June 12, 2025 • 46 mins
When we first spoke to Jacob Bank, founder of Relay.app, the product was ambitious, powerful, horizontal, flexible, but a bit challenging to grasp.
It was early, and the clarity wasn’t quite there. Fast forward a year, and something’s changed.
Jacob’s post caught fire on LinkedIn: customer love, sharp metrics, and unmistakable momentum. It was clear that we had to bring him back on the podcast.
Highlights include: 3 Phases Your Startup Might Go Through (2:06), Two Metrics to Keep Your Eyes On as a Founder (6:10), Educate Your Audience, Then Sell to Them (14:53), Can AI Do More for Me? (24:20), And more…
Stay updated with our podcast and the latest insights in Outbound Sales and Go-to-Market Strategies!
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:10):
welcome back to the predictable revenue. Podcast I'm your host. Collin. Colin. Wow! Let me try that again.
Jacob Bank (00:11):
Only done it a hundred times. Yeah.
Collin Stewart (00:12):
400 times. Yeah, yeah. All right, welcome back to the predictable revenue. Podcast I'm your host, Colin Stewart. Today, I'm rejoined by Jacob Bank from relay. And we are going to have a conversation. And we're going to have an update conversation about kind of what's what's happened. What's changed? Because we talked a year ago just over a year ago, and you were in.
Jacob Bank (00:13):
It's crazy that that was only a year ago. A lot has happened since then.
Collin Stewart (00:14):
And the whole. Thesis for the podcast in the repositioning around the founder stories was, I could have initial conversations and then wait until they take off a couple years later.
and then I saw a post of yours on Linkedin, and the customer love the metrics, the reach. I was like, man. We've got to have the conversation, because I'll admit when we 1st talked, I was like, I kind of get the product. It does a lot of really interesting things. I kind of want to do those interesting things. And I'm typically an early adopter of stuff like this.
I created an account, and I didn't do anything with it, because I'm like I there's so much here I don't get it.
But obviously that's changed. And so I want to talk about what's changed.
Jacob Bank (00:18):
Let's do it.
maybe the most valuable starting point would be. I can tell you about the 3 phases of the company we've gone through since we launched and what was going on in each phase.
So first, st for some context, relay dot app is an AI agent and workflow building platform. It is a horizontal tool which means
any company can use it for a wide variety of use cases. You could use it to extract information from your invoice emails. You could use it to automatically draft Linkedin and Twitter posts for you. You could use it to automatically analyze all of your customer calls and give your sales reps coaching. So it's a very broad, flexible platform, and horizontal products like this are hard to bring to market for 2 reasons at least number one
is, they're hard to market because you have to figure out a positioning and messaging that communicates who it's for in a way that they understand what it's for. And then, second, you need to find a channel by which you can reach them. It's not as simple as saying, well, our Icp is Hr leaders at technology companies between 501,000 people. And we're going to get this list of targets. And we're, gonna you know, cold email them
so for both a positioning and messaging side. And from a go to market Channel side, it can be harder. And then, second, it's harder from a product perspective, because you need to build a single tool that can serve a cupcake bakery all the way up to the customer success team of a fortune. 500 company.
both in terms of functional breadth, industry, breadth, and then breadth in the level of technical sophistication of the users. So that's kind of the context of like what the product is and what some of the challenges are. When we 1st launched in October 2023, the most approachable way. To bring a product like this to market was to anchor it against an existing market leader.
and the way we chose to do that was to position ourselves as the modern alternative to Zapier. Zapier is a well-known product. They're a multi-billion dollar company they are synonymous with. If this, then that automation for work, use cases for a broad audience. Funny enough if this, then that sort of should have been the leader in that space, because they invented the term, but they ended up focusing more on
connected home and consumer use cases. So Zapier, Zapier sort of inherited Iftt's crown in the 2015 to 2020 time. And we had seen this pattern work successfully for many other classes of Sas.
for example, linear is the modern alternative. To Jira. Audio is the modern alternative to salesforce loops is the modern alternative to Mailchimp, etc, etc. In every category where it felt like there was a 1215 year old, incumbent leader. There was an opportunity for a new player with a simpler ui, maybe more AI forward to gain, to gain, mind share.
So that's how we position ourselves in October of 2023. We're the modern Zapier alternative. So if you're the kind of company that picks linear, if you're the kind of company that picks adio because they're well designed, well crafted, really performant high quality for built for, for fast moving, strong teams who care more about
high quality end user experience than they care about enterprise check boxes. We were that kind of that kind of product.
And
that's when you know, back to our previous conversation. That's when we 1st exited the desert, and it started to feel like we had a little bit of product market fit. We got our 100 ish paying customers, we got our 1st dose of user love. Our retention started to be pretty good, like. So we were stacking. We were starting to stack cohorts. And I was like, Oh, yeah, this is. This is turning into a real business.
That was phase one October of 2023 to June 2024, phase 2.
We realize. Okay, we're now on the map as a Zapier alternative. We are getting some users that are zapier refugees because they're unhappy, for whatever reason, with the pricing or the robustness, or the quality or the interface, and there's.
Collin Stewart (00:34):
It looks anywhere.
Jacob Bank (00:35):
A steady trickle of those people over to us, but we felt like that was missing an opportunity in a world where there was such a huge tailwind of interest in AI. It felt like we were dramatically under utilizing that tailwind. So in June
we did a big relaunch that instead of framing ourselves as the modern alternative to Zapier, that differentiated primarily on usability. We became the AI 1st or AI native automation tool. And we built really really high quality. AI steps so that you could take your traditional workflow.
When a form is filled out automatically, add the contact to my Crm, you could add an AI powered lead, enrichment, and qualification that would work really really well and be very simple. To set up.
That led to another inflection point in the curve. It led to an inflection.
largely because retention and engagement went up further. Oh, my camera went out of focus there for a second. Alright!
There we go.
So when we did that, AI 1st launch, it led to another inflection point in the curve. Retention and engagement improved. User love improved.
But I still felt like there was too much potential energy locked up in the system in the sense that
we had 2 charts that would really tell the story. The 1st chart was our overall revenue growth and our active user growth. And that looked pretty good. It was not quite like the vertical line hockey stick, but it was like a very strong up into the right. But the number of new signups we got each week was pretty flat.
and so when you have those 2 in conjunction, it's a good sign and a bad sign. The good sign is, it means you have a high retention product that without a ton of top of funnel. You're still able to grow the business because your old users are sticking around and your new users are coming in. And so you're stacking cohorts on top of each other.
But that doesn't work forever. It also means
that you haven't. You haven't figured out how to break through and create a compounding top of funnel strategy that makes you top of mind for your customers at the right time.
That's when
I wrote a 25 page document for our team and for our investors at the end of the year in December, and the document covered a lot of ground. But the fundamental point of the document
was that I thought we were mispositioning ourselves. Our product was actually really good. I thought we were mispositioning ourselves in a way that was failing to take advantage of the AI agent, Tailwind.
and my claim was, nobody cares about no code automation that sounds robotic and nerdy, and a lot of work. But everyone wants AI agents to do work for them. And even though agents are actually a technical term, they've just become in the popular lexicon something people care about. I never would have predicted that agent was agent is a weird word. It's not what I would have predicted people would have cared about, but they do.
And so my claim was that by repositioning ourselves from the new standard for no code automation to the easiest way to build AI agents that work for you, we would be able to unlock a broader degree of awareness and mind share that would communicate to the true target audience of everyone who wants to use AI to help themselves that we existed. So I wrote this document. I shared it with investors on
like December 21st or December 20 second, something like that just before Christmas.
on December 23, rd I was like, Oh, let me put up a little trial balloon. Let me create a linkedin post. That uses this new framing of the product
and creates a cool gift that shows what this agent can do. So the Linkedin post was something like this is before these posts were all over the place which they are now. It was the post was I made an AI agent that automatically looks up Linkedin profiles for me. Here's how it works. You give it all the information you have about someone, their name, their email, address, their company their location, and it quickly and accurately finds their Linkedin profile. You can use this for recruiting, for sales, for meeting preparation. It's pretty cool. And then it was like literally like a 1 4 bullet post.
It had a gif that showed a Google spreadsheet with
(00:56):
information about the targets on the left, and then it would fill up with the Linkedin posts on the right, and it got 600,000 impressions, and like 7,000, and like 5,000 comments, and I had been posting regularly on Linkedin for the year prior to that, and I had never cracked 10,000 impressions and 100 comments. And so it's just like, Whoa, okay, something about
the framing of an AI agent doing work for you, something about the use case, something about the the corresponding asset, like something about that. And it was funny, because it was literally the 1st thing I tried. I'd written this long document with like, here's my hypothesis that this will resonate, and then, like it literally worked the 1st time, which has never happened in my career in any situation. It doesn't work, it doesn't work every time. And so
that's when we repositioned the product from the new standard for automation to build AI create AI agents that work for you. Our landing page conversion immediately from making that string change at the top of the website. I can't remember the exact numbers, but I think it went up from 7% to 10%,
taking one string at the top from landing page, from landing on the page to converting to signing up for an account like that is a crazy, that is a crazy impact from a 1, a one, a 1 tagline, a 1 tagline change.
And so that was an unlock for us for 2 reasons.
Number one. It was an unlock for us in positioning and messaging. We were no longer the no code automation tools for nerds that weren't quite happy enough. With Zapier we were the easiest and most accessible product for everyone who wants to create AI agents that work for them, and that is a much larger set of people. So that was the messaging breakthrough that also led to a Channel breakthrough
which for the 1st time we had content that people wanted to share on Linkedin.
And so my post started to do really well. But also other people started to post about us because no one
people didn't really post. Here's my cool automation that I built on Linkedin people love posting. I just built my 1st AI agent. It's really cool. Here's what it here's what it does for me. And so
we had experiment. I think we talked about this in in our previous conversation, but to quickly recap it, we had experimented with everything
warm intros from our network cold, outbound email interacting with relevant threads on Reddit partner marketing bottom of funnel SEO, we'd experimented with all of the the go to market channels that you programmatic SEO and content. And you know, editorial content. We'd experimented with all of the go to market channels that that would make sense for a product like ours, and we hadn't found the one that quite clicked.
And then organic Linkedin just clicked. And so
I went from in December. I had, like 4,000 followers, 4 or 5,000 followers, and now I have, I think, 47,000 followers.
I went from getting maybe 10,000 impressions a week to a few 100,000 impressions a week, and then, more importantly, a few 1,000 comments, or at least at least a few 100 on a good week, a few 1,000 comments, and it went from no one hearing about relay anywhere. To every day people would see a couple of posts on their linkedin feed, either from me or from someone else.
and then that has led to
the hockey stick trajectory over the last the last 4 months, because it felt to me like we're finally unlocking the potential energy that was stored in that product that was really high retention and engagement and people really love. And now we figured out how to communicate it to people.
Collin Stewart (01:12):
And you seem just from a you seem so much happier about the product, so much more relaxed than the like, that, like tightly bound up that we all have when we're like, we know our tool is great, but we haven't found the area for greatness yet. You seem like you've relaxed.
Jacob Bank (01:13):
Yeah, and it's
There's a few elements to it. On the one hand.
I always thought we had a great product, and my role within the company as the founder, and CEO is to figure out how to bring that great product to market. And I felt like I wasn't. I wasn't doing justice to the opportunity of of what the team had built. And then, second.
I felt uncomfortable because I was very new to marketing as a discipline I'd always been in product and technology and doing a new, hard thing that you're not experienced in is can be very demoralizing, especially with something like marketing when so many things don't work before you find the thing that does work.
And now it's pretty fun, because now it's it's
so funny how the product market switch flips overnight, like even 5 months ago, we had to beg
for users who had signed up to meet with us to give us feedback about the product. And now, if I offer that call to action in more than you know, 5% of we, we have like a qualification filter. And we offer the the opportunity to talk to a member of our team to like 5% of new users, I offer to 5% of user boom. My calendar is booked out, for, like as many slots I have, I have available. And so but it's just funny how that happens overnight from like I have to beg people to talk to me about this thing to. There's like way, more demand than I can possibly satisfy.
Collin Stewart (01:20):
Earlier. You said it sounded like you had unlocked the Channel like organic Linkedin as a as a channel, which absolutely it sounds like it's working for you. But you were using organic Linkedin before, and it wasn't working. It sounds like the big change, was the positioning, the repositioning of relay. And how it can have an impact on or like how you communicate it to users. That was.
Jacob Bank (01:21):
It was a combination of the repositioning of the product overall, and me identifying
a few topics of interest and post formats that are really resonant, and these are changing. So I had a few. In January I had a few very popular posts that were of the form. I built an AI agent to do. X. Now, those posts don't perform as well anymore, because everyone's doing. They're like, if you go to Linkedin, you'll just see like 50 posts. I built an AI agent to do this. I built that like that post format is now saturated. I got I got my 4 or 5 posts with hundreds of thousands of impressions each. And now that's saturated. Then
I realized there was a huge educational opportunity that
people were starting to recognize that building AI agents was going to be an important skill for the future of their careers, and between all the shiny product Demos, it wasn't clear how to actually get started, and so one of my most probably my most surprising Linkedin post ever was.
This was back in mid-february.
and I posted what I thought was going to be a very generic webinar announcement that would get, I don't know. 30 people interested 50 people interested.
and the post was.
if you're interested in learning more about AI agents. But you don't know where to start. I'm hosting a 1 h live session, and we'll build an AI agent together from start to finish, and you'll have a working thing by the end of it. Very simple. One slide Google slide poster, not well designed at all. Just like build an AI agent with me. Here's the date and time. Let me know if you're interested.
and that post got like 700,000 impressions and 6,000 comments, which those are like big numbers for Linkedin
And that shocked me because I was like, oh, I was just announcing a webinar. But again it tapped into this demand, where people are like, oh, I want to learn. And so I had a few other live session posts that performed really well. And now I do my webinars once a week. I don't promote them as much on Linkedin anymore. But I get, you know, a few 100 people every week, and we kind of have that channel really working now and then. I hadn't had a very successful post for a month or 2
about 6 weeks, and then
I stumbled across across this other post that I thought was pretty cool where a person had created a map of an org chart of AI agents.
but I didn't, really. I thought the idea was cool, but I didn't really like the execution of it. I was like, Oh, that's not real like. No one is like the way I know enough about the space to just like I can tell the way you set that up like. That's not how you're actually using them because it wouldn't. It just wouldn't work in that way. So I thought, Okay, I can use that. And it got a lot of that post had a lot of engagement, so I was like, Oh, I can use that sort of same theme, but I can counter position a little bit and say, like, Here are the 40 AI agents I actually use for marketing, and they're all real. They're all actually running.
And I made that post about 2 weeks ago.
and it got 1.5 million impressions and 16,000 comments, and I don't know for sure I haven't been able to validate this. I don't think that's a record for impressions, but I think 16,000 comments is the most comments ever on a Linkedin post, at least asking catch Bt. And trying to do research, I was unable to find evidence of another Linkedin post that has received more than 16,000 comments on it. And so again, I realized, like
within this content theme of people. Because people are
the way I'm kind of thinking about it is
there's this really shiny object called an AI agent, and people are kind of circling around it. They're like.
what would I use it for? How do I make one? What does it actually look like when people are using it well. And so I don't think any one post format can consistently go viral.
But if you can, if you can tap into this need that people are feeling to better understand this, this really powerful new concept of AI agents, and how they're going to use it, that has turned out to be something that we can durably get people excited about on Linkedin, and that's just my post. But I will say that even when
my posts were starting to do very well. In January, February I still had some anxiety because it still felt funnily enough, with the podcast name predictable revenue. When we were totally reliant on me to have a viral post that felt unpredictable and unreliable, and that felt like a brittle place to be, because there's so much about a viral post that is outside of your of your control.
(01:42):
Now, we're in a different situation for 2 reasons. One is.
we've built up enough of a baseline awareness and a base like between my follower count and the number of people have heard of relay. That individual posts don't have to go viral. They just have to attract the 2 to 500 right people who are like the target audience for that post. So my posts, even if they only hit 10,000 20,000 impressions and 100 comments. If those 100 comments are for the right people like I made a post about a specific class of SEO SEO ranking, tracking agents that you can build
if that catches the right. 200 social media like SEO and content marketing managers like that's perfect. That's the tam like. That's the audience for that post. And then, second, like, I mentioned.
other people are proud to show what they're doing with our product with AI agents. And so every day there are 3 or 4 posts that are that are not from me. And so now
I'm never comfortable. I'm never comfortable as a founder, right? You can never be comfortable as a founder. So a year ago, when we spoke, I felt like we had very strong retention, which is the foundation of any good business, and that we just hadn't figured out how to unlock top of funnel and get people in. Now, I feel like we have very strong retention.
We have very good top of funnel, but there's still a lot of drop off between those 2, because
there's quite a leap to go from.
I've heard of AI agents and relayapp, and they're cool to. I have actually set up a running agent successfully, and it's adding value to me. There's a lot that has to go right in between those 2. And that's the problem we need to solve now, and once we can unlock that we'll have like a further inflection in the curve, and we can talk again in a year.
Collin Stewart (01:50):
Amazing.
Yeah, it's it's such a cool way to think about it in terms of like you've got your retention in place. You've got your the untapped. There's more untapped potential that you're putting into the system.
Jacob Bank (01:52):
I have way. More leads in our Crm that have engaged with multiple my Linkedin posts than I possibly have time to contact.
Collin Stewart (01:53):
Totally.
Jacob Bank (01:54):
That's a good that's compared to where we were a year ago. That's a really good place to be.
Collin Stewart (01:55):
I like to think of product market fit on a continuum. It's not 0 or one. It's not binary, it's
strength, you know. It's it starts at 0, you get to one. And you're like, Hey, we've got something. And now you're at kind of like 7, you know, and you can probably get to 30 or 50.
Jacob Bank (01:57):
My definition was always like the 1st and most important definition for our class of product, which is a self. Serve bottom up. Plg, B. 2 B. Saas product was.
If the whole team disappeared for a month, would we have more users and more paying customers and more revenue than we had at the beginning of that month, because that means there is a sustainable, compounding engine that is working. Now that definition doesn't make sense for every kind of business, but for our kind of business. That means you've got something. If you don't do anything for a month, and you end up with more users and more customers than you had the month before, and that we had already reached when we last spoke a year ago.
but the number would have been only a little bit bigger.
Collin Stewart (02:00):
Like.
Jacob Bank (02:01):
It was like, yes, like our retention was good enough that we wouldn't have lost many people, and our top of funnel was good enough that we would have gained some people, but we would have if we all disappeared for a month we would have grown but grown slowly. But that's that's an important foundation, because it means you got something. You got something, and then you just have to bend the curve. And now we're at the point where, like the curves pretty good. And now I'm just looking at like, oh, if we if we do, if we improve this element of it, we bend it. If we improve this element, we bend it, and now it's cool, because
yes, it's a spectrum. But I think we've crossed from the 0 to the one. And once you've crossed from the 0 to the one, if you just get 1% better every week that starts to have the compounding impact that leads to like long term sustained hockey stick, growth.
Collin Stewart (02:03):
Yeah.
So what comes next for you like to to bend the curve even further.
Jacob Bank (02:05):
Our product. Strategy
is to be the easiest to use and most intuitive way for less technical people to get the value of AI agents and workflows working on their behalf.
So it's the kind of person who's just starting to play around with Chat Gpt, just starting to experience the magic. And they're like, this chat gpt thing is pretty cool when I'm typing to it. Can it do more for me when I'm sleeping like that person who has that thought like, Oh, could it do? Could it do more for me? I want to be there for that person. And say, let me teach you how this technology can do more for you while you're sleeping.
And that means that we need to build
the simplest and most intuitive product experience.
The best
set of inspirational materials of here's what you can do with this thing. And then the best set of educational materials of here's how to do something with this thing. And there's some overlap between inspiration and education like, I make a lot of Youtube videos that both showcase. Here's something you could build an AI agent to do. And here's how to do it.
So those are the Big 3. So from a product perspective.
we are not yet as intuitive as I would like to be. I think we're best in class, like I think we are the easiest to use product. And the reason I know that is because users tell us. I tried with this. I tried with that I tried with that, and I failed with all of them, but I succeeded with related app.
but I still spend, I spend, I do, 6 or 7 customer calls per day, and every customer call. I identify 5 or 6 things where I'm like, oh, shoot! That's still not easy enough. That's still not easy enough. That's still not easy enough, so we need to refine, refine, refine, refine, refine. Second.
we need to show people what's possible. So we have about. I think we have like 70 templates in our gallery right now. That needs to be thousands. There are thousands of things AI agents can do for people. So we want to build up a much bigger template gallery, mostly for inspiration, but also just to be able to import and get up and running quickly both by us and and by our community and our customers. And then. 3, rd I want to build a better educational curriculum. I'm doing these 1 h. Build with me live sessions every week. I'm recording a couple of Youtube videos every week. But
whether or not people end up using relayapp. The thing I need to communicate is
learning how to work with AI agents is not optional. You don't get to be skeptical about this one like you can. You can bury your head in the sand and pretend that AI agents are not going to have an impact on the workplace.
But I think that is a very poor career decision and business decision. So the way I'm framing it is AI agents are happening. They're underhyped. This is my hot take AI. Agents are way underhyped right now, like the world as a whole, does not understand how much work is going to change once we get better at creating and adapting these things.
And so it is a foundational skill that you know how to create and work with AI agents just like it was a foundational skill for the last 40 years to know Microsoft excel. You could not get any sort of job doing any knowledge work. If you were not basically functional in Excel, you couldn't be a marketer, you know, doing content calendars. You couldn't be a salesperson, creating a spreadsheet of leads. You couldn't run a cupcake. Bakery managing your inventory like Excel was a foundational skill for the last 40 years of work
creating AI agents. That work for you will be a foundational skill for the next, however, many years of work. And so that's what I want to communicate. And then I want to build a bunch of
educational resources that make that feel less scary and more like a fun opportunity to learn.
Collin Stewart (02:22):
Yeah.
I mean strong degree. I have AI agents and prompts, and Meta prompts that I use that run my business, and I feel like Superman sometimes, or iron man, where, like the things that I'm able to do
are incredible.
Compared to you know I can do. The other day I got off and I got off a client call, and I took the transcript. I fired it into one of my Meta prompts, and then I pushed that into another system. I have. I use Windsurf on my side, and I've got all my content. I got all my
and it like it wrote a full. Sdr sales playbook in like a minute. And I'm like this. I've done this, and it takes 3 months to do and customize all the different things and the pieces, and it's crazy, ugly, tedious work, and all that tedium was gone, and it was like I spent a lot of time on a really good, prompt, and I can reuse that. I'm like the pace.
Jacob Bank (02:27):
Just.
Collin Stewart (02:28):
Celebrates.
Jacob Bank (02:29):
It's it's it's honestly amazing. Let me tell you about a funny customer interaction I had earlier this week.
The customer works for a recruiting agency, and as part of their recruiting process they have a bunch of candidates and a bunch of clients, and they need to generate documents that indicate when a client is a good fit for a candidate, and they basically create a matching document which is like Here client, here's we found Colin for you. Here's why we think he's a good fit for your open role. Here's the key elements of his resume that makes sense to you, and it probably would have taken a recruiter
30 min to write this document. They had a template and they had a format. But you do have to read through the resume. You do have to match it to the Jd. And the clients, kind of location and culture, and whatever it might be. And so
this client, this customer of ours built an AI agent that automatically takes in all the information they have about the client, takes in all the information they have about the candidates, figures out the matches. And it writes these matching documents at like 95 to 99% of human quality. And she sent me a message that's like oof this one took like 700 AI credits like, is there a way I can reduce the cost of that like 700 sounds like a lot. And I was like first, st yes, here are ways that you can reduce the AI cost. You can send less data. You can use a cheaper model.
But 700 AI credits is like a dollar and 30 cents. And if that thing
5 h for a dollar and 30 cents, that's actually, I know it's it's self-serving for me to say that, but that's a pretty good deal, and.
Collin Stewart (02:35):
That's a great deal.
Jacob Bank (02:36):
Whoa, yeah, when you put it that way, a dollar 30 cents for all that stuff it did for me is a great deal.
Collin Stewart (02:37):
It's not even worth writing an email about a dollar 35 like, if you overcharge me for a dollar 35, I'd be like, yeah, whatever.
Jacob Bank (02:38):
Exactly.
I remember listening to this novel, Ravikant, podcast where he makes this provocative statement that every person should value their time at $5,000 an hour, even if your actual rate is far below that, it's just a good way
for you to frame in your mind that time is so so valuable. Time is so so valuable. And so when you use that frame, it's like Whoa! This saved me 30 min, and I'm only being charged a dollar and 30 cents like. What a huge, what a huge surplus! And that another example
that I was working on yesterday
was this was a Linkedin content research use case where a lot of people now it's become pretty natural to me. But even now I have my ais that are helping me come up with good ideas of content based on what I'm doing, based on what's interesting to other people.
What this Linkedin content research agent does is
it looks at the top 100 or 200 posts on Linkedin every week in aggregate. It identifies how many comments they got, how many engagements they got, what the hook was what the content theme was who the author was.
and it writes a 5 page report saying, like, Here's the kind of hook formats that are working well on Linkedin. Here are the visual aids that are working well on Linkedin. Here are the topics that seem to be resonant, and that's how I came up with the org chart idea like that. One of the things that was turned up in my AI report was like, Oh, it seems like people are interested in this like org chart about AI agent thing. Can you do something cool and unique about that? Imagine
how much you would have to pay an agency or a human
to create that report personalized to me because I don't want just generic content. I want things that's about AI agents and AI workflows and AI models.
What would it cost someone to produce that 5 page report every week? A 1,000 bucks a week like
more more way, more, 5,000 bucks a week like, yeah, I don't even know how much it would cost, but you know how much it costs me 40 cents like it's crazy.
Collin Stewart (02:50):
I had a similar moment the other day I was looking at a list. And I'm like, Okay, I there's 25,000 accounts here. I've got their Linkedin description
normally, you know, and it's a client with a really tricky ideal customer profile. I'm like, traditionally, what would happen is their Sdrs would go and take a look and go. Yep, good, fit, bad fit. And I was like, Okay, well, I can use it. I can use an Api call and run it across here, and I was like, Well, how much is it going to cost me if I use the bigger, more expensive model? And I was like, Okay, it's going to cost me $12 instead of $4. It's like $12, it is. And it looked at 25,000, and it gave me good fit, and the subclass, and I was like.
Jacob Bank (02:52):
Way we should think about it is that $12 and $4 are both $0, right? Like, compared to the value of of what's being generated here. $12 is such a ridiculous bargain.
Collin Stewart (02:53):
Totally. Yeah, I mean, especially if your time is worth $5,000 an hour.
Jacob Bank (02:54):
Yeah.
Collin Stewart (02:55):
I. There's a thing I want to ask you, and maybe we edit this out because it's a bit of a detour and like feel free to tell me to shut up, and I'm annoying.
But you're probably one of the few people I could have this conversation with, because I'm a user of a lot of products like yours. None quite like relay. I was an early Zapier user. I'm currently doing relay-esque things in Windsurf, because I like to have the control.
I use clay.com quite a bit. And it's similar, but very different. Right? It's very much like, Go to market focus.
Jacob Bank (02:58):
Yep.
Collin Stewart (02:59):
The thing that I hate I shouldn't say hate. But the thing that I'm getting I think I'm 1 of the more advanced users of clay. The thing I'm getting that I think could be solved is the manual click labor of setting up clay. The number of things and times I have to go and redo the same thing, and yes, you could. There's templates, and there's things here, but I'm like
I can speak to AI. I can speak to an agent. Why can't an AI agent go and set up clay for me right like? Why can't it go set up? And I saw a video of some guy who is using one of, I think, one of your competitors. I won't say the name, because I think they were in a similar space, and he put all the docs into
opus, and then said, Hey, output a Json file with the config. I tried this with Clay like a year ago, and stuff wasn't there yet. I'm like, so one. Are you working on this already? 2, if not, please, because, like, I'll be a relay power user in a heartbeat if I could do this, and I don't have to do all the manual click labor which feels like.
Yeah.
Jacob Bank (03:03):
So yes, we're working on it. It's actually a pretty. It's
counterintuitively a natural language experience. To
build a workflow or build an agent is harder than a natural language experience like lovable or bolt or V. 0, to build a website that was kind of counterintuitive to me at first, st but then, when I dug into the details, I could figure out why it's harder to build a workflow between knowing the exact capabilities of all the individual tools. And there's way, less training data of open source projects. And so I'm guessing the competitor of ours, you're referencing is N. 8 NN. 8 n. Is a great, a great product and a great company.
It's for a more technical audience than ours. They really lean into like self-hosting and open source. And you can input and output Json representations of your workflows, and therefore you can go to Chatgpt or Claude and say like, Give me a Json of an Nn workflow.
Collin Stewart (03:07):
-
Jacob Bank (03:08):
I don't know how well it'll work to actually plug it in. I'm sure some people have succeeded with it. I'm sure it takes a lot of prompt engineering.
But no, what we want to do is in the relay product. And and another thing I'll note is that pretty much every product in our space has some sort of natural language experience for creating automations.
The problem is what most of them do is you write in like.
I want to create a workflow, that whenever a new email comes in in, Gmail creates a row in a spreadsheet
and then it'll create a partially correct workflow that you don't understand, and
you don't know how to get it all the way to correct. Do you modify the natural language prompt? Do you modify the workflow in the sort of like the structured workflow builder. And so
I think this is the biggest opportunity for us in product activation. Can we create
the 1st really good natural language experience for coming up with an idea of what agent to build, figuring out how to structure that agent, actually creating the 1st draft of that agent iterating on that agent, as you test it, with real world situations, turning that agent on and giving you the oversight over that agent as it's working as you build confidence
can. If we can build the 1st like really good natural language system to do all of that.
that's what's gonna change the activation and and and bend the curve further.
Collin Stewart (03:18):
Yeah, I mean, you're basically building a product manager, because these like, there's all these users that are like, Hey, I want to do this thing. Users have a total misconception about what relay can do from like every different angle, and like.
Jacob Bank (03:19):
And and it's not just about relay. There's a problem for everyone which is
articulating clearly. What you want is not easy.
Yeah, it sounds. It's but like you'd be shocked at the number of customer support requests I get that are just like it's not working, or I want to connect Monday and Calendar.
It's like, Okay, you want to connect Monday and calendar.
Can you give me more information about how you want to connect Monday and calendar, which board and Monday which calendar and calendar do you want to create new events and calendar when a new item is created Monday? Do you want to do vice versa, do you want to update them? And so
one of the things that that we've realized is that the most important problem is not
converting the textual prompt
to the workflow. It's helping the user articulate what they want in a way that an AI can understand.
Collin Stewart (03:27):
Totally. I mean, don't you remember your 1st writing, your 1st specification that you handed to engineers? And then they went and built the thing and came back a week or a month later. And you're like this is not what I meant.
Jacob Bank (03:28):
I remember, I used to, you know, I was a computer science student. And then I used to teach introductory computer science. And I remember, at the beginning of every computer science class, the 1st thing most teachers or professors would do is they would do an exercise where they would stand at the front of the class, and they would say, Tell me how to throw this apple core away as an example, and people would say, Walk over to the trash can and throw the apple core away, and the professor wouldn't move.
Then they would say, Turn 90 degrees to your left, walk one pace forward and throw the apple core away, and then he would like, throw the apple core on the ground. Then they would say, Now turn 90 degrees, go one pace forward, extend your arm fully, release the apple core from your fingers into. And the point of that exercise was to show that
if you don't tell a computer exactly what to do in exactly the syntax that the computer understands, it will not be able to do the right thing. And that was always a part of introductory programming. Now Llms are cool in that.
You can speak to them way more naturally, like an Llm. You can say, please go throw this apple core away in a way that you couldn't that you couldn't say to like a standard computer programming language.
But if you want the Llm. To do something complex, for you
still need to articulate it somewhat carefully. And so I think that's a skill that we really need to help people build where, let's say, you're building
visit experience I've had many times.
People will go to Chatgpt, or they'll go to relay, and they'll write a prompt that's like
create a blog post based on this idea.
And then they're surprised when the blog post is generic. AI slop.
That's because all you told it was one sentence to write a generic blog post about an idea. What could it possibly do other than write generic AI slop? And so that's what it did. And so
part of my my goal in creating educational materials is not just. How do you build a great workflow? Or how do you build an agent. But how do you
think and communicate in a way that it will be easy for the AI to know what you want it to do?
Collin Stewart (03:41):
And that's not necessarily intuitive to help under help the user understand all the different models and the structure, and how things and the level of specificity that you need to get to to go from. I want to build an AI agent to. I am capable of communicating specifically the type of.
Jacob Bank (03:42):
Yeah. And and that's why I think a lot of the
this is going to be crazy to say. But I think a lot of the AI agent. Influencers are doing the community a disservice by creating things that are way too complex. Like, if you're on Linkedin, you'll see these diagrams with hundreds of nodes and dozens of prompts.
No one is going to get that working. No one is going to be able to figure out how to adapt that and get it working. And so I'm very happy when I get mocked in the other direction, because I post very simple use cases. And every time. There's someone in my comments saying, this isn't a real AI agent that's so simple. Couldn't you have just done that in Zapier with 5 clicks, or like I could have done this with one line of python I'm like.
go ahead, go ahead! Do that with one line of python, if if you want to my goal is to communicate broadly applicable, broadly accessible, useful stuff that people can actually get working and get value out of.
And so I kind of know. I know I'm in the right when I'm triggering. Some of the people like the more pedantic folks who are like, well, that's not technically an agent, or that's not really the best way to write this problem like it may not be the best, but it's gonna think it's gonna be the thing that works in a most, in the most understandable way for a less technical user than you are.
Collin Stewart (03:47):
Also define agent.
Jacob Bank (03:48):
Yeah, I mean, and this is the thing where I think people on Twitter and on Linkedin are like
totally missing the point.
There's all these debates about what is a workflow versus what is an agent? Is it goal, oriented tool calling, in which the agent determines the flow control? Or is it something else?
Nobody cares. Nobody cares about the technical difference between a workflow and an agent, and it's sort of a spectrum between them, because you can have a workflow or agent that has some more deterministic components. And then some components where the AI has more judgment of what tools to call, and how many times to call them.
Collin Stewart (03:52):
Here's what I think people care about.
Jacob Bank (03:53):
People care about is.
how am I going to use this AI stuff? And what is it going to do for me? And there are 3 basic modalities that matter number one. You're going to have a Chatbot. You probably are already using Chatgpt or Claude or Gemini. You're starting to build intuition about what a Chatbot is really good for. It can do research for you. It can brainstorm with you. It can be a thought partner and a sparring partner with you. It can kind of create one off text-based content. That's cool.
Second modality is a copilot. If you've used airtable, you've seen the AI assistant on the side. If you've used notion you've seen the AI assistant on the side. If you've used cursor or Windsurf, of course you've seen the AI copilot on the side. That means you are in a Sas tool, accomplishing something like writing a document or writing code, or making a presentation, and an AI, a clippy style. AI is your interactive assistant that's helping you do that.
Collin Stewart (03:56):
That's super useful in a bunch of contexts.
Jacob Bank (03:57):
And then the 3rd category is what I refer to colloquially as an AI agent, and that is not a Chatbot and not a co-pilot. It is an autonomous system that is doing work on your behalf behind the scenes.
and that's the definition of agent that I think matters to people which is like, I'm not going to talk to it in Chat Gpt. It's not my co-pilot on the side of my, you know, Google Docs or Microsoft word, but it is going to automatically run 30 min before every meeting. It's gonna go fetch the information about the person I'm meeting, and it's gonna send me a nice briefing notification over slack. That's what people should think of as an agent, even though what I just described to you is technically a workflow.
Collin Stewart (03:59):
Who cares? It's doing work for me.
Jacob Bank (04:00):
Exactly. That's the definition that matters to people.
Collin Stewart (04:01):
Totally. It's it's I mean, you summed it up so well. The actual definition doesn't matter. It's is this doing work for me?
Jacob Bank (04:02):
Right like. And the reason I'm so excited about AI agents.
I remember when my previous company timeful, was acquired by Google in 2015,
and the head of the division who acquired the company, and I had a really cool conversation about
how we thought productivity tools were in need for a refactor like, yeah, we'd had this email client and this calendar client and this document editor and this spreadsheet for the last 40 years. That was a paradigm invented by or 30 years invented by Microsoft in the mid eighties. That
is that the optimal way for productivity tools to help us get things done Calendar to do list contact manager. And so we were so excited about this idea of a refactor. But the time wasn't right for it. The time wasn't right. And so there was still another 10 years of like building better email clients and building better calendar clients and building better document editors.
And now I think the refactor is coming.
The refactor is coming in that. Yeah. Of course, we're still going to have an email client. But the email client's role in our life is going to be less central, the more central role will be the agent or agents that we have doing work over email on our behalf. And so I think in the future, we're not really going to care which Crm we pick or which email client we pick basically which which database we pick. We're going to care about which agents we build on top of it that do useful stuff.
Collin Stewart (04:09):
Because once the agents get to a certain level it won't matter what Crm you have if the agent can go and recreate your salesforce instance in Hubspot. You could just pay that.
Jacob Bank (04:10):
No. And and and I don't want to underplay the importance of really good data. Modeling and really good infrastructure like crms, are a very hard data modeling problem. They're a very hard infrastructure problem. There's a lot of customizability required. There are still going to be better crms and worse crms. I just don't think they're going to be judged by how easy it is to click the buttons to update the thing, because an agent's gonna be doing that for us.
Collin Stewart (04:11):
Totally. I I think the new ux is like no ux right to your point. It's it's not about what it looks like. It's what it can do for you, and not in the what it helps you accomplish.
Jacob Bank (04:12):
Here's the really cool thing when you describe it like that. The next ux is no ux like that sounds very abstract to a non technical person. But the really cool thing is, we all already have a mental model for this. We've all worked with people. We've all been an intern or hired an intern, or been a contractor, or hired a contractor, or been a boss, or had a boss, and so we know
we've all, you know, had a plumber come to our house and help us fix a leak, so we know how to articulate a problem that we have. I have this leak under my kitchen sink. We know how to communicate the success criteria of the leak being fixed, and that you're going to be paid some money after fixing it. And here's you know, some other information like Out. Here's the bathroom, if you need it, and grab a drink from the fridge, or whatever like. We all have experience
of delegating work to other people, and so we can. And yes, delegating work to an AI is going to be different in some ways, but we all have a ton of fundamental intuition about how to delegate work, and that's what's going to make this whole refactor possible, because we already have a mental model for it. We already work with other people.
Collin Stewart (04:15):
I love that.
I think that's a perfect place to land this episode, Jacob. Thank you so much for coming on and resharing the journey. And like keeping us updated. And absolutely, I'm going to put it in my calendar from a year from now we're going to redo this. We're gonna update the episode because.
Jacob Bank (04:17):
Hopefully, our current growth curve will look cute at that time.
I can't wait. Yeah, thanks for having me. This was a blast.
Collin Stewart (04:19):
Always I could talk for hours. But I want to be respectful of your time in the 7 other, you know. Customer calls. You gotta take today.
If people want to find out more about relay, what's the best place for them to look.
Jacob Bank (04:21):
Follow me on Linkedin and join one of our live building sessions. You can sign up at Eventsrelayapp.
Collin Stewart (04:22):
Awesome. I'm going to throw your Linkedin a post to your the Linkedin post that I saw, and then I've got the social Media starter pack the Social Media Research starter pack, because, as you were describing it, I'm like, that's what I need to do. I'm not doing that. I'm going to go do that after this call.
Jacob Bank (04:23):
Awesome. Well, let me know if you run into any issues, and I'm I'm here to help.
Collin Stewart (04:24):
Amazing. Thank you so much for.
Jacob Bank (04:25):
All right. Thanks, Colin. Talk to you soon.
welcome back to the predictable revenue. Podcast I'm your host. Collin. Colin. Wow! Let me try that again.
Jacob Bank (00:11):
Only done it a hundred times. Yeah.
Collin Stewart (00:12):
400 times. Yeah, yeah. All right, welcome back to the predictable revenue. Podcast I'm your host, Colin Stewart. Today, I'm rejoined by Jacob Bank from relay. And we are going to have a conversation. And we're going to have an update conversation about kind of what's what's happened. What's changed? Because we talked a year ago just over a year ago, and you were in.
Jacob Bank (00:13):
It's crazy that that was only a year ago. A lot has happened since then.
Collin Stewart (00:14):
And the whole. Thesis for the podcast in the repositioning around the founder stories was, I could have initial conversations and then wait until they take off a couple years later.
and then I saw a post of yours on Linkedin, and the customer love the metrics, the reach. I was like, man. We've got to have the conversation, because I'll admit when we 1st talked, I was like, I kind of get the product. It does a lot of really interesting things. I kind of want to do those interesting things. And I'm typically an early adopter of stuff like this.
I created an account, and I didn't do anything with it, because I'm like I there's so much here I don't get it.
But obviously that's changed. And so I want to talk about what's changed.
Jacob Bank (00:18):
Let's do it.
maybe the most valuable starting point would be. I can tell you about the 3 phases of the company we've gone through since we launched and what was going on in each phase.
So first, st for some context, relay dot app is an AI agent and workflow building platform. It is a horizontal tool which means
any company can use it for a wide variety of use cases. You could use it to extract information from your invoice emails. You could use it to automatically draft Linkedin and Twitter posts for you. You could use it to automatically analyze all of your customer calls and give your sales reps coaching. So it's a very broad, flexible platform, and horizontal products like this are hard to bring to market for 2 reasons at least number one
is, they're hard to market because you have to figure out a positioning and messaging that communicates who it's for in a way that they understand what it's for. And then, second, you need to find a channel by which you can reach them. It's not as simple as saying, well, our Icp is Hr leaders at technology companies between 501,000 people. And we're going to get this list of targets. And we're, gonna you know, cold email them
so for both a positioning and messaging side. And from a go to market Channel side, it can be harder. And then, second, it's harder from a product perspective, because you need to build a single tool that can serve a cupcake bakery all the way up to the customer success team of a fortune. 500 company.
both in terms of functional breadth, industry, breadth, and then breadth in the level of technical sophistication of the users. So that's kind of the context of like what the product is and what some of the challenges are. When we 1st launched in October 2023, the most approachable way. To bring a product like this to market was to anchor it against an existing market leader.
and the way we chose to do that was to position ourselves as the modern alternative to Zapier. Zapier is a well-known product. They're a multi-billion dollar company they are synonymous with. If this, then that automation for work, use cases for a broad audience. Funny enough if this, then that sort of should have been the leader in that space, because they invented the term, but they ended up focusing more on
connected home and consumer use cases. So Zapier, Zapier sort of inherited Iftt's crown in the 2015 to 2020 time. And we had seen this pattern work successfully for many other classes of Sas.
for example, linear is the modern alternative. To Jira. Audio is the modern alternative to salesforce loops is the modern alternative to Mailchimp, etc, etc. In every category where it felt like there was a 1215 year old, incumbent leader. There was an opportunity for a new player with a simpler ui, maybe more AI forward to gain, to gain, mind share.
So that's how we position ourselves in October of 2023. We're the modern Zapier alternative. So if you're the kind of company that picks linear, if you're the kind of company that picks adio because they're well designed, well crafted, really performant high quality for built for, for fast moving, strong teams who care more about
high quality end user experience than they care about enterprise check boxes. We were that kind of that kind of product.
And
that's when you know, back to our previous conversation. That's when we 1st exited the desert, and it started to feel like we had a little bit of product market fit. We got our 100 ish paying customers, we got our 1st dose of user love. Our retention started to be pretty good, like. So we were stacking. We were starting to stack cohorts. And I was like, Oh, yeah, this is. This is turning into a real business.
That was phase one October of 2023 to June 2024, phase 2.
We realize. Okay, we're now on the map as a Zapier alternative. We are getting some users that are zapier refugees because they're unhappy, for whatever reason, with the pricing or the robustness, or the quality or the interface, and there's.
Collin Stewart (00:34):
It looks anywhere.
Jacob Bank (00:35):
A steady trickle of those people over to us, but we felt like that was missing an opportunity in a world where there was such a huge tailwind of interest in AI. It felt like we were dramatically under utilizing that tailwind. So in June
we did a big relaunch that instead of framing ourselves as the modern alternative to Zapier, that differentiated primarily on usability. We became the AI 1st or AI native automation tool. And we built really really high quality. AI steps so that you could take your traditional workflow.
When a form is filled out automatically, add the contact to my Crm, you could add an AI powered lead, enrichment, and qualification that would work really really well and be very simple. To set up.
That led to another inflection point in the curve. It led to an inflection.
largely because retention and engagement went up further. Oh, my camera went out of focus there for a second. Alright!
There we go.
So when we did that, AI 1st launch, it led to another inflection point in the curve. Retention and engagement improved. User love improved.
But I still felt like there was too much potential energy locked up in the system in the sense that
we had 2 charts that would really tell the story. The 1st chart was our overall revenue growth and our active user growth. And that looked pretty good. It was not quite like the vertical line hockey stick, but it was like a very strong up into the right. But the number of new signups we got each week was pretty flat.
and so when you have those 2 in conjunction, it's a good sign and a bad sign. The good sign is, it means you have a high retention product that without a ton of top of funnel. You're still able to grow the business because your old users are sticking around and your new users are coming in. And so you're stacking cohorts on top of each other.
But that doesn't work forever. It also means
that you haven't. You haven't figured out how to break through and create a compounding top of funnel strategy that makes you top of mind for your customers at the right time.
That's when
I wrote a 25 page document for our team and for our investors at the end of the year in December, and the document covered a lot of ground. But the fundamental point of the document
was that I thought we were mispositioning ourselves. Our product was actually really good. I thought we were mispositioning ourselves in a way that was failing to take advantage of the AI agent, Tailwind.
and my claim was, nobody cares about no code automation that sounds robotic and nerdy, and a lot of work. But everyone wants AI agents to do work for them. And even though agents are actually a technical term, they've just become in the popular lexicon something people care about. I never would have predicted that agent was agent is a weird word. It's not what I would have predicted people would have cared about, but they do.
And so my claim was that by repositioning ourselves from the new standard for no code automation to the easiest way to build AI agents that work for you, we would be able to unlock a broader degree of awareness and mind share that would communicate to the true target audience of everyone who wants to use AI to help themselves that we existed. So I wrote this document. I shared it with investors on
like December 21st or December 20 second, something like that just before Christmas.
on December 23, rd I was like, Oh, let me put up a little trial balloon. Let me create a linkedin post. That uses this new framing of the product
and creates a cool gift that shows what this agent can do. So the Linkedin post was something like this is before these posts were all over the place which they are now. It was the post was I made an AI agent that automatically looks up Linkedin profiles for me. Here's how it works. You give it all the information you have about someone, their name, their email, address, their company their location, and it quickly and accurately finds their Linkedin profile. You can use this for recruiting, for sales, for meeting preparation. It's pretty cool. And then it was like literally like a 1 4 bullet post.
It had a gif that showed a Google spreadsheet with
(00:56):
information about the targets on the left, and then it would fill up with the Linkedin posts on the right, and it got 600,000 impressions, and like 7,000, and like 5,000 comments, and I had been posting regularly on Linkedin for the year prior to that, and I had never cracked 10,000 impressions and 100 comments. And so it's just like, Whoa, okay, something about
the framing of an AI agent doing work for you, something about the use case, something about the the corresponding asset, like something about that. And it was funny, because it was literally the 1st thing I tried. I'd written this long document with like, here's my hypothesis that this will resonate, and then, like it literally worked the 1st time, which has never happened in my career in any situation. It doesn't work, it doesn't work every time. And so
that's when we repositioned the product from the new standard for automation to build AI create AI agents that work for you. Our landing page conversion immediately from making that string change at the top of the website. I can't remember the exact numbers, but I think it went up from 7% to 10%,
taking one string at the top from landing page, from landing on the page to converting to signing up for an account like that is a crazy, that is a crazy impact from a 1, a one, a 1 tagline, a 1 tagline change.
And so that was an unlock for us for 2 reasons.
Number one. It was an unlock for us in positioning and messaging. We were no longer the no code automation tools for nerds that weren't quite happy enough. With Zapier we were the easiest and most accessible product for everyone who wants to create AI agents that work for them, and that is a much larger set of people. So that was the messaging breakthrough that also led to a Channel breakthrough
which for the 1st time we had content that people wanted to share on Linkedin.
And so my post started to do really well. But also other people started to post about us because no one
people didn't really post. Here's my cool automation that I built on Linkedin people love posting. I just built my 1st AI agent. It's really cool. Here's what it here's what it does for me. And so
we had experiment. I think we talked about this in in our previous conversation, but to quickly recap it, we had experimented with everything
warm intros from our network cold, outbound email interacting with relevant threads on Reddit partner marketing bottom of funnel SEO, we'd experimented with all of the the go to market channels that you programmatic SEO and content. And you know, editorial content. We'd experimented with all of the go to market channels that that would make sense for a product like ours, and we hadn't found the one that quite clicked.
And then organic Linkedin just clicked. And so
I went from in December. I had, like 4,000 followers, 4 or 5,000 followers, and now I have, I think, 47,000 followers.
I went from getting maybe 10,000 impressions a week to a few 100,000 impressions a week, and then, more importantly, a few 1,000 comments, or at least at least a few 100 on a good week, a few 1,000 comments, and it went from no one hearing about relay anywhere. To every day people would see a couple of posts on their linkedin feed, either from me or from someone else.
and then that has led to
the hockey stick trajectory over the last the last 4 months, because it felt to me like we're finally unlocking the potential energy that was stored in that product that was really high retention and engagement and people really love. And now we figured out how to communicate it to people.
Collin Stewart (01:12):
And you seem just from a you seem so much happier about the product, so much more relaxed than the like, that, like tightly bound up that we all have when we're like, we know our tool is great, but we haven't found the area for greatness yet. You seem like you've relaxed.
Jacob Bank (01:13):
Yeah, and it's
There's a few elements to it. On the one hand.
I always thought we had a great product, and my role within the company as the founder, and CEO is to figure out how to bring that great product to market. And I felt like I wasn't. I wasn't doing justice to the opportunity of of what the team had built. And then, second.
I felt uncomfortable because I was very new to marketing as a discipline I'd always been in product and technology and doing a new, hard thing that you're not experienced in is can be very demoralizing, especially with something like marketing when so many things don't work before you find the thing that does work.
And now it's pretty fun, because now it's it's
so funny how the product market switch flips overnight, like even 5 months ago, we had to beg
for users who had signed up to meet with us to give us feedback about the product. And now, if I offer that call to action in more than you know, 5% of we, we have like a qualification filter. And we offer the the opportunity to talk to a member of our team to like 5% of new users, I offer to 5% of user boom. My calendar is booked out, for, like as many slots I have, I have available. And so but it's just funny how that happens overnight from like I have to beg people to talk to me about this thing to. There's like way, more demand than I can possibly satisfy.
Collin Stewart (01:20):
Earlier. You said it sounded like you had unlocked the Channel like organic Linkedin as a as a channel, which absolutely it sounds like it's working for you. But you were using organic Linkedin before, and it wasn't working. It sounds like the big change, was the positioning, the repositioning of relay. And how it can have an impact on or like how you communicate it to users. That was.
Jacob Bank (01:21):
It was a combination of the repositioning of the product overall, and me identifying
a few topics of interest and post formats that are really resonant, and these are changing. So I had a few. In January I had a few very popular posts that were of the form. I built an AI agent to do. X. Now, those posts don't perform as well anymore, because everyone's doing. They're like, if you go to Linkedin, you'll just see like 50 posts. I built an AI agent to do this. I built that like that post format is now saturated. I got I got my 4 or 5 posts with hundreds of thousands of impressions each. And now that's saturated. Then
I realized there was a huge educational opportunity that
people were starting to recognize that building AI agents was going to be an important skill for the future of their careers, and between all the shiny product Demos, it wasn't clear how to actually get started, and so one of my most probably my most surprising Linkedin post ever was.
This was back in mid-february.
and I posted what I thought was going to be a very generic webinar announcement that would get, I don't know. 30 people interested 50 people interested.
and the post was.
if you're interested in learning more about AI agents. But you don't know where to start. I'm hosting a 1 h live session, and we'll build an AI agent together from start to finish, and you'll have a working thing by the end of it. Very simple. One slide Google slide poster, not well designed at all. Just like build an AI agent with me. Here's the date and time. Let me know if you're interested.
and that post got like 700,000 impressions and 6,000 comments, which those are like big numbers for Linkedin
And that shocked me because I was like, oh, I was just announcing a webinar. But again it tapped into this demand, where people are like, oh, I want to learn. And so I had a few other live session posts that performed really well. And now I do my webinars once a week. I don't promote them as much on Linkedin anymore. But I get, you know, a few 100 people every week, and we kind of have that channel really working now and then. I hadn't had a very successful post for a month or 2
about 6 weeks, and then
I stumbled across across this other post that I thought was pretty cool where a person had created a map of an org chart of AI agents.
but I didn't, really. I thought the idea was cool, but I didn't really like the execution of it. I was like, Oh, that's not real like. No one is like the way I know enough about the space to just like I can tell the way you set that up like. That's not how you're actually using them because it wouldn't. It just wouldn't work in that way. So I thought, Okay, I can use that. And it got a lot of that post had a lot of engagement, so I was like, Oh, I can use that sort of same theme, but I can counter position a little bit and say, like, Here are the 40 AI agents I actually use for marketing, and they're all real. They're all actually running.
And I made that post about 2 weeks ago.
and it got 1.5 million impressions and 16,000 comments, and I don't know for sure I haven't been able to validate this. I don't think that's a record for impressions, but I think 16,000 comments is the most comments ever on a Linkedin post, at least asking catch Bt. And trying to do research, I was unable to find evidence of another Linkedin post that has received more than 16,000 comments on it. And so again, I realized, like
within this content theme of people. Because people are
the way I'm kind of thinking about it is
there's this really shiny object called an AI agent, and people are kind of circling around it. They're like.
what would I use it for? How do I make one? What does it actually look like when people are using it well. And so I don't think any one post format can consistently go viral.
But if you can, if you can tap into this need that people are feeling to better understand this, this really powerful new concept of AI agents, and how they're going to use it, that has turned out to be something that we can durably get people excited about on Linkedin, and that's just my post. But I will say that even when
my posts were starting to do very well. In January, February I still had some anxiety because it still felt funnily enough, with the podcast name predictable revenue. When we were totally reliant on me to have a viral post that felt unpredictable and unreliable, and that felt like a brittle place to be, because there's so much about a viral post that is outside of your of your control.
(01:42):
Now, we're in a different situation for 2 reasons. One is.
we've built up enough of a baseline awareness and a base like between my follower count and the number of people have heard of relay. That individual posts don't have to go viral. They just have to attract the 2 to 500 right people who are like the target audience for that post. So my posts, even if they only hit 10,000 20,000 impressions and 100 comments. If those 100 comments are for the right people like I made a post about a specific class of SEO SEO ranking, tracking agents that you can build
if that catches the right. 200 social media like SEO and content marketing managers like that's perfect. That's the tam like. That's the audience for that post. And then, second, like, I mentioned.
other people are proud to show what they're doing with our product with AI agents. And so every day there are 3 or 4 posts that are that are not from me. And so now
I'm never comfortable. I'm never comfortable as a founder, right? You can never be comfortable as a founder. So a year ago, when we spoke, I felt like we had very strong retention, which is the foundation of any good business, and that we just hadn't figured out how to unlock top of funnel and get people in. Now, I feel like we have very strong retention.
We have very good top of funnel, but there's still a lot of drop off between those 2, because
there's quite a leap to go from.
I've heard of AI agents and relayapp, and they're cool to. I have actually set up a running agent successfully, and it's adding value to me. There's a lot that has to go right in between those 2. And that's the problem we need to solve now, and once we can unlock that we'll have like a further inflection in the curve, and we can talk again in a year.
Collin Stewart (01:50):
Amazing.
Yeah, it's it's such a cool way to think about it in terms of like you've got your retention in place. You've got your the untapped. There's more untapped potential that you're putting into the system.
Jacob Bank (01:52):
I have way. More leads in our Crm that have engaged with multiple my Linkedin posts than I possibly have time to contact.
Collin Stewart (01:53):
Totally.
Jacob Bank (01:54):
That's a good that's compared to where we were a year ago. That's a really good place to be.
Collin Stewart (01:55):
I like to think of product market fit on a continuum. It's not 0 or one. It's not binary, it's
strength, you know. It's it starts at 0, you get to one. And you're like, Hey, we've got something. And now you're at kind of like 7, you know, and you can probably get to 30 or 50.
Jacob Bank (01:57):
My definition was always like the 1st and most important definition for our class of product, which is a self. Serve bottom up. Plg, B. 2 B. Saas product was.
If the whole team disappeared for a month, would we have more users and more paying customers and more revenue than we had at the beginning of that month, because that means there is a sustainable, compounding engine that is working. Now that definition doesn't make sense for every kind of business, but for our kind of business. That means you've got something. If you don't do anything for a month, and you end up with more users and more customers than you had the month before, and that we had already reached when we last spoke a year ago.
but the number would have been only a little bit bigger.
Collin Stewart (02:00):
Like.
Jacob Bank (02:01):
It was like, yes, like our retention was good enough that we wouldn't have lost many people, and our top of funnel was good enough that we would have gained some people, but we would have if we all disappeared for a month we would have grown but grown slowly. But that's that's an important foundation, because it means you got something. You got something, and then you just have to bend the curve. And now we're at the point where, like the curves pretty good. And now I'm just looking at like, oh, if we if we do, if we improve this element of it, we bend it. If we improve this element, we bend it, and now it's cool, because
yes, it's a spectrum. But I think we've crossed from the 0 to the one. And once you've crossed from the 0 to the one, if you just get 1% better every week that starts to have the compounding impact that leads to like long term sustained hockey stick, growth.
Collin Stewart (02:03):
Yeah.
So what comes next for you like to to bend the curve even further.
Jacob Bank (02:05):
Our product. Strategy
is to be the easiest to use and most intuitive way for less technical people to get the value of AI agents and workflows working on their behalf.
So it's the kind of person who's just starting to play around with Chat Gpt, just starting to experience the magic. And they're like, this chat gpt thing is pretty cool when I'm typing to it. Can it do more for me when I'm sleeping like that person who has that thought like, Oh, could it do? Could it do more for me? I want to be there for that person. And say, let me teach you how this technology can do more for you while you're sleeping.
And that means that we need to build
the simplest and most intuitive product experience.
The best
set of inspirational materials of here's what you can do with this thing. And then the best set of educational materials of here's how to do something with this thing. And there's some overlap between inspiration and education like, I make a lot of Youtube videos that both showcase. Here's something you could build an AI agent to do. And here's how to do it.
So those are the Big 3. So from a product perspective.
we are not yet as intuitive as I would like to be. I think we're best in class, like I think we are the easiest to use product. And the reason I know that is because users tell us. I tried with this. I tried with that I tried with that, and I failed with all of them, but I succeeded with related app.
but I still spend, I spend, I do, 6 or 7 customer calls per day, and every customer call. I identify 5 or 6 things where I'm like, oh, shoot! That's still not easy enough. That's still not easy enough. That's still not easy enough, so we need to refine, refine, refine, refine, refine. Second.
we need to show people what's possible. So we have about. I think we have like 70 templates in our gallery right now. That needs to be thousands. There are thousands of things AI agents can do for people. So we want to build up a much bigger template gallery, mostly for inspiration, but also just to be able to import and get up and running quickly both by us and and by our community and our customers. And then. 3, rd I want to build a better educational curriculum. I'm doing these 1 h. Build with me live sessions every week. I'm recording a couple of Youtube videos every week. But
whether or not people end up using relayapp. The thing I need to communicate is
learning how to work with AI agents is not optional. You don't get to be skeptical about this one like you can. You can bury your head in the sand and pretend that AI agents are not going to have an impact on the workplace.
But I think that is a very poor career decision and business decision. So the way I'm framing it is AI agents are happening. They're underhyped. This is my hot take AI. Agents are way underhyped right now, like the world as a whole, does not understand how much work is going to change once we get better at creating and adapting these things.
And so it is a foundational skill that you know how to create and work with AI agents just like it was a foundational skill for the last 40 years to know Microsoft excel. You could not get any sort of job doing any knowledge work. If you were not basically functional in Excel, you couldn't be a marketer, you know, doing content calendars. You couldn't be a salesperson, creating a spreadsheet of leads. You couldn't run a cupcake. Bakery managing your inventory like Excel was a foundational skill for the last 40 years of work
creating AI agents. That work for you will be a foundational skill for the next, however, many years of work. And so that's what I want to communicate. And then I want to build a bunch of
educational resources that make that feel less scary and more like a fun opportunity to learn.
Collin Stewart (02:22):
Yeah.
I mean strong degree. I have AI agents and prompts, and Meta prompts that I use that run my business, and I feel like Superman sometimes, or iron man, where, like the things that I'm able to do
are incredible.
Compared to you know I can do. The other day I got off and I got off a client call, and I took the transcript. I fired it into one of my Meta prompts, and then I pushed that into another system. I have. I use Windsurf on my side, and I've got all my content. I got all my
and it like it wrote a full. Sdr sales playbook in like a minute. And I'm like this. I've done this, and it takes 3 months to do and customize all the different things and the pieces, and it's crazy, ugly, tedious work, and all that tedium was gone, and it was like I spent a lot of time on a really good, prompt, and I can reuse that. I'm like the pace.
Jacob Bank (02:27):
Just.
Collin Stewart (02:28):
Celebrates.
Jacob Bank (02:29):
It's it's it's honestly amazing. Let me tell you about a funny customer interaction I had earlier this week.
The customer works for a recruiting agency, and as part of their recruiting process they have a bunch of candidates and a bunch of clients, and they need to generate documents that indicate when a client is a good fit for a candidate, and they basically create a matching document which is like Here client, here's we found Colin for you. Here's why we think he's a good fit for your open role. Here's the key elements of his resume that makes sense to you, and it probably would have taken a recruiter
30 min to write this document. They had a template and they had a format. But you do have to read through the resume. You do have to match it to the Jd. And the clients, kind of location and culture, and whatever it might be. And so
this client, this customer of ours built an AI agent that automatically takes in all the information they have about the client, takes in all the information they have about the candidates, figures out the matches. And it writes these matching documents at like 95 to 99% of human quality. And she sent me a message that's like oof this one took like 700 AI credits like, is there a way I can reduce the cost of that like 700 sounds like a lot. And I was like first, st yes, here are ways that you can reduce the AI cost. You can send less data. You can use a cheaper model.
But 700 AI credits is like a dollar and 30 cents. And if that thing
5 h for a dollar and 30 cents, that's actually, I know it's it's self-serving for me to say that, but that's a pretty good deal, and.
Collin Stewart (02:35):
That's a great deal.
Jacob Bank (02:36):
Whoa, yeah, when you put it that way, a dollar 30 cents for all that stuff it did for me is a great deal.
Collin Stewart (02:37):
It's not even worth writing an email about a dollar 35 like, if you overcharge me for a dollar 35, I'd be like, yeah, whatever.
Jacob Bank (02:38):
Exactly.
I remember listening to this novel, Ravikant, podcast where he makes this provocative statement that every person should value their time at $5,000 an hour, even if your actual rate is far below that, it's just a good way
for you to frame in your mind that time is so so valuable. Time is so so valuable. And so when you use that frame, it's like Whoa! This saved me 30 min, and I'm only being charged a dollar and 30 cents like. What a huge, what a huge surplus! And that another example
that I was working on yesterday
was this was a Linkedin content research use case where a lot of people now it's become pretty natural to me. But even now I have my ais that are helping me come up with good ideas of content based on what I'm doing, based on what's interesting to other people.
What this Linkedin content research agent does is
it looks at the top 100 or 200 posts on Linkedin every week in aggregate. It identifies how many comments they got, how many engagements they got, what the hook was what the content theme was who the author was.
and it writes a 5 page report saying, like, Here's the kind of hook formats that are working well on Linkedin. Here are the visual aids that are working well on Linkedin. Here are the topics that seem to be resonant, and that's how I came up with the org chart idea like that. One of the things that was turned up in my AI report was like, Oh, it seems like people are interested in this like org chart about AI agent thing. Can you do something cool and unique about that? Imagine
how much you would have to pay an agency or a human
to create that report personalized to me because I don't want just generic content. I want things that's about AI agents and AI workflows and AI models.
What would it cost someone to produce that 5 page report every week? A 1,000 bucks a week like
more more way, more, 5,000 bucks a week like, yeah, I don't even know how much it would cost, but you know how much it costs me 40 cents like it's crazy.
Collin Stewart (02:50):
I had a similar moment the other day I was looking at a list. And I'm like, Okay, I there's 25,000 accounts here. I've got their Linkedin description
normally, you know, and it's a client with a really tricky ideal customer profile. I'm like, traditionally, what would happen is their Sdrs would go and take a look and go. Yep, good, fit, bad fit. And I was like, Okay, well, I can use it. I can use an Api call and run it across here, and I was like, Well, how much is it going to cost me if I use the bigger, more expensive model? And I was like, Okay, it's going to cost me $12 instead of $4. It's like $12, it is. And it looked at 25,000, and it gave me good fit, and the subclass, and I was like.
Jacob Bank (02:52):
Way we should think about it is that $12 and $4 are both $0, right? Like, compared to the value of of what's being generated here. $12 is such a ridiculous bargain.
Collin Stewart (02:53):
Totally. Yeah, I mean, especially if your time is worth $5,000 an hour.
Jacob Bank (02:54):
Yeah.
Collin Stewart (02:55):
I. There's a thing I want to ask you, and maybe we edit this out because it's a bit of a detour and like feel free to tell me to shut up, and I'm annoying.
But you're probably one of the few people I could have this conversation with, because I'm a user of a lot of products like yours. None quite like relay. I was an early Zapier user. I'm currently doing relay-esque things in Windsurf, because I like to have the control.
I use clay.com quite a bit. And it's similar, but very different. Right? It's very much like, Go to market focus.
Jacob Bank (02:58):
Yep.
Collin Stewart (02:59):
The thing that I hate I shouldn't say hate. But the thing that I'm getting I think I'm 1 of the more advanced users of clay. The thing I'm getting that I think could be solved is the manual click labor of setting up clay. The number of things and times I have to go and redo the same thing, and yes, you could. There's templates, and there's things here, but I'm like
I can speak to AI. I can speak to an agent. Why can't an AI agent go and set up clay for me right like? Why can't it go set up? And I saw a video of some guy who is using one of, I think, one of your competitors. I won't say the name, because I think they were in a similar space, and he put all the docs into
opus, and then said, Hey, output a Json file with the config. I tried this with Clay like a year ago, and stuff wasn't there yet. I'm like, so one. Are you working on this already? 2, if not, please, because, like, I'll be a relay power user in a heartbeat if I could do this, and I don't have to do all the manual click labor which feels like.
Yeah.
Jacob Bank (03:03):
So yes, we're working on it. It's actually a pretty. It's
counterintuitively a natural language experience. To
build a workflow or build an agent is harder than a natural language experience like lovable or bolt or V. 0, to build a website that was kind of counterintuitive to me at first, st but then, when I dug into the details, I could figure out why it's harder to build a workflow between knowing the exact capabilities of all the individual tools. And there's way, less training data of open source projects. And so I'm guessing the competitor of ours, you're referencing is N. 8 NN. 8 n. Is a great, a great product and a great company.
It's for a more technical audience than ours. They really lean into like self-hosting and open source. And you can input and output Json representations of your workflows, and therefore you can go to Chatgpt or Claude and say like, Give me a Json of an Nn workflow.
Collin Stewart (03:07):
-
Jacob Bank (03:08):
I don't know how well it'll work to actually plug it in. I'm sure some people have succeeded with it. I'm sure it takes a lot of prompt engineering.
But no, what we want to do is in the relay product. And and another thing I'll note is that pretty much every product in our space has some sort of natural language experience for creating automations.
The problem is what most of them do is you write in like.
I want to create a workflow, that whenever a new email comes in in, Gmail creates a row in a spreadsheet
and then it'll create a partially correct workflow that you don't understand, and
you don't know how to get it all the way to correct. Do you modify the natural language prompt? Do you modify the workflow in the sort of like the structured workflow builder. And so
I think this is the biggest opportunity for us in product activation. Can we create
the 1st really good natural language experience for coming up with an idea of what agent to build, figuring out how to structure that agent, actually creating the 1st draft of that agent iterating on that agent, as you test it, with real world situations, turning that agent on and giving you the oversight over that agent as it's working as you build confidence
can. If we can build the 1st like really good natural language system to do all of that.
that's what's gonna change the activation and and and bend the curve further.
Collin Stewart (03:18):
Yeah, I mean, you're basically building a product manager, because these like, there's all these users that are like, Hey, I want to do this thing. Users have a total misconception about what relay can do from like every different angle, and like.
Jacob Bank (03:19):
And and it's not just about relay. There's a problem for everyone which is
articulating clearly. What you want is not easy.
Yeah, it sounds. It's but like you'd be shocked at the number of customer support requests I get that are just like it's not working, or I want to connect Monday and Calendar.
It's like, Okay, you want to connect Monday and calendar.
Can you give me more information about how you want to connect Monday and calendar, which board and Monday which calendar and calendar do you want to create new events and calendar when a new item is created Monday? Do you want to do vice versa, do you want to update them? And so
one of the things that that we've realized is that the most important problem is not
converting the textual prompt
to the workflow. It's helping the user articulate what they want in a way that an AI can understand.
Collin Stewart (03:27):
Totally. I mean, don't you remember your 1st writing, your 1st specification that you handed to engineers? And then they went and built the thing and came back a week or a month later. And you're like this is not what I meant.
Jacob Bank (03:28):
I remember, I used to, you know, I was a computer science student. And then I used to teach introductory computer science. And I remember, at the beginning of every computer science class, the 1st thing most teachers or professors would do is they would do an exercise where they would stand at the front of the class, and they would say, Tell me how to throw this apple core away as an example, and people would say, Walk over to the trash can and throw the apple core away, and the professor wouldn't move.
Then they would say, Turn 90 degrees to your left, walk one pace forward and throw the apple core away, and then he would like, throw the apple core on the ground. Then they would say, Now turn 90 degrees, go one pace forward, extend your arm fully, release the apple core from your fingers into. And the point of that exercise was to show that
if you don't tell a computer exactly what to do in exactly the syntax that the computer understands, it will not be able to do the right thing. And that was always a part of introductory programming. Now Llms are cool in that.
You can speak to them way more naturally, like an Llm. You can say, please go throw this apple core away in a way that you couldn't that you couldn't say to like a standard computer programming language.
But if you want the Llm. To do something complex, for you
still need to articulate it somewhat carefully. And so I think that's a skill that we really need to help people build where, let's say, you're building
visit experience I've had many times.
People will go to Chatgpt, or they'll go to relay, and they'll write a prompt that's like
create a blog post based on this idea.
And then they're surprised when the blog post is generic. AI slop.
That's because all you told it was one sentence to write a generic blog post about an idea. What could it possibly do other than write generic AI slop? And so that's what it did. And so
part of my my goal in creating educational materials is not just. How do you build a great workflow? Or how do you build an agent. But how do you
think and communicate in a way that it will be easy for the AI to know what you want it to do?
Collin Stewart (03:41):
And that's not necessarily intuitive to help under help the user understand all the different models and the structure, and how things and the level of specificity that you need to get to to go from. I want to build an AI agent to. I am capable of communicating specifically the type of.
Jacob Bank (03:42):
Yeah. And and that's why I think a lot of the
this is going to be crazy to say. But I think a lot of the AI agent. Influencers are doing the community a disservice by creating things that are way too complex. Like, if you're on Linkedin, you'll see these diagrams with hundreds of nodes and dozens of prompts.
No one is going to get that working. No one is going to be able to figure out how to adapt that and get it working. And so I'm very happy when I get mocked in the other direction, because I post very simple use cases. And every time. There's someone in my comments saying, this isn't a real AI agent that's so simple. Couldn't you have just done that in Zapier with 5 clicks, or like I could have done this with one line of python I'm like.
go ahead, go ahead! Do that with one line of python, if if you want to my goal is to communicate broadly applicable, broadly accessible, useful stuff that people can actually get working and get value out of.
And so I kind of know. I know I'm in the right when I'm triggering. Some of the people like the more pedantic folks who are like, well, that's not technically an agent, or that's not really the best way to write this problem like it may not be the best, but it's gonna think it's gonna be the thing that works in a most, in the most understandable way for a less technical user than you are.
Collin Stewart (03:47):
Also define agent.
Jacob Bank (03:48):
Yeah, I mean, and this is the thing where I think people on Twitter and on Linkedin are like
totally missing the point.
There's all these debates about what is a workflow versus what is an agent? Is it goal, oriented tool calling, in which the agent determines the flow control? Or is it something else?
Nobody cares. Nobody cares about the technical difference between a workflow and an agent, and it's sort of a spectrum between them, because you can have a workflow or agent that has some more deterministic components. And then some components where the AI has more judgment of what tools to call, and how many times to call them.
Collin Stewart (03:52):
Here's what I think people care about.
Jacob Bank (03:53):
People care about is.
how am I going to use this AI stuff? And what is it going to do for me? And there are 3 basic modalities that matter number one. You're going to have a Chatbot. You probably are already using Chatgpt or Claude or Gemini. You're starting to build intuition about what a Chatbot is really good for. It can do research for you. It can brainstorm with you. It can be a thought partner and a sparring partner with you. It can kind of create one off text-based content. That's cool.
Second modality is a copilot. If you've used airtable, you've seen the AI assistant on the side. If you've used notion you've seen the AI assistant on the side. If you've used cursor or Windsurf, of course you've seen the AI copilot on the side. That means you are in a Sas tool, accomplishing something like writing a document or writing code, or making a presentation, and an AI, a clippy style. AI is your interactive assistant that's helping you do that.
Collin Stewart (03:56):
That's super useful in a bunch of contexts.
Jacob Bank (03:57):
And then the 3rd category is what I refer to colloquially as an AI agent, and that is not a Chatbot and not a co-pilot. It is an autonomous system that is doing work on your behalf behind the scenes.
and that's the definition of agent that I think matters to people which is like, I'm not going to talk to it in Chat Gpt. It's not my co-pilot on the side of my, you know, Google Docs or Microsoft word, but it is going to automatically run 30 min before every meeting. It's gonna go fetch the information about the person I'm meeting, and it's gonna send me a nice briefing notification over slack. That's what people should think of as an agent, even though what I just described to you is technically a workflow.
Collin Stewart (03:59):
Who cares? It's doing work for me.
Jacob Bank (04:00):
Exactly. That's the definition that matters to people.
Collin Stewart (04:01):
Totally. It's it's I mean, you summed it up so well. The actual definition doesn't matter. It's is this doing work for me?
Jacob Bank (04:02):
Right like. And the reason I'm so excited about AI agents.
I remember when my previous company timeful, was acquired by Google in 2015,
and the head of the division who acquired the company, and I had a really cool conversation about
how we thought productivity tools were in need for a refactor like, yeah, we'd had this email client and this calendar client and this document editor and this spreadsheet for the last 40 years. That was a paradigm invented by or 30 years invented by Microsoft in the mid eighties. That
is that the optimal way for productivity tools to help us get things done Calendar to do list contact manager. And so we were so excited about this idea of a refactor. But the time wasn't right for it. The time wasn't right. And so there was still another 10 years of like building better email clients and building better calendar clients and building better document editors.
And now I think the refactor is coming.
The refactor is coming in that. Yeah. Of course, we're still going to have an email client. But the email client's role in our life is going to be less central, the more central role will be the agent or agents that we have doing work over email on our behalf. And so I think in the future, we're not really going to care which Crm we pick or which email client we pick basically which which database we pick. We're going to care about which agents we build on top of it that do useful stuff.
Collin Stewart (04:09):
Because once the agents get to a certain level it won't matter what Crm you have if the agent can go and recreate your salesforce instance in Hubspot. You could just pay that.
Jacob Bank (04:10):
No. And and and I don't want to underplay the importance of really good data. Modeling and really good infrastructure like crms, are a very hard data modeling problem. They're a very hard infrastructure problem. There's a lot of customizability required. There are still going to be better crms and worse crms. I just don't think they're going to be judged by how easy it is to click the buttons to update the thing, because an agent's gonna be doing that for us.
Collin Stewart (04:11):
Totally. I I think the new ux is like no ux right to your point. It's it's not about what it looks like. It's what it can do for you, and not in the what it helps you accomplish.
Jacob Bank (04:12):
Here's the really cool thing when you describe it like that. The next ux is no ux like that sounds very abstract to a non technical person. But the really cool thing is, we all already have a mental model for this. We've all worked with people. We've all been an intern or hired an intern, or been a contractor, or hired a contractor, or been a boss, or had a boss, and so we know
we've all, you know, had a plumber come to our house and help us fix a leak, so we know how to articulate a problem that we have. I have this leak under my kitchen sink. We know how to communicate the success criteria of the leak being fixed, and that you're going to be paid some money after fixing it. And here's you know, some other information like Out. Here's the bathroom, if you need it, and grab a drink from the fridge, or whatever like. We all have experience
of delegating work to other people, and so we can. And yes, delegating work to an AI is going to be different in some ways, but we all have a ton of fundamental intuition about how to delegate work, and that's what's going to make this whole refactor possible, because we already have a mental model for it. We already work with other people.
Collin Stewart (04:15):
I love that.
I think that's a perfect place to land this episode, Jacob. Thank you so much for coming on and resharing the journey. And like keeping us updated. And absolutely, I'm going to put it in my calendar from a year from now we're going to redo this. We're gonna update the episode because.
Jacob Bank (04:17):
Hopefully, our current growth curve will look cute at that time.
I can't wait. Yeah, thanks for having me. This was a blast.
Collin Stewart (04:19):
Always I could talk for hours. But I want to be respectful of your time in the 7 other, you know. Customer calls. You gotta take today.
If people want to find out more about relay, what's the best place for them to look.
Jacob Bank (04:21):
Follow me on Linkedin and join one of our live building sessions. You can sign up at Eventsrelayapp.
Collin Stewart (04:22):
Awesome. I'm going to throw your Linkedin a post to your the Linkedin post that I saw, and then I've got the social Media starter pack the Social Media Research starter pack, because, as you were describing it, I'm like, that's what I need to do. I'm not doing that. I'm going to go do that after this call.
Jacob Bank (04:23):
Awesome. Well, let me know if you run into any issues, and I'm I'm here to help.
Collin Stewart (04:24):
Amazing. Thank you so much for.
Jacob Bank (04:25):
All right. Thanks, Colin. Talk to you soon.
Popular Podcasts
United States of Kennedy
United States of Kennedy is a podcast about our cultural fascination with the Kennedy dynasty. Every week, hosts Lyra Smith and George Civeris go into one aspect of the Kennedy story.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com