Ernest Koe (00:01.191)
Hey guys, welcome back to another episode of the pot contacts. Let's do it again. Hey everybody, welcome back to another episode of the Contacts Podcast. I almost totally flopped that one. yeah, today we've got a few things that we want to cover. I think. I think there was this post by Addie Osmani, that I think is really relevant to the conversation today about you know how AI is getting adopted and whether the psychosis is over. I think

Todd Geist (00:08.814)
Yeah.

Ernest Koe (00:30.377)
That's important question. also we did have the webinar from from Claris yesterday on on sort of an early sneak peek on where they think Agentic development is going in FileMaker. we should touch based on that because it applies to things related to what we're doing.

Todd Geist (00:47.756)
It was also, we should say, it was done by ex podcast ex the contest podcast host, well, who used to be our very own Martha Zink, who now is is at Claris. So have to definitely definitely talk about that. Yeah.

Ernest Koe (00:51.672)
yeah.

Ernest Koe (00:55.763)
Yes.

Ernest Koe (01:01.201)
It was Martha Zink doing that webinar which was really quite exciting and well received, but we'll get into that a little bit. and yeah, I think we have some thoughts around agents and maybe kind of reacting to some of the conversations that are happening around us. whether they're real, whether they'll ever see the light of day beyond coding, you know, who knows? That's that's anything else, Todd, that we wanna get into?

Todd Geist (01:26.19)
I don't think so. Let's just get into it. let's see, where to start? I think maybe one thing just I think we're I think it's maybe to set the stage, like in the Claris world, we're just really getting going on being able to to work with agents with FileMaker apps. And so that's a that's a very new thing, with stuff like Proofkit and other things that we've been doing and with Claris's announcement. and so it's still

It's still fresh and brand new and and and very new stuff that we can do. The rest of the world has had a good six months now to really grind on tokens. And you hear things like token maxing and productivity maxing and AI psychosis and all these things that have kind of been through the zeitgeist, if you will. And we're starting to get some early returns on you know what's happening with that. And I think,

It it is still it's this is I mean, we're still talking about this latest wave is really still less than six months old. So this is all very, very new. But we definitely are seeing, you know, early hype cycles. I won't necessarily say peter out, but there's definitely there's been sort of a reconnection with reality in at least some areas. and it's c I think it's clear that the impacts of AI

will move through different in different industries at different paces, which I suppose is no surprise, right? Because that's the always the way technology rolls out, right? Each industries have different level of preparedness or susceptibility to a new technology wave. And so you know, different industries will will adopt faster than other industries. So so I think you know, we're kind of in in in in in in the claris world it's very new.

Yeah, outside the Claris world, we've had six months and some of the early results are in and people are generating I think some interesting ideas and thoughts around how humans fit in to the whole story and you know what what might be the broader use case for agents outside of say coding, so for example. So I think that's where we are. we're seeing things like the the Uber

Ernest Koe (03:30.739)
Mm-hmm.

Todd Geist (03:51.156)
COO reported that they weren't exactly clear that the token maxing they were doing, which is this phase that companies were doing just a couple months ago of basically getting their employees to spend as many tokens as they possibly could, they weren't exactly clear that they were producing any productivity gains or any real value for all the money that that that they were spending. I think part of this is because it's hard to measure, right? And that that's part of the reason why it's difficult to know.

And also part of this I think token maxing is dumb. Like that's like you know, like tet you know, having leaderboards where people were like, you know, basically judged on how many tokens they were burning is just not a very good idea. that that

Some places we're we're promoting.

Ernest Koe (04:39.4)
To be Yeah, to me to be to be clear, I I don't I've never that idea has never made sense to me. I mean I we know early a few months ago we kept hearing if you are not just if if your senior engineer is not like burning three times their their salaries and tokens, then they're kinda worthless. Like I have like, okay, how does that work? Right. So your cost

your your operational op acts goes up three X. Well, not three X, four X or whatever it is now. And your revenue does what? So I mean like unl unless you're burning free money from somebody, you know, like how are you ever going to sustain that path when it you know, productivity is one thing, but productivity is basically constrained by your your cost budget. And so if you if that productivity doesn't translate to actual new revenue,

Todd Geist (05:11.906)
Yeah. Forex, yeah.

Yeah, exactly. I mean

Yeah.

Ernest Koe (05:37.58)
where where are you going to go with y with all that quote unquote productivity? You know, that's I I that never made sense to me. I mean I I yeah.

Todd Geist (05:44.952)
Yeah, exactly. Yeah. Yeah, I think I think I think the i it's just you know, it's it's I I the part of it is is that it's really hard to connect the dollar spent to the outcome right now. Like the tooling doesn't exist for that, right? Like it's a big challenge. We we've talked about like how how do we track the the tokens we're spending on particular services project. It's it's really quite hard. that tooling just doesn't really exist.

Ernest Koe (05:59.123)
Sure.

Todd Geist (06:13.132)
So that'll come along and make some of these things a little bit easier easier to test. But I think one thing that's clear is that just because you can you can generate three, five, ten times more code doesn't mean you have three, five or ten times more good ideas, right? That are that are worth that are worth that are worth doing.

Ernest Koe (06:29.5)
Well well well

Th that is absolutely true. you know, maybe l let me sort of ground this in in in a world that may be closer to where we live, right? In sort of the the business and enterprise world. And especially in agency and and sort of sort of the consulting business. so we we've had this anxiety for a while, I think even at proof where, you know, it's clear that productivity has gone up, quality has gone up.

you can you can you can quibble whether it's not just faster faster, it's also better or w whatever wherever that that you know the the benchmark lies in terms of what HN decoding is doing for us. But but I can say in very simple terms that things that used to take us a week to do as a senior developer or pr engineer, we could probably execute that work to that same level of quality in

a a you know, in in a fraction of that time. You know, I'll call it maybe twenty five percent, fifty percent. I I don't know. But it's it's it's at least tw two X, you know, for equivalent output. the tooling's better, the agents can just do it. Like, you know, like like take a simple simple example. Like building a custom Kanban board. Kanban board. The artisanal way takes five days. Today, probably three hours tops, end to end testing and all that, right?

simple simple use case. So there's no question that it's faster and probably better if you can vet the code and make sure everything's good. The problem is what happens when all of that stuff gets done at the same velocity and the same level of quality, your backlog hasn't increased by four X to compensate, you know? So So what do you do? You know? So

Todd Geist (08:21.986)
Right.

Ernest Koe (08:28.431)
You know, and and I I and I think so that's one trajectory, right? That that that that's that's one problem. That yeah, you can do it faster, but it's not clear that you have enough customer driven demand to make up for you know, the work that you've just done faster.

Todd Geist (08:49.4)
Yeah. You may you may just not have anything else that needs to get done. Right. And I think that's definitely one one one thing. I do however think that there's it's also we're bottlenecked by by creativity in in a lot of ways.

Ernest Koe (09:08.677)
bo Yeah.

Todd Geist (09:10.274)
You know what I mean? Like like it still takes a human to put attention on things to get things done well, right?

Ernest Koe (09:13.521)
Well well this is well this is I think this is really important. And and my my analogy there earlier was that it it's possible that things are getting faster, things are getting better with AI coding. But it doesn't translate directly to revenue because there are other bottlenecks in place that w that that affect your pipeline, affect your business, affect affect the decisions that you make in terms of what to build and how to build it, right? All those things are still constraints.

Like you can b sure, you can build something, a feature faster. So what? Does it get adopted? The customers like it? Is it the right thing? I mean all those things require meetings, require time, require analysis, require require, you know, buy in, require socialization. And those things haven't gotten faster. In fact they may have gotten slower.

Todd Geist (10:01.346)
No, they they can't because it's because that's that's all a very human thing. And I think part of the the challenge is like that this was actually a quote or or a paraphrase of a quote by by the by the Dax who's who's one of the open code founders. You can find him on Twitter. And one of the things that he said was, We've shipped a lot of things that we shouldn't have shipped. Like there's just a lot of stuff in here that's that doesn't make sense. Like it's so easy to add a feature to something.

Ernest Koe (10:05.415)
It's all very human.

Todd Geist (10:31.18)
Now. And that used to, when you had a feature and you worked hard to get a feature over the line and ready to go to production, you shipped it because you you all it had already passed through a whole bunch of vetting and and it also took a lot of effort and you needed to get something back for that effort. So you shipped it. and that that might have made sense, you know, a few years ago, five years ago, ten years ago. But it it it just doesn't anymore because you can make a feature very, very quickly. And so the much harder problem of figuring out like

Does this feature fit with the rest of the features in the product? Does it make the product better or worse? You know, the it is not, it is not a fundamental truth that a new feature makes always makes a product better. That is not true. It's usually the opposite, at least beyond some certain thing, right? Like, like, like the Dak said something else was kind of interesting. He said, he said, your product, whatever it is, probably only has one really, really killer feature.

Ernest Koe (11:15.065)
In fact, I think it's usually the opposite.

Todd Geist (11:31.724)
And if you don't know what that is and get people to into that, then you're you're gonna have lots of problems. So everything that you add outside of that is noise, right? I think we're all experiencing that. We're seeing tools like even like dev tools, like there's a tool called portless, which is kind of it's a way to to spin up dev servers without using a lot of ports, you know, the the stuff that comes after the dot com in the URL. It's the it it's very useful.

However, it's it should be very simple and it's not. It's actually very, very complex. If you pull up the help or the CLI, it's like really long. There's all these options that you now have to understand how to use it. And my guess is that they that they could shave half or more of those features out and make a better product if they just focused on what was actually useful and not every single thing that they could put in there. Right. It's like that kind of stuff.

Ernest Koe (12:27.781)
Yeah. I so I think I go back to s some earlier thoughts we've had, at least thesis we w we've had about this, right? I I think that from a programming and coding and and software development point of view, it is compressing some of the engineering work. And and effectively so. I mean there is real gains on that front. Like we can certainly prototype faster, we can do we can take more swings at the bat, we can have more ideas that we can validate and verify. All those things are true.

I think it can it lead to a better product because of that? Absolutely. I I I mean I it you know, that that I can do three versions of something and and and and run it through some Asian framework and swarm to do some testing and and get some feedback and share it with people, that's all great, right? But it doesn't necessarily speed up the business process and the business it doesn't compress the business timeline for things to to

to make its way through society or to to a human workspace. I think that is the thing that's still very, very hard and still very, very important to to get right, you know. I go ahead.

Todd Geist (13:41.698)
Well I was just gonna say, yeah. So I mean there's there's you know, there's a so there's a bunch of like businessy things as to why

Just being more productive doesn't lead automatically to revenue gains, right? Or the or to better products. Like there's still a a huge process that goes into it. that seems very, very clear. And I think people are discovering that more and more. And, you know, just because you can ship something doesn't mean you should. so I I think there's other things too that have to do with like there's a limit to what to that process that we talked about, like figuring out.

what is the right thing to build and and actually like also like

There's a so there and then there's I I I think there's something that's what am I trying to get to? I'm trying to get to there's something about the attention that's required to run a project, to run a software project is in some ways going up because you have to instead of getting to spend some time sort of doing kind of almost rote tasks that you kind of almost helped you think through the problem by coding by hand, right? You now have to.

basically front load a lot of that into a detailed spec and a detailed architecture design that you then give to the agents to then go do it faster than you could. The problem is is that you still you still have to do all that attention stuff up front. And that takes time and that takes skill. and so there's actually in some ways I've heard people say, I'm thinking I'm actually thinking harder than I've ever had to think before about what I'm doing.

Todd Geist (15:22.902)
'Cause it just takes almost like more intention to drive the agents towards the goal that you want them to, rather than you just noodling around and kind of arriving at at the thing almost through a combination of of engineering and craft. You do you know what I mean?

Ernest Koe (15:41.295)
Hundred percent agree. I I mean I feel it very viscerally and personally. I mean I I think that there was a time where I felt like I had just unlocked some high bandwidth mental capacity in my head and I could do so much more because I had several agents running doing things and and it felt really unstoppable. I can say very unequivocally that that was not sustainable. Like I you know

Todd Geist (16:07.596)
No, that's right. Yeah.

Ernest Koe (16:08.442)
I mean, the other day I was pinging you like, Hey Todd, did we did we do a podcast on Friday? And last week. Yes. I was in I I was literally losing like this this is a true thing. Like I I for a moment like sort of lost track of the things I was holding in my head of what happened. And and I I I really think that I I I can't do that much anymore, you know.

Todd Geist (16:12.246)
It's right. Like it's like yes, and you were in it. It's right. Yes, you were you were yeah.

Ernest Koe (16:37.892)
don't run multiple agents now these days as much. I I two or three tabs at most if if I'm doing something. And it's and and and the other thing is as as as complexity increases because you're doing more, I'm finding that it's harder for me to actually keep track of what's happened. Yeah. So like thank God for commit logs, thank God for everything. So I can't I can't stop and come back and go, what what happened here? So I I just kind of fast forward, right? So if

Todd Geist (16:57.784)
Yeah. Exactly. Yeah.

Ernest Koe (17:07.512)
If you apply that take that to the logical conclusion, right? I I don't I think I think it's it's it's practically impossible. Maybe others are much better. It isn't I I just have, you know, the lim my attention limitations are just just mine and it doesn't it's it doesn't it doesn't apply universally. But if it does then I think that we are there are certainly human limits here that prevent this from actually being effective at certain scale.

Todd Geist (17:38.092)
Yeah, I I there I think there are, but I also I have a kind of interesting anecdote. I'm gonna reach

Todd Geist (17:45.932)
I I think is something that we're gonna get to in this space at some point. So I've been learning to pl to play the yuke, which means I'm learning to play chords, you know. And so this is a very physical thing that I'm doing, right? Like I have to learn to move my fingers in a certain way and I'm getting a sound out of it. And, you know, obviously I'm playing music and I'm trying to learn how to play music. But one of the things that I I was at a I was at a

a performance a couple weeks ago well, maybe a week or so ago. And I've been going to music shows my whole life. Like like like concerts were a very big part of my life growing up. And I mean, w we had a music store when I was a kid. Like I mean, we had a cop we had a concert hall when I was a kid. My my stepfather was a disc jockey, a folk music disc jockey. So music has been a big deal. But I never learned to play anything in terms of actually physically learning.

To play chords, right? So I had this really interesting experience where I was watching somebody play and I was watching their hands movement, their hands move, and I experienced the music in a different way. What I felt was a richer experience. I understood the music better because I understood how my I could feel it in my body in a way that I could not. There's a connection to it.

Ernest Koe (18:46.854)
Sure, yeah.

Ernest Koe (18:59.078)
Mm.

Ernest Koe (19:09.262)
There's a connection. Yeah.

Todd Geist (19:12.48)
And part of what's happening with agents is we are losing connection to the code because we're not typing it. Right. So we are having to work at a higher level abstraction. And there's no question that there are, like we've been saying, there's no question that there are that there are gains. And like, I don't believe we're going back. Like I don't believe we're going back to an experience where we're typing on keyboard. But I don't think this sort of flat, high level everything in your brain and in a markdown file.

is where we end because there's other ways that humans process information phys you know, through hand movements, f especially hand movements. Really, really big part of what it is to be a human is to work with your hands. And so the keyboard, we're kind of we're kind of losing some of that. But I wonder if there's new ways in which we can learn to experience what we're doing in a way that will enrich the experience and improve our attention. This is what I was getting at was

You know, I'm just not sure our attention is designed well for this sort of flat experience that we're having, a bunch of markdown files and a whole wall and walls of text and diffs and on a screen. I don't think that's enough. And that's why we're feeling this kind of attention constraints that you were just describing. Things like that.

Ernest Koe (20:29.081)
I think that could be it. You know, or we are still learning a different how to how to connect with the at a different level of abstraction that we haven't fully figured out yet, you know? Like because I I I actually I don't f I'll give you example. you and Eric love using Whisper and you know, to to talk to your agents. I can't do it because I I feel disconnected when I don't type. Right. So for me Yeah. Right.

Todd Geist (20:38.638)
Yeah, yeah.

Todd Geist (20:48.14)
Yeah, yeah.

Todd Geist (20:55.958)
It's yeah, I'm the opposite. Typing never worked for me. Yeah. Yeah.

Ernest Koe (20:59.583)
And that was that was my way of connecting to the craft. Like when I when I I still write freehand most of my blog posts or or things like that, because a I hate the slop that comes out without without real prompting guidance. And I feel like by the time I write the prompt I've written the whole thing anyway. So I'm just the right write the thing. You know, but but I I get you. I I think I think there is something about craft that connects us deeply to our work that

Todd Geist (21:14.716)
Ha

Ernest Koe (21:28.431)
we are stru that's that we may need to that we're maybe relearning and trying to find a new way to to to get reconnected to it, which you know.

Todd Geist (21:36.99)
Right. And I was so I was struck by I found actually I I need to go back and find this again. so there are there are new code visualizers that people are playing around with. I think because for this very reason they're looking for a way to to reconnect with their code. And one of them that I found was pretty interesting. It'll make a three-dimensional heat map of your code base. And the the people that were doing this said one of the things that they do is they actually print.

Ernest Koe (22:06.703)
Mm.

Todd Geist (22:06.776)
They 3D print this map of their code base. It looks like a city landscape. And, you know, different heights or different complexities and different modules or different colors, et cetera. And so they actually print it so they can actually experience it three-dimensionally physically. And they feel that that's that that's helpful to them into understanding the code base. Which I think is pretty interesting, right? Like it may just this may just be a bunch of copium trying to get.

Ernest Koe (22:17.06)
Mm-hmm.

Ernest Koe (22:30.013)
Ha ha.

Todd Geist (22:35.626)
at this, but I do think there's something about about maybe just attention. And that's actually what what Addie Addy Osmani from Google was had a had a blog had a had a had an article about this. He called it the orchestration tax. But one of the things he one of the points that he made in this in this post was we need to start thinking about not just how context engineering or agentic engineering

Ernest Koe (22:49.071)
Ready's money, yeah.

Todd Geist (23:05.678)
But attention engineering, which is the same thing that I'm getting at here, is we need to start thinking about how we get our attention to be well suited for the job that we're trying to do. And so whether that is like being clear on like when you actually do your best thinking, you should make sure that that's your free time. You should like if you're a morning person, that's when you should do this. You shouldn't do it when you're tired at the end of the day.

Things like that. But I wonder if it doesn't go further. I'd wonder if it's not like we actually need some three-dimensional way of of of of paying attention to our code or some tactile way of paying attention to our code, which will help us kind of unlock this maybe, maybe new levels of being able to understand and interact with the code base.

Ernest Koe (23:57.285)
I I think that rings true. I I I don't know how you do the latter, but I certainly feel like I'm actively trying to reset the way I work. You know, I think I got into a good place before November of last year when like, you know, the whole thing the whole thing exploded and I was getting into AI psychosis where I was sleeping earlier, I was waking up earlier, I was I was actively meditating in the morning, I was getting my tea the right time.

I had a good rhythm and I set myself up to be working in the morning and all that got blown to smitterings. November came like a spent three I wasn't sleeping. I was going crazy. I was like coding through to the middle of the of the night and then waking up early and doing all over again and and feeling high about it. but yeah, I I I I think that my attention regime, a lack of attention regime really hurt me after a while. I I could feel it.

could feel like I this is this is really not working and I need a reset. And so I think I've been trying. I think it's actually been better. Like I don't I don't feel so anxious about, you know like it was a lot of FOMO before. Like you know, like it's not so like like it was like if I don't do this now Yeah. Exactly what am I doing? Like what am I what's my value on this on this on this planet? You know, why am I not what am I here? You know, what do I serve? You know and all that.

Todd Geist (25:11.47)
If you're not running twenty agents overnight, what are you doing? You know.

Ernest Koe (25:25.089)
Anyway, I think long story short, I I I think that in a weird way I'm coming back to the idea

If we don't have a good practice around attention personally, but also across the company and across our teams, this is just not sustainable and doesn't it's it's it's it's not going to yield the results of productivity that we think it it you know, people thinking that it does.

Todd Geist (25:53.708)
No. And and I think like, you know, I'm a I'm a pretty serious meditator and one of the one of the reasons is because attention is really the only thing you have. Like everything else is everything else is transient, right? Even attention itself is transient, but at least it's yours. It's the most it is the most in demand or the most scarce resource that as a human you have is what you pay attention to.

Ernest Koe (26:02.327)
Right.

Ernest Koe (26:21.484)
Yeah, I gotta tell you, Todd, I I got so I felt like I got in got to a good place of delegating and creating space for myself and being good about my attention and then and not taking on too many things that I can't execute well. And then AI came and then I made myself like fifty new jobs. You know, I'm maintaining Tesseract, I'm maintaining that, I'm doing this, I'm like

I I see a problem I can AI this and I'll I'll but build skill around it and now I'm maintaining that. You know, so all of a sudden I've double and triple the number of things I'm keeping in my brain and trying to get ahead off. I I think that's not good. I think that's that's the opposite of what we're trying to do. So yeah, it goes back to this thing that people keep saying, like, AI is gonna take jobs away. I don't know. You know, I think some jobs are probably at risk.

There's obviously going to be disruption. Right? But if you if you believe in this thesis that attention is what matters, then you're really constrained by the capacity for humans of human attention. And if you can do more, then you need more humans to pay attention to the things that you want to do more of.

Todd Geist (27:37.378)
Yeah, I think I think we'll I I mean I I think that's I I'm sort of both positive and n negative on that at the same time. Is that I think the because I I I think it is just gonna be a big disruption. And, you know, there were lots of ways to have careers that

Ernest Koe (27:47.812)
Mm.

Todd Geist (27:59.906)
you know, maybe didn't require a lot of attention. And you still were able to, you know, you could still have a house and a family and a car and things like that. And and I I I worry that there's fewer and fewer things that are going to be deemed economically valuable if they can be automated away through whatever it is, AI or robots or whatever. I think we're a ways away from that. But you know, there there's and I do think

What do I mean? So I I do think attention is the most valuable thing you have, but but that doesn't necessarily trans translate any given person's attention into economic value. And that's always the the real sort of tension in capitalism in particular, right? Like there there's being a human and being even like a good human that has a positive impact on your community doesn't always in fact

you could argue, rarely connects economic value. And that's always been a problem. And I worry that this is going to make it a little bit worse. Or maybe a lot worse. so we'll see. We'll see. But I, I, I, I, I'm still like sort of like I'm still I there's just a lot of problems that need to get solved in this world. And so I am for there being ways in which to solve problems faster. And I do believe AI is that.

I just think we're a long way from it you know being able to do all this stuff without humans in the loop. You know, I think it's gonna it's going to be something we're gonna have to drive maybe forever. So I I don't know. It's you know, it's it's sort of like I still I really love what I can do with the AI tools. It still feels really, really good. but now it's also you know, like

it has to equate to something that is going to allow our company to thrive. or I'm not sure what, you know, what value it actually brings at the end of the day. And I think it will, but I just don't think it's clear like in every business that this is you know, outside again, outside of coding, I still think I still think software development is the killer app for AI. I don't think there's a better use case currently out in the world than than AI coding.

Todd Geist (30:23.724)
So I'm still pretty bullish on that. But I think we do have to get real about what that means in terms of it's not the code that matters. It's the problem you're solving. It's the attention that you spend on it that is the critical piece. And so we just have to we have to get through that part. We have to learn how to how how to do that. And and we're still early. We're still really early in that game.

Ernest Koe (30:46.893)
I think we're still really early. Yeah, I I will say that yes, I think software development and engineering is

biggest benefits are accruing too with AI currently. But every business runs on software. So I mean I think you can see, you know, if nothing else matters, yeah.

Todd Geist (31:04.717)
Yeah, that's true. Yeah. Well, I you know, I I I would love to get some actual data on that because I think it's really easy for us to say that, because that's what we do, right? But if I think about like what my wife does, I mean, sure, she runs her business on software, QuickBooks and WordPress and really not really much need to do beyond that currently, you know.

in terms of the software that run. And I think a lot of businesses are that way. So I don't actually know like how I would just love to know like how many businesses really run on the kind of software that we make, for example. Like what's that percentage?

Ernest Koe (31:45.11)
Yeah, no, I'm not I'm not saying that all companies or most companies run on bespoke software and that is their their core differentiator or the way you think about where their primary competitive advantage comes from. You know, certainly I think that's a class of those and and those are the companies that we tend to work with. But I'm saying that even even if you take a f a a sort of a a standard way of thinking about

Todd Geist (31:53.965)
Yeah.

Todd Geist (32:05.346)
Yeah, yeah.

Ernest Koe (32:14.099)
software as just renting a tool that does a basic business activity or task. You know, manage your accounts, manage your books or process an invoice or whatever that is and it's a fairly routine thing. Like when I go to a restaurant, I'm not reinventing the you know, the the the POS system. for you know, like they're not gonna do that. I mean there's tons of

really great POS systems out there. I know people who do that. But there's a really great POS systems out there. And you really I mean the economies of scale don't really will never probably allow you to custom build something that makes you quote different or differentiated, right? it's really well motion. So so the the economics of it already make that an impossible situation. However, the people making POS, the companies making point of sale systems, now

Todd Geist (32:52.898)
Restaurants have extremely low margins, just to say

Ernest Koe (33:07.331)
hopefully iterating faster and getting more wax at problems that they're trying to solve. And you know, I think I think hopefully the value of that then becomes accrued or or translated to the to the customer. Right. So I think I I think it creates competition. It it it hopefully will get value out better value out to to people faster. And I actually think that the the s the disruptions that we're seeing in the SaaS world click up, linear

All these companies, Notion, even that. I think they're going through a version of this because the pressure on them is to innovate better because candidly they've really had the economy of scale thing working for them to the disadvantage of consumers. My point is simply that I think whether or not you are building a bespoke business software solution for yourself or just renting one, I I do fundamentally think that the agentic engineering revolution will benefit consumers ultimately 'cause they'll have more choice. They'll have more choice w yeah. Yeah. In I in either direction, whether buying or building, right? And in either direction they'll

Todd Geist (35:17.442)
They will. I I believe that too. I believe that too.

Yeah, yeah. Yeah.

Ernest Koe (35:28.651)
Lost my headphones for a second. They'll have more choice. You wanna shift gears a little bit?

Todd Geist (35:31.756)
Yeah. Yeah. So I mean, w it may be that we're sounding down on on on AI at the moment or this has maybe a little bit of a downer episode. I wanna be clear that I don't that's not what I feel. What I do feel is a sort of coming to grips with these questions that that we've been asking. Just because you can you can pound nails ten times faster doesn't mean you have ten times more ideas of what things to build. Like

You still have to have good ideas. You still have to actually build them and test them and try them and iterate on them and support them and all those things. And none of that gets markedly easier with AI, except, you know, the fact that you can write more code and you can experiment more. That's it. That's it. The rest of it is still the very hard problem.

Ernest Koe (36:17.718)
You still have to build yeah. You still have to deliver a service or a product that has enough value that can justify the cost of that AI effort that you just the the amount of tokens you just burn, right?

Todd Geist (36:24.834)
Yes, right.

That's right. That's right. And so like, you know, some people who maybe haven't made the leap yet into AI might say, like, well, okay, then I can skip this. I can just keep coding by hand. I wanna I just wanna put the caution out there that I don't think that's the case. Like we have a new tool and it is better than the old tool. We just don't know how to use it as well as we will in the future. But I can't imagine going back to doing this the old way. I mean, I really can't. The the the productivity gains

in terms of writing code are significant. And they're so significant that if if if LLMs disappeared tomorrow, there would be a there would be it would be a hard few months as people had to, as people had to readapt and we would lose all those gains and it it would be a real problem. So the what we're saying here is not sit this one out, to be very clear. Not you if you're in the software development business, if you're in the file maker business, you can't sit this one out.

But you can have a more realistic understanding of where the gains really are. And so you can sort of you can build up intuitions around what it will help you do better and what it won't help you do better and why you're still a necessary part of this equation. So

Ernest Koe (37:41.653)
In the in in the file maker space specifically, I I think my message is more like w we now have a tool that puts us on a level playing field with almost every other thing that is out there and then some and and this is the critical piece. You know, i it's i I think it would be a mistake

Todd Geist (37:55.629)
Right.

Ernest Koe (38:06.998)
To not take advantage of that because that's always been the constraint that we have had to sort of work around or try to to compensate for in different ways. But now you can build modern apps. You know, now you you gotta do that well, you gotta do that thoughtfully, you gotta do that with the right amount of attention and all those things. It's gotta be engineered well. But it's a different story when you can't. and and that's the

Todd Geist (38:31.886)
That's right. Yeah.

Ernest Koe (38:34.27)
zero to one, I think, breakthrough that we need to understand. Now, again, this isn't addressed the larger AI story and the whole agentec space and how it affects businesses and all that. I think there's a lot of, you know, stuff that happens there. But man, if we don't get on this path, then you are playing, you know, with shorthanded basically, you know. Yeah.

Todd Geist (38:55.426)
Pretty much. Yeah. That's pretty much what it comes down to. So that's why the the news recently in the file maker world, you know, we've been talking about this for quite a bit with Proof Kit. It's now available and out, and you can go and download it and you can experience some of the productivity gains that we are that we're talking about here. And then a couple few we I think we talked about this on the last podcast, but so Ryan, Ryan, the CEO of Claris, dropped a blog post.

three or four weeks ago now talking about Claris's commitment to building the hooks into FileMaker to allow you to use Cloud Code and cursor and all the coding tools that you might want to use to work with FileMaker. And then I guess this was just yesterday. I've been traveling so I've kind of lost track of time. But just yesterday the our our friend of the show

Ernest Koe (39:43.062)
Yes, yesterday, yeah. I do remember this one. I do remember this one.

Todd Geist (39:53.47)
Ex-host Martha Zink gave a webinar Claris, which if you attended, was pretty remarkable in a couple of ways. One is that they started to demo some of the what's coming and they talked about the fact that they're working with us to help them speed up the process of you know, baking in some of the ideas that that we've had from a third party perspective into a first party perspective that they would do instead of us.

And then I think the other thing that just is worth saying as we jump into this is this was an extremely well attended webinar. Like I think normally the FileMaker webinars are, you know, four to five hundred people show up for these. This one had over a thousand. And they did a poll, and the poll was you know, basically 86% were excited or already doing these kinds of things, excited about doing these things or already doing these kinds of things.

within the Claris marketplace. So just to say the webinar was very successful, both in terms of the the number of people that showed up and their sort of positive reaction to what was shown. So let's start with that. Like that's a big deal, yeah?

Ernest Koe (41:05.238)
Yeah. Yeah, it's a big deal. It's a big deal. I mean, I don't think I had many doubts that that would be the case, but you never know with these things, you know. Like we we're drinking w we're AI pilled, we're drinking the Kool Aid and and I remember when with all these great ideas about proof chat that would be the best thing ever. It's like kinda Right.

Todd Geist (41:16.046)
Sure, yeah.

Todd Geist (41:26.606)
Yeah, it turns out proof chat, by the way, is an is basically a non-mover. Doesn't people don't use it. And I think we know why that is, but and we can talk about that maybe a little bit. But yeah, so we're always, you know, we had we had the idea for proofkit and and the sort of agentic coding. We really saw that after December. We were like, regular file maker people are now gonna be able to code web viewer apps, which which which our web team has been doing for a long, long time.

Ernest Koe (41:55.658)
Right, right.

Todd Geist (41:56.268)
But so we had this idea in December and we got very excited about it. But, you know, like you say, we we get excited we got excited about proof chat too, and it turned out to be you know, a big basically people use it, but it's not the hit that we were hoping it would be. so seeing this webinar with so much excitement was definitely good affirmation for us.

Ernest Koe (42:21.429)
Yeah, so wha what else to say about that? I think as a reminder that Claris is working hard to bring proofkit like functionality into into the product as a first party experience and we're very excited about that. I think that you know, I sometimes get questions like, Well, does this compete with Claris and how how do you guys s see that? just unequivocally I'm gonna write a little blog post about this, but we don't see this as

us competing with Claris or or vice versa. But we're we're we're actually encouraging them to take as much as from Proofkit as possible and and either re-expressing it or implementing it or integrating into the product, right? Because I think fundamentally these things have to be a first party experience. otherwise you know, we have to work around some of the rough edges in the product to to get this working.

Todd Geist (43:05.454)
They do. They absolutely do.

Ernest Koe (43:13.941)
And it it's it's it's great as we we are doing it now and it's very, very useful and and very powerful. We're not stopping with ProofKit. But yeah, a lot of the stuff that we're doing should be baked into the product and certainly we have theories and and opinions about what what that needs to do, but I think this is exactly the right move for the whole community of of of Claris and and and FileMaker. You know, and the way I think about it is and maybe Ty have some thoughts here, it's like

Proof kit will evolve and we'll continue to have our own opinions about w what we extend, what we do differently from from what what is stock. But it's like we want the stock car to be pretty good and then if we want to add a spoiler to it and add new rims, then we'll do that, you know. But it's gotta be a d really great car out of the lot before you do much to it, you know.

Todd Geist (43:55.662)
Yeah. Yeah.

Todd Geist (44:04.854)
Right, like, you know, to maybe take that analogy a few steps too far. you know, like in order to get the proof kit car to dri to move, we had to like we had to like build a steering system. Right. But even though maybe our goal was only to, you know, make it go really fast and have a nice spoiler and to, you know

Ernest Koe (44:09.281)
It's dangerous.

Ernest Koe (44:30.337)
Right. Right.

Todd Geist (44:30.486)
whatever whatever speed demons really like. But the first thing we had to do is we had to actually have it steerable, basically. And so we had to put the steering. We had to like put the wheels in that turn. Yeah where that should just be part of the car, right? Like the car should provide that. Steering should be part of the car, not the aftermarket stuff. Like you wouldn't you would like I mean you might you might get an aftermarket steering wheel, but you want your your car frame like the thing that rides on.

Ernest Koe (44:41.237)
re engineer the steering system.

Ernest Koe (44:49.289)
Right. Right. That's right.

Ernest Koe (44:56.096)
Yeah.

Todd Geist (44:59.458)
to be able to steer. Like you don't want that to be aftermarket. Yeah, yeah, exactly. It should be steerable. And so we had to put in steering before we could do the rest the stuff we wanted to do. And brakes, you know, like like we had to put in brakes. Like like that should the just the car should just have that. Right.

Ernest Koe (45:00.757)
We want it to be stirable to begin with. Right.

Ernest Koe (45:10.251)
Right.

Actually, that's a great analogy because that's just exactly it. Like brakes and the fundamental systems that allow a car to do what cars do that just need to be part of the carness of car, right? Yeah, we want we want a new radio that has Apple Play and not Android car play or whatever. That that we can do. That's aftermarket and

Todd Geist (45:25.152)
Yeah, the yeah, exactly. The carnists.

Todd Geist (45:33.526)
Yeah. Yeah. Yeah, that's third party. That's like that's that's where that's where aftermarket makes a lot of sense or third party ecosystems make a lot of sense. But there's stuff that belongs in the core. And we've done this before. Like this is something that that we've done over the years with Claris where we've we've sort of been ahead. Like when we did OttoFMS first version, there was no admin API for the server. So we had to build our own. And as soon as Claris provided that, we

Ernest Koe (45:37.704)
Third party.

Ernest Koe (45:49.139)
Over and over again. Yeah.

Ernest Koe (45:58.24)
Right.

Todd Geist (46:02.538)
stopped providing our own. Why? Like that's just now we can go farther and we'll do the same thing here. yeah, yeah, we'll do exactly. We did transactions forever. Now we do it the way Claris does it because that's better. And I think there's something worth especially at this time where like we've noticed this because again people can build anything that people are building rather like it's very clear that for most

Ernest Koe (46:04.429)
Sure.

Transactions, the same thing.

Todd Geist (46:32.328)
I I I don't know most. A significant number of the developer shops out there will maybe take a look at ProofKit and then they will build their own because they want to build their own. And that's fine. Like we think people should should do whatever they want. But if the market is going to be changed by something like ProofKit, it has to come from the vendor, right? Otherwise, everybody's in their own silos. This has always been part of the problem in the in the Claris market to some degree.

Is that people just want to build their own things. They don't really wanna, they don't really want to adopt something from another third party unless they have to. that's always been something that's been been kind of a challenge. And so if agentic coding is going to be a thing that helps the entire market, that has to come from Claris. Otherwise, everybody will have their own piece. There'll be no network effects, there'll be no ecosystem that springs up around it. It's critical.

That this stuff gets gets baked in. And so we're super thrilled that they're doing it, right? Like that's exactly what we hoped would happen. And and so yeah, it's an awesome win. We're super excited for what they're doing. And we're super excited to keep pushing on the other side of whatever it is they give us to build on. We will continue to push on that and see what's next around the corner for, you know, the next round of this cycle.

Ernest Koe (47:30.401)
It's critical. It's critical. Yeah. Super thrilled.

Ernest Koe (47:54.261)
Yeah, and Proofkit will be I think there's no reason Proofkit will not be compatible with anything Claris does because it's really the thing that we just built on it. Yeah.

Todd Geist (48:00.758)
No, we'll just build on we'll just use it. We'll use whatever they provide as the as the you know, the brakes, right? Like the steering wheel. We won't we will not force you to rip out the steering wheel that comes with Claris and use ours, like that would be dumb, right? Like we just wanna keep building and pushing what's possible ahead of that curve and rely on what's there.

Ernest Koe (48:07.584)
Right. Exactly.

Ernest Koe (48:22.08)
But until until Claris does drop their official AI kit, dev kit, or whatever they call it, Proofkit is I think the best way for you to get some experience on how to do Agentic development, especially on the application front end side of things in FileMaker. So I think you should definitely take a look at Proof Kit, go download it, and if you switch to Claris' version later, things will just hopefully just work, you know, and it wouldn't be

Todd Geist (48:39.384)
Yeah. Yeah.

Todd Geist (48:50.774)
Yeah, I mean, we'll be able to come up with we'll be able to you know, we'll w whatever we do in the future, it'll it's it's sort of like this may not be clear if you're not deep into this world, but it is absolutely going to be compatible with whatever Claris does. maybe right out of the box. But if not, what what there will be is essentially a prompt somewhere that says, We built this with Proofkit and now Proofkit's got a new version that uses the Claris pieces. Here's a

Here's a a document about how you migrate the code from the old way to the new way. And guess what? This is something the agents are really, really good at. Migrations are AI is super good at it. So it's gonna be able to do it. Like, like these kinds of tasks are what AI excels at, you know, pretty wild. Like as just a sort of another example of how good they are at it, bun, which is another tight another JavaScript runtime.

was written in a language called ZIG and they were having some problems with the language, you know, the lower level language they were using. They needed they needed better type safety and and memory primitives. And so they switched to Rust and AI rewrote in six days, they ported a very complex JavaScript runtime from from being built in ZIG to being built in Rust because AI is very good at these kinds of things. So whatever the shape, the final shape of what

Claris's version of this is going to be, we'll be able to you'll be able to reshape it. Whatever you're building with proof kit will be extendable or you know, will be be able to be migrated to that version. So no reason to worry about that.

Ernest Koe (50:35.252)
Yep. Yeah, so if if you want to get an early look, and even maybe get a sense of what we are doing internally with our teams and just kinda figure out, you know, what is proof doing, how do they build, how do they work with clients, Proofkit is probably the the best thing you can do. You know, download it. You you you are seeing, you know, front row seats to to how proof develops and how we ship to the world. You know, so

Todd Geist (51:01.858)
Yep. Yep. It's it's got a lot of the stuff we do baked in. And I think it's just, you know, again, we say this, but I want to keep iterating on it. Like it's just the beginning. Like I people I think people will experience a lot of success if they go through the, you know, the the guides that we have and how to get started. but y they I I wanna be clear, there's things to learn here and we're gonna do the best that we can to produce

materials on how to do that, things like this podcast, but we're also gonna do we'll do more videos, more more demos, office hours types things where we can get more into the weeds on how proofkit actually actually works and how, you know, you can prompt different things like I wanna make a data table with like sortable columns. How do I do that? Well, we'll we'll have documentation on how to do that or or prompts and hints about how to do that, that kind of stuff. So yeah.

So it's the beginning. This is the beginning of of a new journey. but luckily AI, the other thing AI is really good at is teaching. Actually, can really help you learn how to do this stuff. So I just encourage everybody to at least give it a try, you know, download it, start a new file. I'd start I'd start in the greenfield first, just so you can just feel very not worry a lot, know. Yeah, and just like you know, not care too much about.

Ernest Koe (52:22.653)
Give us feedback. Yeah.

Todd Geist (52:28.812)
what you're doing, just have fun. Make some little project you've always wanted to do but never had the time. Try that or something like that. Just to get started. I think that's the easiest and freest way to go.

Ernest Koe (52:39.603)
Yeah, and give us feedback. This is this is really a developer enablement quote unquote motion for us. You know, we are using this to make our tooling and our teams better. And we want you to succeed and we want you to be to come along with us on this ride and any feedback that you have ha have on this will also help us and everybody else who's getting on board this train. So love to hear it.

Todd Geist (53:05.196)
Yeah. And and you know, people ask us sometimes, how do you why do you guys give away so much for free? Like we literally get that question. People, I can't believe you did OttoFMS for free. I can't believe proofkit's free. Somebody's like, one of the people is like, how do I pay for how do I make sure this doesn't go anywhere? Like, you know, they want to pay for it. They want to give us something to keep doing this work. And so it's maybe not always clear why we do this. I think the, you know, there's many reasons, but number one is, you know.

I mean, I think I have personal reasons why the file maker community is important to me. It's been the my almost the entirety of my professional career, except for the first five or six years out of college, has been in FileMaker and I I know a lot of people here, a lot of a lot of friends met over the years here. It's very important to me on a personal level. But on a business level, you know, all of our customers are file maker cup people. Like we have a vested interest in this platform being

as powerful as it possibly can be. Like that's more important than us charging for these products. It's way more important that this platform be powerful. it just serves us better economically from a business perspective. It's quite clear that that's true. So that's why we do it, you know.

Ernest Koe (54:21.777)
Amen. so yeah, what what's next here?

Todd Geist (54:26.552)
What else to cover? Anything? I don't know. I just finished the Mars trilogy. Such a great that's my butt you know, people do they do the sick pics on the syntax podcast, you know. recommendations. Yeah. The three books you must read are Red Mars, Green Mars, Blue Mars. read it. I don't know. I I've loved it the second time even more than the first time. and I just I had it, I was listening to an audio on Audible.

Ernest Koe (54:38.087)
Yeah, the three books that you must read, you know, that sort of thing.

Green Mars and Blue Mars.

Todd Geist (54:56.332)
And I turned the speed down to point nine X. So it was actually usually I'm like a two X or you know on a on an audio book. But I loved it so much that I slowed it down. I literally spent the this entire year so far, almost almost six mo five months in. I've been listening to though to that series. And it's it's finally over and I'm a little sad, but well worth it. And they talk to AIs a lot in that, in that in that series. AI's

Just like a thing everybody has.

Ernest Koe (55:28.755)
Yeah. That that would be my dream, right? Where AI just melts into the background and it's just just like a tool that you use to do real work, you know, to connect with people.

Todd Geist (55:29.623)
Nice vision.

Todd Geist (55:37.73)
Yeah. Yeah. You want to build a new town, you just talk to the AIs and they fire up the robots and and they start making a new town on the side of a cliff and a crater in Mars. That's I we're a long way from there, but I still I like to dream, you know. Yeah.

Ernest Koe (55:53.264)
I think

Ernest Koe (55:57.032)
I think this is a good place to probably end things for now. I think we'll have more things to talk about next time, maybe around Agents and

Todd Geist (56:00.856)
Yep. Sounds good.

Yeah, maybe we'll do maybe we'll get technical on to like what some of these proofkit things are. Now that there's actually hundreds of people using Proofkit, we maybe we'll do some get into some of the technical like how to prompt certain things, what to expect kind of stuff, things like that. Be good to do. Smash that button, tell your friends. Yeah. Tell all your friends the the the context podcast.

Ernest Koe (56:23.945)
So subscribe, I don't know where the button is. Subscribe. Smash that button. Tell your friends. Check out the show notes. we do have a playlist.

Todd Geist (56:34.668)
Yeah, do all that stuff. Tell everybody. and let us know, like also like let us know what you'd like to hear from us. Like, you know, we'll talk about whatever we like. Like I I we could do a whole podcast on the Mars series. We could do a whole episode, we could do many episodes of the Mars series. I'm not sure that would resonate with everybody. So let us know what you know, what questions you have, like what things you'd like to hear us discuss, you know, respond to, things like that. We'd love to get that kind of stuff. So thanks everybody.

Bye bye.