Activation Energy: How AI Gets You Back Into Old Projects Vibing with AI Code, Episode 11: Brendan & Baird Brendan: today I was thinking, you know, we could talk about back into some projects. Um, I've actually dove into, I don't know, three, three old projects, and I even started a fourth all yesterday, and I thought that might be a good topic to at least start with. Um, so maybe I'll just kinda dive in really quick since I was doing that. You can speak at, you know, what you've got and see where we land here. But, you know, I, I've got, like I said, three… I think it was three. Yeah. There w- actually no, there's… One of them was basically just, like, my command center, which the whole point of that, and I, I feel like I talk about it enough, but I, I'll repeat it for anybody new and whatever, a memory type of thing. Like, it, it has the MD files all over the place, so it's good. I knew that… And why I'm bringing this one up is because I have w- with my personal setup, gotten really lax about, like, doing the daily routines that I originally set up. So the way that it was kind of set up was do a morning routine and then do an end of day routine, and then those kind of like put the, the or the markers in the MD files to then pick up and have a memory. So that was one instance where it was, like, not a project, but I also still wanted to be like, "Hey, what have I missed in the last week?" like my, my personal emails and life stuff. Baird: Yeah Brendan: projects, you know, I've got like a video game tracker thing that's just for myself, that one was one where it was like mid trying to, like, do stuff. It's been, I don't even know, six, seven, eight weeks maybe, something like that. And then another one is a, a project that we don't need to get into the details too much, but hopefully I'm getting back into it, you know, we can talk about it more. But one that we… In fact, when we did the, the episode where we talked about using Get Shit Done as the, the, um, skill or, or like… I Baird: Yep Brendan: it's a library more than Baird: It's a plugin Brendan: or Baird: Yeah Brendan: you go. Yeah. Baird: Yep Brendan: Like, that also helped because that plugin already has kind of and markers Baird: Yeah it it it plans everything and does the roadmap and has that like memory as kind of a core feature to it Brendan: Exactly, yeah. it was interesting to just see these three different things where pre-AI, if we didn't have any of this, probably would've never done any of it, for one. Baird: Bishop Brendan: I Baird: Yep Brendan: past the point of no return. I never would've been able to recover from it. And i- in some ways it's, I think, also because I… All of my projects, I, at least the ones I've started in the last, say, nine months, do basically tell my, uh, Claude, I guess, in, in the repo, like, "I want you to be an orchestrator, and I want you to, like, Baird: Sure Brendan: So it's al- I've Baird: Yep Brendan: trained it in a way Baird: Yeah Brendan: to help me do this. But just, like, seeing how it actually works and actually, like, is successful is really Baird: Yeah Brendan: because not only can I be like, "Hey, give me an update of, you know, what, were doing last," which it's great at. A- and you know, I can also just be like, "Hey, you know, okay, great. That's awesome, but then, like, what was this command we did like Baird: Yep Brendan: 10 months ago?" Or whatever. Baird: Exactly Yeah Brendan: it, it, like, actually knows. And so, you know, like deploys are things that I, I inevitably when you build a new system, even in this case, like it's on Vercel, but like I haven't even done that, so it's like, wait, what was the way to do it? You know, stuff like that is… It's been great. So I, I mean, that's kinda what I wanted to start with. That's where Baird: Yeah Brendan: Like, what are your thoughts? Have you had any similar experiences with this? Baird: Yeah I mean I look at it like this You you get to get ask it you know what is it that you do here when you when you come into a project A and I agree completely I haven't done it as much lately but remembering when you go Like whether it's a side project whether it's a work project no matter what it is remembering that like Hey I can just ask the AI like what have we done in here And it can go through give you like essentially either like the summary of these are the things this is what it does this is how you like all that type of stuff or the deep dive of you know what are all the features you know what's the architecture It's so powerful A and I think it's almost like a meta layer on top of it but it's also you know like you have to think of it which is part of the problem And so it's like remembering you you we almost need to build like a skill or something where it's just like the what am I doing here type skill where it runs through and and does that as you know part of the process And and I will actually say I don't know if you've noticed it working in Claude Code recently they now have memory baked into Claude Code as well Like I've actually been watching it as it's going in in a project and remembering things that we actually talked about seeing it writing stuff there and I think that's super valuable Like you know me memory is such a weird thing right now where it can be extremely valuable but it can also be like are you remembering the right things You know like is it actually like doing like having the right context for what you're talking about And and I find it probably helps in Claude Code cause you're always in a project You know like you're working in that directory so it has a very clear delineation of like what memory you're working off of which is huge But like if you're in like the regular chat flow that memory can be like usually gets pretty right but it can also conflate things pretty easily as well So that's an interesting thing I've noticed just as a Brendan: It, Baird: tangent to this Brendan: No, I think that actually makes sense, and I, I saw something that is directly related to this, um, yesterday when I started. So actually also, because I was able to quickly get up to speed with these three things that I already mentioned, I had another project in mind that I was like, "I mean, I should try-- I should throw Fable at this," is where Baird: Sure Brendan: went. Baird: Yep Brendan: I've seen a lot of positive things. I really hadn't poked around with Fable much. Baird: Yeah Brendan: And so I decided to, A, I wanted… I was just kinda curious, like, on the $20 plan which I'm on for personal stuff like this, how far can I get with Fable? Baird: For sure Brendan: it, it got really far Baird: Yeah Brendan: in, in parallel with running all the different things. But the, the thing I wanted to bring up that was interesting is that, and this is something I've been doing more with my work, um, setup with Claude and everything, is I have the command center kinda just like run everything. Because in that context, writing less code. I'm, you know, I'm the, the manager, and I am more reviewing things. But I do also want to like reach out to the repo and say, "Hey, where is this thing I need to do?" Or, "Remind me of this thing." And it's really great about that. so when I was starting this with Fable, I, I actually decided to try starting from my command center and saying, "Hey, you help me plan because I also want you to know about this stuff." Baird: Sure Brendan: But then when I got to the point of, know, net new, this doesn't exist yet, what do I do? Should I keep, should I keep running it from here in my command center and like h- you know, let that version of Claude do its thing? And it actually came back to me and it said, "Actually, I don't think that's a great idea because then the memory is gonna be mixed up with all these different things Baird: Sure Brendan: just the one project." Baird: Yep Brendan: that's something that I thought was interesting and that it called that out specifically, and it made a lot of sense. And I think, again, back to what I do like with my work machine, sense to do it that way because I want everything all Baird: Yeah Yep Brendan: this case, it's like, no, you know, the command center stuff for my personal is basically like my emails and my calendar Baird: Yeah Brendan: like that. I don't need that in this project. Baird: Yep Brendan: I thought that was Baird: No that makes perfect sense I a and that's such a good thing to learn to look for is and I mean it is this shocking No it's context You know like managing context what context are you putting in But it's easy to kinda like what you were doing there it's like oh like it makes sense to start here But sometimes you kinda forget to ask those questions Like is it actually the right place Do I need to do something different And like that that's you know for me AI these days that's one of the huge things that it it benefits me is like being able to ask those questions and get a second opinion And is it always right Not necessarily but it gives you a different way to think about it But it's also and I I think the Brendan: duck that actually talks Baird: For sure E exactly If if you're you know experienced enough you can tell when you should follow it and when you shouldn't You know like Brendan: Right Baird: I find more often than not when I'm talking to the AI I've got a pretty good idea of what the answer's gonna be or the right answer's gonna be but you get something different And you know cause you know I I don't know everything o obviously and so I don't pretend that I do but I at least have like a an inclination about it So it but it's definitely interesting to see how it responds to it But I I think the bigger thing before we d dive into anything deeper is just like if you abstract what we're talking about here activation energy is what we're really talking about It's like how can you pick something up and get started that you wouldn't have done preAI by having that as a a you know power that you can use A and like I've always found it where it's like I wanna do this but you know how do I even get into it And we we talk about this all the time It's like if you haven't played a video game in a while it's like Brendan: Yeah, Baird: know I wanna play that game but what what are the buttons again You know like what where was I in the story Like it's that type of feeling and like I always do that and then I'm like Ah well I won't even play it because it'll take me 10 minutes to get back into it Same thing happens on Brendan: 10 Baird: coding side of it Well yeah Yeah I mean that that's probably more realistic I never play for two hours so I don't have the chance But Brendan: That's Baird: but yeah exactly But it's that kind of feeling the like diving into any project where it's just like oh man I haven't touched this so long What was I even doing And you know I I think it's actually an interesting way to approach it too And honestly like I'd be curious to see like how you feel about it as you do more with it on your side is like with how quickly AI keeps advancing on the code front it's like how can you even when you jump back in A have it tell you where you were at You know get you back up to speed really quickly Brendan: Yep Baird: But also like were we even doing the right thing You know like was this actually the best way A and Brendan: That Baird: especially with Brendan: yep Baird: yeah w with with Fable being out now and this is actually something that I just did with Fable this last week is you know I've noticed too on the 20 plan you know I didn't have it build much but I had it plan a bunch of stuff And I mean I had it running for probably like two hours planning stuff and I still had room to do other stuff with it like you know with Sonnet and stuff like that But what I noticed is cause I was curious I was like you know How much better is this And so the work project that I have where it's like the AI stuff for our internal I just tasked it like Hey go out look through everything that we've done here and just like give me recommendations about what's you know what could be optimized what could be better what's like what are more you know modern approaches for it And I mean it came up with like two huge plans like one for my the like rag and all that type of stuff on the like the true interactive portion and then one on the like there's a lot of background processes that get kicked off with cron jobs And the way it was built originally was like it's pretty sequential You know there wasn't really any agentic work happening Like there was AI involved It would be like Okay go out grab some news grab some data do some crunching on it and then you know disseminate it places Working with Fable I was able to actually design a new system that is more of like a true agentic background process that isn't just like sequential steps It actually makes better decisions on like is it worth continuing to do the next couple steps You know how do we actually approach it A and so I definitely noticed that Fable seems to do a really good job planning Like I said I kicked that off did most of the planning and then ran it through Sonnet to actually start implementing it And I say Sonnet 5 given that that just came out a little bit ago That definitely I know a lot of people are like not as happy with it but I've actually seen pretty good results so far with it Brendan: Yeah, that, that's cool. Uh, and I think that, that, that's interesting. You're, you're actually making me think of something too. Like, when, when I went back to the, the project u- was using Get Shit Done, Baird: Sure Brendan: it… The… I am using Opus 4.8, Baird: Mmhmm Brendan: and I, I had the feeling though, like, as I was going through, so, you know, w- if you haven't used that plugin, you know, what it does is it does a lot of good planning things, and it asks a lot of questions. in this case, I'm using a, a Rails 8 app, and I've been using… I've been doing Rails forever, I haven't done any Rails 8 things, and so I know that there's new things like Kamal and, um, or even, like, Active… or not Active Storage, but Solid Queue in place of, Baird: Sure Brendan: if you're familiar with those. They're just background job workers. But was asking me questions about that kind of stuff, which is totally cool, but at the same time, I'm sitting there thinking like, "Is this overkill? And are we doing it right?" And like, should I have Fable kind of reassess? Because it has been… You know, w- we talked about the activation energy, but it's also been, I don't even know, two months at least since I really dove in and, like, started this project. And it's not even built. It's, like, halfway built with architecture that maybe I should be reassessing already because it's Baird: Yep Brendan: a point where it's done, right? So Baird: for sure Brendan: that was a thought I had as well, that I, I, I still might even throw Fable at and just be like, "Double-check my work here," kind of thing Baird: They've got a day and a half more to to get through it so Brendan: Yep. Baird: then we're back on the API so which will be an interesting Brendan: mean, they, this is like the second time they pushed it back though, so Baird: Yeah Well I I think part of it too not to go too deep on this type of stuff you know Brendan: Oh, Baird: 56 Sol just came out um well it's been out for a little bit but they actually just released it to everybody it was yesterday I think Brendan: Okay Baird: and s that's supposed to be pretty comparable to Fable But as of now OpenAI isn't putting it on API only So it'll be interesting to see to your point that you know Fable's been pushed back once already Will they do it again and potentially even permanently do it given that there's now competition for another model that's at least close Like I've seen some conflicting reviews Like I know there's several people that are like This is the best model ever now like the Sol model But I've also seen some people that are like It's really really good but there's still things that I'm gonna throw Fable at instead of it And I think it's an important distinction of like we're getting to a point where I mean we've been there for a while but like you have to choose the right model for the problem versus being like Fable's perfect for everything or GPT 56 is per Like it's keep it's evolving constantly but it's also some are better than others And you know I don't know about you but all of the benchmarks while I get we need to have benchmarks so we have a idea I tend to never feel when using models that they're like the benchmarks tell the full story You know it's like sometimes I I have a model that people are like This is terrible and it feels really good to me But then there's other times where I'm like This is supposed to be a great model and it feels terrible So I think an important part and we've talked about this a lot in general is like just play with it you know Like actually experience it and see how it feels to you to figure out what's gonna do what you need to best Brendan: Yeah. Yeah, and that's the thing I still struggle with a little bit is, like, knowing the differences are between the models. Baird: Yeah Brendan: if I the exact same thing with, you know, let's just use Sonnet, Opus, and Fable, know, I would… I'm sure I would see differences. But in, like, the things that I'm doing right now, and even just, like, my day-to-day when I've, I've used Fable a little bit at work, Baird: Yep Brendan: like, reviewing things, Baird: Sure Brendan: in, like, in that instance, I really didn't see much. Now, granted, it could've been that the code I was looking at was f- we'll just call it simple enough that, like, it, there really wasn't much nuance there. you know, a- and I know there are differences, but I do struggle a little bit myself to really have a feeling for, this one's great, or this one lets me down in these places. So may- maybe what I'm getting at is what you said a minute ago, like, knowing when to use which models. Like, I, I base it mostly off of token usage almost Baird: For sure Absolutely Brendan: ways. And, and that's fine, you know, that, that's one way to, to do it. Um, and, and maybe as I build some of these projects more, you know, these side projects, I will, will probably get back to the point where I will start seeing that. But when all you're doing it on the day-to-day for work is, like, the command center stuff and just, like, stuff, like, I… You don't need Opus for that. You Baird: No Brendan: Sonnet for that. Like, and so it is hard to compare, I guess. Maybe that's, that's what's going on Baird: Yeah Brendan: But Baird: But but I think that makes perfect sense And I've found and I think it's uh we actually talked about this way back in our Gemma 4 conversation when we were talking about that I'm actually finding more and more and I I think that the consensus is getting to Fable level is actually the most efficient especially if you care about tokens Like if you don't care about tokens if you've got like a unlimited account or infinite money run Fable through everything Who cares But Brendan: Yep Baird: you actually need to be costconscious which I think more and more people are realizing they do Fable is awesome for creating plans and checking work And that work can actually be like okay plan it pass it off to like Sonnet cause it's cheaper build it and then go back to Fable and say Okay now check this Find holes And then like you you start to get to the agent loop that everybody's talking about But Brendan: yeah. Baird: in in my mind it's it's a more like controlled agent loop It's not like run infinitely because that Like I don't know if you've noticed it but if you keep asking the AI if it needs to do something it's always gonna find something Like I've never seen an AI be like Brendan: that Baird: This is perfect It's it's gonna start like veering off into Hey we could do it this way Like it it's just So for me I still am not to a point where I'm like I'm gonna let agents run on something like a programming task infinitely Brendan: Yeah Baird: for what you're doing for like a command center for like that you know second brain that memory type thing that's a perfect place to let it run you know Like not have to babysit it Because I mean I'm not saying that you're not gonna find hiccups there cause I I've done that stuff too and it's like Brendan: still hallucinate and all Baird: Yeah it'll be like This is super important And I'm like This doesn't make any sense at all But you know you then you gotta weigh up like how much is it helping you versus hurting you when you have to deal with that type of stuff Which and you know the majority of the time it's far more help than it is annoying But yeah Brendan: Yep Baird: just that feeling of I think we're getting to a point with these bigger models where they're not your daily drivers unless you've got infinite money Otherwise they're like your planners and your you know I mean it it's almost what you were saying like they're kind of the orchestrator They're not doing the work They're like managing subagents that are smaller models doing the work And you know I've I've noticed a lot both with uh Claude Code and Codex subagents are now like I mean they've been around for a long time like it's not like this is new but the actual you know harnesses are leaning into them so much harder And I think it really is like uh Fable's doing the orchestration and it's actually passing off to smaller models Because and I think that's partially why we're not seeing the usage get you know consumed so quickly is because it is doing that work with smaller models kind of automatically So and that makes a huge difference Brendan: Yeah, for sure. also reminded me too, like, um, I, I don't use it for my personal stuff, but I, I still use Conductor Baird: Oh sure Brendan: actually use it as much, but, Baird: Yeah Brendan: another harness and you can use… Baird: Yep Brendan: can- I, I actually have Codex in there, um, or not Codex, I guess, but, Baird: Uh Brendan: yeah, G- GPT, Baird: GPT5ish Brendan: I, I don't use it as much, so I Baird: Sure Brendan: uh, whatever the, the latest one Baird: Yep 55 Yeah Brendan: that's beside the point. Um, but what I have noticed though is, like, because that harness is, you know, built specifically as a, a tool to get code, you know, the, the programming workflow done, it automatically, depending on what task you do, it will pick the model for you. So Baird: Sure Yep Brendan: I think, I forget if we joked about it or not joked about it, but even talked about it on here or not, or just joked about it outside of this, you know, recordings. But a while there, it would actually, when you would have it write a commit message, it would write the commit message using Haiku, but then it would Baird: Sure Brendan: attribute the code writing to Haiku because that's who wrote the commit message. And so Baird: Like he was like Come on guys you haven't updated me in forever I'm still here Brendan: it's like it made sense to downgrade to Baird: Yeah Brendan: commit message, but Baird: For sure Brendan: attribution was total failure. Like, no, no, you didn't write the Baird: That's it This is why Haiku doesn't write the code Brendan: Exactly. But then, you know, uh, well, to be honest, like, that's, that's probably more of a harness failure. Like, Baird: For sure Brendan: should know, but either way. Yeah, what r- the more interesting thing is there's always been a review button since I've used Conductor, Baird: Sure Brendan: it's just an easy way to just prompt Claude or Codex or whatever to actually run a review. what I've noticed is now that Fable is a thing, and at least when you're using Claude, you know, it actually turns it to, to Fable in the Baird: Okay Brendan: than leaving it at Opus, Baird: Sure Brendan: also kind of funny I'm remi- remembering because when Fable was disabled, Conductor never fixed that. So every time you'd click review, at least for me, would launch Fable and say, "We can't do that." Baird: Sorry we're we're not gonna do that for you Yeah Brendan: So that, that's on the Conductor team, and Baird: For sure Brendan: d- upgrade my version. I don't wanna like, you know, throw Baird: Yeah for sure Brendan: way or anything, but Baird: No but but I I think I mean it it's I'm sure that's hap I mean it I know it's happened in so many places But it's an interesting kind of unique issue with where we're at programming right now It's like thinking about it when have we ever had to think about like I mean you think about like what programming language you're using or compiler or stuff like that But it's never been like What model am I using to write something You know So it's a whole different class of issue where it's like what A and I just like I didn't really run into an issue but in one of the products I've built you know Thinknato it was using I think GPT4 Mini or something just to do like really basic type stuff and they just deprecated And so it's like oh yeah I've gotta actually go back in and update my models to make sure that all of this stuff's running And it's I mean dependencies we've always dealt with dependencies so it's not really that different but it is a little bit more of a nuance A and I'm curious to see you know ChatGPT has already done like kind of the auto model stuff like routing I'm curious to see if that really is where we end up you know where it's just like in one way or another we're gonna abstract the model out so that the AI's actually figuring out what AI it should be using versus everybody has to figure out individually Brendan: Yeah. Yeah, exactly. I, I think, a- and we've touched on this too, is like it does feel like the harnesses are the bigger thing to evolve now. Like yes, the models are still going to get better, Baird: Yeah Brendan: are already, and even before Fable they were so good that really does come down to like the skills you're using, the harness you're Baird: For sure Brendan: And, and there's, there's still a lot to l- to ask for, like the… It's lacking a lot in most of Baird: Yeah Brendan: Um Baird: it it's the context problem but context has evolved so drastically from it used to be j I mean and here's the funny thing though Brendan: first episode when we were Baird: Yeah Brendan: or Baird: Yeah if we're like Oh yeah the co I mean it it's still true And the the funny thing to me is is that it still works the same too Like we're still just dumping context into the chat stream like you know the system message or whatever it is But now we just have like shortcuts where like you've got a skill that bundles that up into something and then it just gets injected when you invoke it or when the AI knows you need to invoke it Like it's interesting to see how these things are not changing and yet changing radically in interesting ways And you know just context management when the agent's doing it is also like a whole nother tangent too where it's like when you compact that's really cool but it's also really dangerous sometimes And Brendan: Right Baird: so w we're doing like You know everything's advancing but it's also still doing the same stuff It's just like those abstractions on top of it Brendan: Mm-hmm. and, you know, th- this is also reminding me of when, when I did get back into the project that, you know, is using Get Shit Done. So I, I basically, um, you know, I got up to speed like we talked about, and then I kicked off the planning phase, I guess it was, which Baird: Yep Yeah Brendan: it got to the execute phase. And a lot of the execute phase was actually spinning up, I think, like four different, uh, sub-agents. And the interesting thing there is like I… And this gets to what we were just talking about with like, is Claude or even, um, ChatGPT, you know, are, are they going to pick the model for you? And at least in this case, like I was running Opus 4.8 when I started it, then I could just watch my tokens just, or, or my, my session usage just burn right through it, Baird: Oh yeah Brendan: interesting 'cause like I was doing Fable stuff in a different window, but not using, you know, no- nothing nearly as Baird: Yeah Yep Brendan: I had four sub-agents, it was like, think I was maybe at 40% usage and then, you know, it burned it to I think it got to like 99, was about to s- ask it a question like, "What's gonna happen here?" Like, "Are you gonna be able to pick this back up?" And it got done, and I was like, "Okay, phew, sweet." But Baird: I I will actually say that's one of the things and I'm curious if Claude Code's gotten here Cause like I constantly hit limits Like I'm always pushing it cause I also have this compulsion where I'm like Okay how much do I have left and how many days do I have left And like I'll just push things being like I'm gonna get my full usage out of this And so Brendan: is kinda like a min-max game sometimes, Baird: is yeah for for sure Uh but I've noticed that Codex within the last couple months if you've kicked off a job and it hits the limit it'll Brendan: Yep Baird: you the message and stop you from executing anything else but it'll just keep going It'll go until it actually finishes whatever active pr work it was doing and then it'll be like Okay you you can't do kick off anything else But it does actually execute until it gets done I noticed yesterday and this was when I was executing on the Fable stuff I had planned Brendan: Yep Baird: it kinda did the same thing But it was like the main orchestrator kept working but it kept trying to kick off subagents and those would fail And I'm a I I actually haven't had time to go back and look I'm a little concerned that it didn't actually finish the work because like it told me at the end like after I got you know my session reset it's like Oh yeah we're all done We're good And I'm like Did you actually finish the work that the subagents didn't And so I'm I'm gonna have to explore that But it did actually like go through the full process before it quit even though I was out of my sess session usage So I think it might be getting there as well which I think is super helpful Like the killing right in the middle was just so frustrating because it's like I mean I get it You're out of usage But at the same time it's a terrible user experience Brendan: Exactly, and that, that was my thought too. It's like if I've essentially… Like, these were like three web searches basically Baird: Yeah Brendan: that it was conducting, and I'm sure there's a way that they could probably get there. But at the same time, like, the, the, e- these sub-agents don't have the context while Baird: Yep Brendan: that, Baird: No Brendan: got 10% left," or whatever. Baird: Yeah Brendan: you would… If, if you were actually gonna follow through on that, you'd have to like, what, cut them off, like Baird: Pretty much Brendan: it and be Baird: Oh yeah Brendan: sum it up now," but that's still gonna burn tokens. Baird: Well Brendan: It's Baird: it doesn't sum it up it just literally stops Like it does not Brendan: but Baird: Yeah Brendan: if, if you could Baird: Yeah Brendan: that, that's also gonna burn tokens to like, "Okay, Baird: Oh yeah Brendan: and like, Baird: Yep Brendan: you know. It, it is kind of a weird thing, and obviously we're not gonna solve it, but yeah, that's interesting to see. Like, you've run into it clearly more than I have, but that was… And I didn't quite, I guess, run into it, but it was like just getting up to that point where it was like, "Okay, what's gonna happen here? I'm kinda curious." Baird: Yeah Now the Brendan: maybe going all the way back to where we started is like could we then, when we get our session tokens back or whatever, our usage back, could we pick it back up? Probably wouldn't be that hard. Baird: Yeah You just say keep going and it'll pick it back up But but to your point though like I I've definitely had it where I'm like in the flow generating stuff actually like actively going back and forth then it stops then I have to wait for a couple hours and then I forget that I was even running it And so it's like that is a problem where like it's losing that thread and then you've gotta go back into it and it's like Okay what were we doing again And like so I I think it does all kind of tie into what we're talking about is like use AI to get you off the ground Like get some mome momentum going whether it's a new project whether it's you know man I don't feel like working today but it's like okay kick this thing off get something started and then you get that dopamine hit of like okay here's the thing And you're like oh cool Okay Let's do the next one And and you keep pushing And so like lowering that activation energy just to get something started is such a huge value to me for using AI Brendan: 100%, yeah. And maybe that's a good place to just kind of wrap things up. Baird: Yeah Brendan: I, I think it, it is definitely making people way more productive. It's making both of us way more productive, which is like, it's not even an argument at this point. Baird: Yeah Brendan: but there are still challenges, and I, I think that's the-- in some ways that's almost the more interesting thing now is like how do we navigate these challenges? How do we get the activation energy? 'Cause, 'cause even then it's not always easy and, y- you know, still like we, we don't need to dive too much more into this, but like it… is the, yes, we can get AI to help us with it, even then sometimes it's just kind of like, uh, but then I gotta like remember where, where like the next thing was or whatever. And, Baird: Yep Brendan: can help with that too. But Baird: For sure Brendan: so I don't know if you have anything else you wanna add, but, um, yeah. Baird: No I think we're we're good Brendan: Okay, cool. Yeah, this is, this has been fun. It's good to talk about this kind of stuff and just, you know, the, the little day-to-day things you run into that I'm sure a lot of people that, you know, might listen to or watch this video have also run into it. So, Baird: Well and I yeah I I think it's important too just just to throw the note in there is that like I know a lot of people talk like they've got AI locked down 100 Like it's perfect and they never run it Like it just runs my life for me and I don't agree with that I think that that's either you're not using it very hard or something I so it's really that hopefully people can relate to this You know like see that the real Like we use it all the time and there's still bumps you know And y Brendan: Yep Baird: fun to explore Brendan: Yeah, exactly. It's a, it's like a new version of the challenges we used to have with just coding without AI Baird: For sure Brendan: Now the challenge is more around just like how do we optimize? How do we get the most out of these agents? And, you know, and just the fact that they're so good at what they do now Baird: Yeah Brendan: pretty awesome. So Baird: For sure Brendan: all right, cool. Well, you know, thanks for watching and listening and, you know, feel free to go to Apple Podcasts or wherever you listen to your podcasts. Give us five stars if you want. That'd be awesome. Um, you know, like and subscribe on YouTube uh, yeah, we'll see you next time.