you know the the weights of llama 4 into an ASIC or something like that. Historically it's been mostly new algorithms. Yeah.

Well you know or like distillations you know for GPT4 or like GPT4 is so much smaller than like the original GPT4. like GP4 was like a 1.

7 trillion parameter model and GPT40 I think is like what 50B or something like that and then yes there all t like quantization and there's like a lot of optimizations at the level of the stack we know that these innovations didn't happen at the hardware level in the time span where we saw these costs fall uh can you talk a little bit about evaluation of new models we saw llama 4 dropped on Saturday how does your team actually test that do you just roll it out to a small set of users or you just playing it with yourself.

Do you have eval built out? Um how are you testing new LLMs when they drop? Because it seems like it's every week this this year. Totally. I I think like so we do have our own eval built out for like cool set of use cases that we're going after and there's a couple hundred test cases there.

Um and we also test by vibes quite a lot. I will say I feel bad for saying that because the I have friends like the Lama team but boy like Lama for the vibes are really bad. like the events are good and the vibes are quite bad actually. What what what do you think's driving that?

Is that just too much focus on pre-training which is hitting a wall on the scale side or is there something else going on? There are all rumors that they injected some data from the eval in the training set at the last minute. Yeah. Isn't that like data poisoning or something like that or yeah saturation? Yeah.

I don't know if you saw like the I saw the allegations. I apologize to my LMA friends, but uh the VP of AI research at Meta resigned and so have a couple members of the core research team seemingly in protest for what they deem to be unethical practices. Oh, interesting. We'll have to follow up on that.

Oh, and is that you think that just comes down to their AI team being on just having so much pressure due to Meta's investment in capex that it's like if you're not best in class, it's just clearly Yeah. You know what what are you doing? Basically, if you're not first, you're last.

I I think Zuck is is massively raising the temperature and I think he's basically threatened to like destroy the whole or like fire the or like similar stuff like that if they don't hit certain targets by a certain date and I think that's led to this outcome.

How uh how how do you think about AI adoption in enterprise but just sort of B2B SMB broadly because I'll give you an example.

So, and I brought this up a few times on the show, but dur during the stib studio Giblly moment, uh, if you had a post go viral enough, people would would respond to it and not know where it was being created, seemingly unaware of OpenAI chat GPT broadly.

When you talk to business owners today that are maybe outside of Silicon Valley, and you explain to them what's possible for AI with AI, are they are they surprised? Are they aware of this stuff already? Do you think they're they're quick to adopt uh these products? Uh you know, I'm curious. People are aware definitely.

Um we we're seeing some very interesting dynamics play out in the market where um SMBs are aware and they are eager to realize immediate ROI. So SMBs are actually the ones who have in my mind the most real deployments out there. They are actually deploying agents for mission critical stuff all day long.

Um enterprise customers are interesting because there is I think AI agents there is a little bit of a return of shelfware.

Um so you know shelfware is like that thing that happened in like the 2000s or like the 90s like Microsoft and Oracle like just like released this software very expensive and everybody the enterprise customers buy it because the pitch is compelling but then it gathers no adoption internally.

that just sits on the shelf and and I think that's happening again with AI agents because a dynamic was seeing play out quite a bit and I don't want to say names but there are some very hot AI startups out there in the enterprise that are basically shelfware like their revenue is skyrocketing and they're meeting very little adoption internally and what's happening is that the board is on Twitter all day and they're freaking out and they're on the back of the CEO and they're like you've got to figure out this AI thing whatever it means to figure out this AI thing and then the CEO turns around and talks to his executives and he's like you guys have got to figure out the AI thing and basically the [ __ ] rolls downhill and immense uh uh budgets get unlocked like hundreds of millions of dollars like figure out the AI thing.

Um and so you know we are seeing these huge AR numbers these huge growth numbers.

Um adoption is is low very often and even when adoption is their satisfaction is is low because the the products are not quite mature yet and even when the products are adopted they're not typically for very uh uh mission critical tasks yet like enterprises are notoriously and I think for good reason quite risk averse and so they're going slowly here as far as like actual adoption and deployment is concerned.

Yeah, that makes sense. Do you think about h how do you think about uh general risk of rolling out these products?

Right, if you're using cursor internally to write code and help you ship faster, there's, you know, if you make if if the if cursor makes a mistake, you know, you sort of can have the opportunity to correct it. The customer doesn't isn't necessarily made of aware of the mistake.

Whereas, um there's a bit more risk with with what you're doing, right? if if a customer gets an email and it doesn't make any sense or they're on the phone uh and and they're talking to uh an agent and the the agent is just sort of stumbling or whatever.

Um I'm curious how you think about, you know, testing uh I'm sure you're doing a massive amount of testing internally and then making sure that that these things are functioning before they're really um you know being rolled out uh at the customer level, but uh I I'm curious uh what your approach is there.

Well, first of all, I I always tell people to compare apples to apples, which is like the the the bar is much lower for agents than it is for other software. For other software, you want like three or four lines of reliability or what or whatnot. Agents, the comp is not other software, it's it's humans, right?

And when you hire humans and very often enterprise customers, they have their support done by BPOS in the Philippines or in India or whatnot, the quality is not 100%. Right? And so you've got to compare, you've got to compare this to that.

So the vast majority of the time for use cases where AI agents are deployed right now they more than clear that bar of human quality and then some right because they are obviously they are available 247 they respond to tickets in like a couple of seconds and then they give you perfect observability so you can actually see exactly task by task what your AI agents did did like all day you can't say the same for for for for humans and then on top of that with this noted which is like hey headline it's fine AI agents work pretty Well, but you know on top of that what you do is obviously you deploy eval you deploy human in the loop for like the most critical parts of a workflow.

So for example customer support agent if you're Uber we don't have Uber as a as a customer but I used to work at Uber. If you're Uber and a customer complains about like a sexual assault, don't have an AI just just like escalate that to a human as right.

Um or if a customer is asking for a refund on a transaction of more than call it like $500, escalates that to a human, right? And you can have these hard guard rails in place that are like this action always structurally requires a human in the loop for the AI agent to perform it. That makes sense.

Uh can you talk a little bit about distribution? uh with a lot of these B2B products, you hear uh oh yeah, anyone can set up a Shopify store, anyone can set up uh QuickBooks, but oftent times these products are actually vended into companies from agencies, consulting groups.

There's actually a like a middleman that's that's doing the implementation and the maintenance of the software. And in some cases, the companies don't even know that their emails are being sent by Mailchimp under the hood or Claio or any of these different uh SAS products.

Are you finding do you think that'll be a material part of your business or or are you thinking about that? Yeah, it is a material part of the business today. We have a partner network. If you go to like lindy. ai/partners, we've got like dozens of implementation partners that that help us implement these things.

I think it's temporary though because AI agents are soon going to be simple enough. Again, if you have a drop in replacement for a human worker, it's going to be no more complicated than collaborating with your human teammate. Um, so I I don't I don't know that you're going to need this for all that long.

Uh, how do you think about collaboration between agents? I was reading the AI 2027 thing and they were talking about uh oh well like when there's a billion AGIS uh they'll just set up a Slack workspace and they'll talk to each other literally in Slack. Yeah.

Uh you you gave that demo of the three agents working on separate things and then I imagine that their results are kind of put together. Are you thinking about communication between the agents during the process?

Because you could imagine that oh, it finds a great it finds a great place to stay while you're traveling, but that requires taking a slightly different flight and there's a trade-off there.

And if you had a really great human working on booking you a trip that involved a flight and a and a stay that they might want to talk to each other about the trade-offs there and kind of do some sort of optimization.

You don't just want to book the flight and then also book the best hotel or whatever because they might not be the same. Yeah. Oh, and so this is currently something we support. So for example, like we do a lot of recruiting, you know, engineers or designers or sales people, anything while hiring everything, hit me up.

Uh but you know, I have a Lindy that's really good at sourcing candidates and then I have a Lindy that's really good at reaching out to them. So she perform research about them. She knows exactly how to reach out to each person and all of that stuff. And so these two lindies are are in touch with each other.

So one finds candidates, the other reaches out to them. Um yeah, I I think that's going to be super important and that's part of why we are so passionate about having all of the agents in the same platform.

You could say you don't need them in the same platform because they can use platforms like Slack and whatnot to to collaborate, but and that is part of the AI 27 uh 2027 essay, right? He he talks a lot about neuroles. He says like agents are not always going to go communicate with each other in uh human readable tokens.

Um and and that's also part of our thesis and so that's going to be much easier for AI agents to do if they are part of the same platform. Um I think ultimately this like the AI agent team topology is going to be one of the main drivers of the performance of your of your AI agent setup.

And I think at first it's going to be on you as a user to figure out that topology. In a way you're going to be doing the the job of a CEO. It's like hey what's the reporting structure? Who communicates to whom? How? Like all of that stuff you're going to have to design eventually and in the not too distant future.

You are going to have an AI agent design design that for you. So we call that the AI chief of staff. It's just going to manage all of your AI agents and you can imagine this recursively, right? Just like agents managing agents managing agents. Uh what's your PDoom? Are you scared?

Uh does AI does AI scare you or you just like no we're putting them to work? I'm terrified.

Um my pedum is about look I fluctuate the error bars are huge but like look even if it was low like 10% that's unacceptably high if you listen to the dor the dorkish podcast on 2027 that's the part I love dorkish I love the podcast but that's the part where I'm like what at some point they talk about pdoom and one guy is like yeah I'm 70% and the other guy is like I'm an optimist I'm only 20% and then they move on I'm like can we spend the whole conversation talking about the fact that like very credible very smart people who are who spent a lot of time thinking about this literally think that there's a 20% chance that we as a civilization go extinct in 10 years.

Like I'm I'm terrified. I I I think I I think it's very under discussed. I think it's it's pretty and I think that's Is it possible is it possible that Lindy at some point pivots to being a defense company, but you guys are deploying agent swarms to to fight the uh robot apocalyp our robot adversaries?

Look, you know, we're selling to SMBs. who you are but a time agent company.

Um um I mean uh I I I want to talk about something very anodine compared uh uh just uh the idea of like putting an agent on a cron job that feels like something that's that's potentially uh so simple but very value creative just okay I have a task that I need to do and I want to make sure it happens every single day.

Are you seeing traction in that vertical or are you messaging that type of activity and and and what are the what are some of the cool use cases like do you have any like cron jobs set up where an agent does something for you even if it's simple uh just checking a newsletter checking a website just doing something pretty simple but it's something that happens reliably every day that I have a ton I have a ton so I have one that actually I received the email from this morning that checks out my favorite podcasts uh every week and Uh because I I don't really listen to podcasts anymore and so just like takes the podcasts, transcribes transcribes it and sends me a summary of my favorite podcasts every week.

Yeah. Um I have a couple that ping me on Friday nights that so one of them looks at my calendar and tells me like and looks at my priorities and tells me whether my calendar is aligned with my priorities. One of them looks at my personal CRM and at my calendar and like hey you met with this guy this week.

He seems interesting. You should hit him up. That's cool. One of them, one of them looks at all because I have a Lindy in all my meetings and so she looks at all my meetings for this week and she sends me a digest of all my meetings this week. Uh we have one that like every day, this one is really powerful actually.

It's my one of my favorite. It's companywide. So every day it ingests a ton of information from the company. So every external meeting we had with customers, every support ticket we received, like a lot of communication and then it sends a companywide digest to the whole company like the general channel on Slack.

So like in a couple of bullet points you get like that heartbeat of the company. That's been one of the most surprising use cases to me is AI agents are really good at just ingesting a ton of data very rapidly and summarizing it. So writing reports is is a really really strong use case right now.

Do you need a larger context window for that larger task or does it work with any of your models? I mean modern models have like 200,000 tokens context windows. That's enough for like 150 pages book or something like that. That's that's plenty. Okay, last question. We'll let you go. Um, what about the name?

This seems like the least Lindy company possible. It's like very new age and futuristic and sci-fi. Uh, is it a reference to the Lindy effect or did you just like the name? No, we have a teammate named Lindy actually. Oh, okay. Yeah.

And so we were just like looking for a name for way too long and then she was in the room. We were brainstorming like, "How about Lindy? " And so yeah, all right. Lindy. That's great. I love it. Can never leave now. Yeah, she's got to stay. Do you have a This is great. This is great. Thanks so much.

We got to get some lindies going internally at TVPN and yeah, where can people sign up, get started? What's pricing look like for your average consumer B2B? Yeah, go to lindy. ai and don't worry about the price, right? It's it starts at $50 a month. Yeah.

And we also offer to my point about the implementation like you know if you've got stuff you want to automate, hit us up. We'll we'll use to implementing men for people. That's amazing. Amazing. Well, thanks so much for joining. Great to chat fluff. This is fantastic. Talk to you soon. Bye. That was a great day.

He he he writes about it a lot on his on his uh on his ax. It's uh always scary when you talk to somebody who's close to the metal and terrified of the metal. Yeah. But I I feel safe knowing that he could turn his swarms against, you know, that's all that matters is there's more good robots than bad robots and we win.

Yep. But hopefully hopefully we we outnumber them. Uh anyway, any other posts you want to do from the timeline or should we get out of here? No, I honestly was just looking at Poly Market pulled up. I was I was tracking. I don't know how this feels. Um, it's crazy.

Do you know that Poly Market right now is only has a 1% chance of a US recession before May 2025? Well, that's like it would be very very hard for it to happen in May because a recession they would have to like rewrite the data like Yeah. Yeah. Yeah. Yeah.

So recession is defined as like uh I I think it's actually it's usually defined as like multiple quarters of of economic contraction. So so it's basically impossible. Yeah. Yeah.

It's unless they were to rewrite unless they were to basically get new old GDP data because I mean these like the market is a leading indicator right now. You're seeing people I mean even on the show we're talking about will companies pull back? Maybe they will. Will consumer stop buying foreign made cars?

Will they stop buying cars overall? Are is there gonna be inflation? Like we have no idea what's going to happen to the to the business community or the or the American consumer.

And so for GDP to just suddenly shrink it would be very rare unlike what happened in COVID where you know all the businesses were closed immediately and there was a massive drop in GDP very very quickly that all came back.

Uh this is very much I mean you talk to the smartest people in the tech and finance world and they're all saying some version of let's wait and see. I don't know the the market's down. It doesn't seem great but maybe it'll work out. I don't know. I'm kind of just doing the same thing. I'm focused on AI or whatever.

And so I'm sure there's a lot of people that Yeah. They're they're upset that the market's down, but they're not fearing for their jobs or really like pressing pause on that purchase or that vacation. Yeah, they're just kind of continuing on.

Anyway, uh we'll be digging into it more throughout the course of the week, so stay tuned. And other than that, thanks for watching. Thank you. Uh it's going to