AI Moves Fast. Expertise Still Matters.
Mark and Ryan unpack AI tool fatigue, the pressure to keep agents busy, and why faster output still needs domain expertise, critical thinking, and human judgment.
Start with the full episode, jump into the best moments, or use the chapters to move through the conversation.
Best entry points
Short on time? Jump straight into the parts of the conversation most likely to pull you in.
Thrive In The Gray.
“Mark: discomfort with ambiguity versus thriving in gray areas.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteYour Agent Can Wait.
“Ryan: pressure to keep agents busy 24/7 versus productive work.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteAi Needs Your Context.
“Mark/Ryan: annual planning requires domain knowledge and real business inputs.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteSuperpowers. Not Substitutes.
“Ryan: team superpowers, not replacement; tools and judgment.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this sitePixels Aren’T The Finish Line.
“Ryan: attractive AI pixels still need vector layers and animation production.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteDon’T Outsource Your Judgment.
“Mark/Ryan: critical thinking cannot be blindly delegated; trust but verify.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteThe Plan Meets Reality.
“Ryan: Bruce Willis/Armageddon analogy about rigid plans and changed tooling.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteThe Last 20% Takes Expertise.
“Mark: 80% prototype versus difficult last 20%, then team change pressure.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteMore Efficient. More Demand?
“Ryan/Mark: anxiety, tech demand, Jevons paradox.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteTest The Bleeding Edge.
“Ryan: retain expertise with a toolset while testing new bot agents.”
Approximate discussion-section timestamp. Standalone short publication pending.
Play on this siteBest moments
10 clipsYour Agent Can Wait.
“Ryan: pressure to keep agents busy 24/7 versus productive work.”
Play on this pageTest The Bleeding Edge.
“Ryan: retain expertise with a toolset while testing new bot agents.”
Play on this pageMore Efficient. More Demand?
“Ryan/Mark: anxiety, tech demand, Jevons paradox.”
Play on this pageThe Last 20% Takes Expertise.
“Mark: 80% prototype versus difficult last 20%, then team change pressure.”
Play on this pageThe Plan Meets Reality.
“Ryan: Bruce Willis/Armageddon analogy about rigid plans and changed tooling.”
Play on this pagePixels Aren’T The Finish Line.
“Ryan: attractive AI pixels still need vector layers and animation production.”
Play on this pageAi Needs Your Context.
“Mark/Ryan: annual planning requires domain knowledge and real business inputs.”
Play on this pageDon’T Outsource Your Judgment.
“Mark/Ryan: critical thinking cannot be blindly delegated; trust but verify.”
Play on this pageSuperpowers. Not Substitutes.
“Ryan: team superpowers, not replacement; tools and judgment.”
Play on this pageThrive In The Gray.
“Mark: discomfort with ambiguity versus thriving in gray areas.”
Play on this pageShow notes
Faster isn’t the same as finished
AI makes it easier to get more done—and easier to feel like you should never stop. Mark and Ryan discuss tool fatigue, the temptation to keep agents busy around the clock, and the friction of changing how a team works.
The conversation moves from software review to animation and annual planning, with a recurring distinction: something that looks finished is not necessarily ready. Their practical throughline is to use agents to extend a team’s capabilities without outsourcing its judgment.
In this episode
- Productive output versus keeping agents busy.
- Tool adoption has a learning cost; novelty can add noise.
- Generalists and specialists bring different, complementary expertise.
- Visible polish isn’t a reliable measure of project completion.
- AI-assisted review still needs accountable human judgment.
- Juniors need opportunities to build and retain domain expertise.
- Better tools should give teams capacity, not just raise expectations.
Watch the full episode on YouTube. Chapters and transcript timestamps follow the supplied episode transcript; key moments identify discussion sections, not frame-accurate short boundaries. The clip archive below links to full-episode discussion sections; standalone short publication is pending.
About the transcript
The full speaker transcript below is supplied source text with conservative corrections to obvious show and tool names. Spoken phrasing and profanity are retained; less-certain names remain as supplied. Host commentary about technology and learning is discussion, not independently verified research.
Episode 26 clip archive
These ten distinct entry points use the reviewed clip inventory and matching covers. Standalone shorts remain pending.
- C01: Thrive In The Gray.
- C02: Your Agent Can Wait.
- C03: Ai Needs Your Context.
- C04: Superpowers. Not Substitutes.
- C05: Pixels Aren’T The Finish Line.
- C06: Don’T Outsource Your Judgment.
- C07: The Plan Meets Reality.
- C08: The Last 20% Takes Expertise.
- C09: More Efficient. More Demand?
- C10: Test The Bleeding Edge.
Full transcript
Welcome back to another episode of the Not Brothers Podcast, where today we're talking about the change of pace and technology that has been taking place. with all of the introduction of new AI tools and models and agents and agent agents that we've talked about in the past on the show. we find ourselves in an interesting spot where especially knowledge workers.
are feeling like everything's moving at just a lightning pace, or you're feeling pressure to move faster than maybe you're comfortable with. Or, you know, every day you wake up and there's a new tool and you feel like you're falling behind. And I think that I've certainly felt that when you're like, you know, here's a new tool that I haven't even heard of. Like, holy shit, what am I gonna do about now I need to go learn that one and figure out how to use it and what's the best practices and is it better than the current ones?
And I think that can be a bit of a dangerous death spiral to be caught up in. so we'll talk about that a little bit today.
Yeah, I mean you're you're in a a slightly different place than I am in terms of overall technology adoption. You're on the bleeding edge side, but I certainly feel it on the daily as well, with you know, tools that are out there and from a business strategy and planning and and working and trying to manage accounts. What's the expectation on the account side of of trying to move faster? We literally just even had a an account this week that
use the phrase like I feel like we're burning daylight. And so I'm like, well, actually you you owe us a bunch of stuff to be able to keep moving faster. But the expectation, because everything in the world can move so quickly now, is that we can move faster than we used to be able to. And and that's true. And there are as we were talking pre-show, there are pros and cons to that.
A a pro is that you literally can get more things done in in the same amount of time than you've ever been able to in the in human history. The con is that's a dangerous spiral to to try to keep that pace and feel like you're not burning the candle at both ends at all times.
Yeah, I think we talked it's been some time ago now, but we talked about the the challenge that agents introduced where you almost find yourself feeling compelled to keep agents busy twenty four seven. Where, you know, it used to be you would you know, your average person would clock in at nine o'clock, they'd work throughout the day, they'd clock out at five, and then that's
kind of it. Maybe there's an email or two here and there, but by and large, you're kind of unplugged. And now it sort of feels like you're compelled to have a 24 hour working cycle because your agent should always be busy. And if your agent pops up, if you don't have an agent that's busy all night, then well you just wasted productivity, you just wasted potential tokens, you just wasted potential time. and I think it's a it's a really dangerous rat race to get caught up in.
it's almost impossible to keep them busy 24-7 on something that's productive, right? I've seen people who have crazy loops and will just throw them at stupid problems. I'm a little more conservative with how I'll dispatch agents. I kind of want them to do something productive, not just create a bunch of noise for the sake of burning tokens and creating noise so I can, you know, show off how many tokens I burnt. I'm looking for productive output.
Right? How many how many tasks got accomplished, how many PRs got pushed that d that wound up getting merged? If it just generates a bunch of pull requests that are all garbage and then they all just get thrown away, like that's not really valuable to me. but I think
that we're
We're also seeing this in things like tool adoption. you know, we just rolled out to our team some new tooling this week. it was the OpenClaw V2, and a new agent built on top of that that our team can use and use the new multiplayer mode together and some other small customizations that we've made to allow it to work well within our business. But you know, you hear through the grapevine.
Things and there are certain people like why well, why do we need this now instead of you know some of the other agents that we have or your standard Claude Code subscription, or you know, why can't we just keep using that one? And it's it's a fair question, but it's also kind of eats pokes at like the heart of this issue, which is that things are moving faster than they've ever moved in human history. and to a certain extent, I think that there is value.
In in maintaining tools, finding a tool set that works for you, sticking with that that tool set, and achieving a level of expertise using and wielding that tool set. Sometimes that can be more valuable than the latest and greatest thing. But then I think there also is a necessity to sort of ride that bleeding edge a little bit and at least experiment with it. you never know what's going to come out of it. There are certain things that I've seen.
You know, the bleeding edge for me right now are all of the the bot agents. You know, everybody we've had OpenClaw and hermes for quite some time. We have grokbot, dots, duet, and muse. So trying all of those out, they're all a little bit different. I've found some things with Grokbot that I haven't been able to do with other agents, but it's just been really useful for. so those are workloads that I've been able to change and offload from
my current day to day, which have been really, really nice. those are also things that I can potentially take and pull into my, you know, OpenClaw or other agent harnesses too.
Yeah, I I got on the the beta list for duet and I've I've played around with it little bit. I don't know exactly how useful it is to me yet. Right now it actually feels like it's adding more noise because I haven't really, you know, spent the time to really tone it or to to to teach it exactly what I'm looking for and what I haven't. I mean I've only had access to it like three or four days, so but I can see the potential.
Of being able to have a proactive agent running something on your behalf. But I can also see how that can be incredibly overwhelming for people because something that's designed to be helpful as a cutting edge tool actually takes some work to figure out.
Mm-hmm.
Otherwise it really is just more noise. It's like, okay, you're reminding me of something I already knew about and was going to handle anyway.
Or you're reminding me that there's this email from my accountant that I also know is in my email and that it conflicts with calendar invites that I already have. Great. That's sort of helpful, but I already knew that. It's in my inbox. so it there's there's this balance between, you know, adoption, creating more or less noise, more or less work for yourself. and I I think to a lesser extent, people that are on our team that are having
or any team, honestly, that are struggling with adopting new tools, like, you know, whether it's in our case the open call two point z or experimenting with a duet or any of those other things, I I don't think I don't think change management is necessarily the culprit in a lot of cases. I think it's it's related to this idea of this feels like a lot more work because now I have to go learn this thing.
That's gonna take away from my otherwise normal day-to-day. And so I'm not able to accomplish things that I may normally be accomplishing because I'm learning this new tool stack. And I don't know whether this new tool stack is actually gonna be productive for me or not. And people like you who are just constant tinkerers, you you you just consistently want to tinker and and figure things out and see what's gonna work best and and and how like how it fits in your workflow or it doesn't fit in your workflow.
But I don't think the average person is is wired that way. I think the average person I won't say gets stuck, but gets comfortable in a particular flow and they've established that flow. They know what it is, they they know what to expect from it. And when someone wants to disrupt that flow, it gets uncomfortable.
Yeah, I think I mean we've we're no stranger to this at all, right? We've throughout the the years that we've had Oodle, we've moved tool stacks
A number of times.
Probably too
often. But
I mean in hindsight we should have just stayed in base camp the whole fucking time. We s we started in two thousand nine with the full base camp stack, right? Like the ca base camp campfire and high rise. We should have just stayed there the whole time. And we'd have never had a problem. But,
Yeah.
you know, we were led astray by plenty of salesperson. and we learned a lot along the way, but you know, part of that has been
I think that's that constant change management, right? To us, moving from one tool to another for a for a series of benefits is a no brainer. It's like, all right, well, we're just gonna pack our shit up, we're gonna move it over, we're going to establish some new processes, we're gonna figure out how some of this stuff works. Maybe we don't have everything figured out a hundred percent when we move in, that's okay, we'll decorate the place when we get there.
and as we've grown, particularly when you know, when we hit like twenty five, thirty people, it started to feel
excruciating to make that sort of change, right? When we would move from a you know, even if you added a platform, I remember adding Notion to the table at one point because it was a better way to take notes and collaborate and do some things and just the the amount of pushback was astounding. And eventually I was like, you know, listen, this is an optional tool that you can use or not. I don't fucking care. I'm just telling you I'm using it.
And I found benefit in it. If you want to use it and find benefit in it, great. If you don't, I don't care. and I think that like where we're at now sort of exacerbates that there's going to be even more change management management necessary, and in a landscape where people are probably even more concerned, right? I think that you know now you're going into using new tools.
using new tools that you're not comfortable with, changing how your workflows are supposed to work, in some cases, changing fundamentally how you do work,
Mm-hmm.
while also being told that you're gonna be replaced by the robots and you're not gonna have a job. And I think that all of those stack together to create a lot of anxiety, most of which is unnecessary, right? What we've seen proven time and time again is technologies like these create more
More demand than they than they replace, right? It's happened with every single t cutting edge technology like this in the past. And it's happening again now. Developers are in all-time high in terms of demand. Because you have companies who can build bespoke software who couldn't build bespoke software before. so
What's that principle called? We we did a whole episode on this. I can't remember the name of the thing. Yeah. Jeep,
It's Jevins Jevins Law or Jevon's paradox.
Jevon's paradox is what it is.
Yeah. So I think that's one that like people are gonna you know, we've had lots of discussions within our team, right? The dev team in particular, we had to have after Rails World, we kinda came back and delivered the message that was delivered that was bestowed upon us by the Prophet DHH.
Ha ha ha.
And we and we told the team straight up, we're like, listen, if you're writing code manually or you're staring at your bot I took it a step further and said if you're watching your agents do work, you're doing it wrong. And if you're not careful, you will be passed up and replaced. Because somebody else who's doing it differently will do it.
And it was interesting, even within our small team, some of the conversations that came out of that, and just you know, you can feel some of the discomfort. And, you know, we have a a wide variety of experiences within the team. And usually the the it's more of a struggle for your more senior talent. Because those people have ways that they used to do things. They have, you know, routines and things that they've learned and expertise. And there becomes this question of like, well, what's the point in the expertise anymore?
whereas your junior people, they're ignorant to what they don't know. So to them, they're like, Great, this is what I've been doing anyway. It's because I don't know what the fuck it's doing. I just tell it what to do and it does it and it and it works. which is not to say that there's not a value for domain expertise. I think you know, I was just working on a a thing last night with one of the other members of the Omarchy team, and I have a a
project that I built called Oma Real, which is, you know, a screen recording tool to make it easier to like record your screen along with your webcam and then kind of do some light editing. does the sort of like zoom in, follow your mouse, put the webcam over top. really basically the only videos that I would ever make, either for our team or for, you know, products or whatever it is.
So I created that tool and I I recently ported it over. I ported the back end over to Rust. The front end is all still QT and there's some interactions with Hyperland and and being able to show things but not record them that are a little weird and other stuff. But anyways, OutFox and I were kind of going back and forth and that dude is wicked smart. he's the creator of Quick Shell, he's you know on the team at Omarchy now working on working with us.
And he was just kind of picking through some of the things in the code and and critiquing things from his perspective. And there were a few gotchas there that he found that he was like, you know, this thing, never use this. This is a huge like performance problem because it causes every frame to get passed off from the CPU to the GPU, and it's a huge bottleneck. And it creates hitching and and performance issues. And I was I'll be damned. I've been chasing that issue.
And there were a few others that he kind of called out in that process. And he knows these because he's encountered those before. He has this domain expertise around Wayland and Hyperland and QT and sort of this combination of all these libraries and things and how they work and the inner workings of them. So with that domain knowledge, he was able to give some suggestions. He was like, you know, ask the agent, like
you know, what domain things we might have of might have not followed appropriately or something. I forget exactly what it was how it was worded. And it came back with about a dozen. It found about a dozen things and you know, we vetted them and found that most of were indeed things that should be fixed. They were crucial. the application works from my perspective, but they're all little time bombs. They're all just sitting there waiting for the timer to go off.
Or, you know, their landmines, right? It's waiting for somebody to step into the right situation with the right set of circumstances and then explode. so that domain expertise in maintaining that just helps avoid that and helps to drive the agents more directly to say, hey, these are the things I need you to look at. Another example excuse me. Another example for me lately has been.
we've been working on compatibility for the the new N1X laptops from NVIDIA. And that's in addition to Snapdragon and the GB10, the DJX Sparks, which are all ARM-based systems. So Arch 64 versus X86. And in that process, I've been driving a lot of the kernel dev behind it and you know reviewing.
patches and things that come in. And luckily the the main kernel maintainer on Omarchy hooked me up with some skills and tools and things to help dig in and and find. it's helped me identify problems that I would have never noticed myself because I don't have the domain expertise. But he does. and being able to combine those two together I think is really, really powerful. Kinda went off on like a tangent there, but
Well, I think you're you're scratching at something that's a little bit different though, right? A couple episodes ago, I said that the generalist is is kind of going through an enlightenment era. Like this is this is the age of enlightenment for the generalist. Someone that knows a lot about or a little bit about a lot of things and can use Asians to fill in a lot of the gaps. The the opposite of that is domain expertise. And that's that's exactly what you're hitting on. And I think there's there's usefulness in both places. It just looks different.
than it used to. And, you know, one of the most compelling things, but also for those that don't like change, things that we are we're experiencing right now is that you can take something eighty percent of the way without really involving someone with domain expertise. Not really, right? You can you can put it get it to a place where it feels navigable, functional, and almost like something that is is polished enough to be able to give to the wild.
But then to your point, you give it to give it to someone that has domain expertise, and that last 20% is incredibly difficult to get right. And
Shredded. Yeah.
it's what takes all the time. And so you can get the 80% and you know call it a couple days, and the 20% takes weeks to get right. And while that's always been true in some ways, that that that curve is happening far faster than it's ever happened. And the volume of things that you can get through.
This is probably the reason that even our team probably feels a little bit overwhelmed at times because you're there's so much thing so much that you can do so fast that it feels like we're changing things on a weekly basis. And really that's not the intent. That's not the purpose. The purpose is to continually evolve and get better. and and to continue, especially for the tools that that we've built to enable them internally to do work differently, it's to make those tools better so that their jobs are are easier.
than they once were. And yeah, it's it's just it's an interesting conundrum.
think it's it's no secret either that time like time estimates are completely screwed. You know we I was talking with somebody on our team recently and and they were like you know I just the timelines always feel off and you know I kind of just want my like my weeks back. And I was like I don't think that's ever gonna happen. I I think that you know
And we don't even have a good process. Like we used to, you know, we used Scrum for a long time. We used sort of like a shape up methodology for a while. And you know, now it's kind of wild, wild west. because what we're what we're trying to figure out is how do you time things in in a gentic world? And even agents don't know how to do it, right? If you spec out a project right now with Claude, he'll come back and he'll be like, you know, all right, this is
Eight sprints worth of work and it'll take two months. And you're like, buddy, this is gonna be you for the next twenty-four hours.
Right.
And it so like they're vastly off when it comes because they're still making estimations the way that we used to make estimations. And those aren't exactly accurate, but it's also not accurate to say, you know, sometimes I mean sometimes you can one shot a project and it's done in, you know, a matter of hours.
It would also be very dangerous to just assume you're gonna one shot every project because it's doesn't go that way. Most times it doesn't go that way. So there becomes this challenge of like how do you measure results? Sometimes you're not taking sort of a linear path, you're taking more of a matrix approach to things.
When even even that is I mean,
It's a it's a real challenge.
we're talking about change management in in a lot of ways, and even that is incredibly challenging because the average person wants a process to follow. And when the process breaks, and they always do, it's just a matter of either time or volume or size or something, the process is gonna break. Where we have right now is tooling that changes so frequently and capabilities that change so frequently.
that any process you try to put in place breaks on a very regular basis. And adapting in that world in a professional services environment is incredibly difficult. Now we we could put our head in the sand and say, nope, we've always done it this way. And so we're gonna henceforth continue to do it this way because we know it works. That that is a solution. Yeah. That is a solution.
Well, you're out of business.
And it'll it'll work for some period of time until it doesn't.
Yep.
But I think that's where, you know, I've certainly had some run ins with our PM team. because they are that mind, right? They want to know what's my waterfall look like? What's that timeline look like? How can I hold you accountable for delivering this thing in accordance to our delivery plan? And, you know, I always come back and I feel like I feel like Bruce Willis in the Armageddon movie.
Right when they're about to abort the mission because they haven't achieved the proper depth as outlined by the NASA drill time card. And he was like, What the fuck is this? Like, this isn't this isn't how this works. This is just what happens when you drill. So you guys continue to jerk off in here and talk about this shit. I gotta fucking hold a dig. And like we've certainly had plenty of conversations that have ended that way where I'm like, listen, guys, like the tooling's changed.
The world's changed. I can't give you a nice neat little timeline, nor am I really interested in trying to make one up. So I'll just let you know. And we'll go from there. And
And
and you're picking on a completely different issue, but it it is it is one hundred percent an issue. And that's like how in the w in the world of professional services and in our world, world of professional services that we complete on behalf of clients, how do we how do we make sure that we're delivering to clients with some reasonable expectations while also giving our team liberty to experiment, change, you know, establish different processes than existed last week based on new tooling.
Like those are those are all really they're all diametrically opposed to each other, right?
Like
Well what
you don't want to do is you don't want to wind up, you know, accidentally creating a slope where
Correct.
you have you have priced yourself into and timed yourself into a situation where basically you're just a middleman for the AI agent and you're delivering directly to clients AI outputs because now you're useless. and also the quality of work just isn't gonna be there, right? There's always there's always a smell.
That comes with like this is a direct re derivative result of something. We have a client right now that we're doing a bunch of animation work with, and we've done plenty of concepts and and some assets have been AI generated and those sorts of things, but it does become a difficult thing to help people understand because in some cases, like I can go to Claude and say, you know, mock up this frame of a video and it creates it in 10 seconds.
But you know, if you look at it closely, the you know, the doors the doors don't have door handles and the windows open the wrong direction and there's a sign hanging on nothing. And you know, there's all these little things that are just kind of off, right? there's also this thing's this is animated, right? What Claude generates is a bunch of pixels. It's a flat element.
So now we have to take that and extract it into its pieces, right? If you have you know foreground elements and something moving behind it, now you like each of those elements have to be individually illustrated and turned into vector objects, not pixels, which is still by and large done by hand. There's some AI systems that can help with that, but like you're still tweaking and fixing by hand. Then you're positioning those assets within you know a frame.
And then moving and doing whatever else needs to be done. So, you know, how do you it it becomes really tricky because there's this this paradigm that has always existed where people equate the
The
The state of a project to the state of the UI. Right? Where in that sense, right? If I generate a frame and it looks right, I'm like, well, we're 90% of the way there. Like, no motherfucker, we're 10% of the way there. I realize that to you this is 90%, but it's not, it's 10%. We gotta do all that other shit to even get to 90. And I think that becomes more and more difficult to bridge the gap of.
And make sure that you know, especially when working with clients who maybe aren't domain experts, haven't done this stuff before, that it's being translated in a way that makes sense and they can understand, you know, how where do these tools actually act as accelerators? So when we think about things like testing, we think about things like content migration, AI is an insane accelerator. It helps helps do that stuff.
To a a crazy degree more than anything we've ever done before. When it comes to asset generation and actually building things, sometimes it can be holdback. sometimes that's not actually something that helps move you forward. It it can get you stuck in like, well, what about this idea? And what about this idea, and what about this idea? And now you're weighing three different ideas versus like, all right, guys, at some point we gotta fucking pick one.
Yeah. you're also you're you're picking on something throughout this conversation that I think is important. And that's as knowledge workers, how how do we, and we're still figuring it out, how do we message and make sure that clients and other people in the community or potential partners or anyone else in an agentic world understand that the value of that domain expertise is still incredibly important. Because while you can
fire up your own clawed account and produce some things. To your point, unless you know how to do all the things on the other end, it looks like you're much further along than you actually are. And it's that domain expertise to be able to tie it all together and actually make something super useful and super refined that is the difficult part. And and why you need those knowledge workers that that not just know how to use those tools in a slightly different way, but also know to look for the things that are are
gonna be the landmines that we talked about and can get ahead of those landmines before they become a thing.
I mean it's an interesting thought because a as you were saying that, I thought immediately about a pull request I had to close last night. And not to say anything negative of it. I you know, the that person certainly did a lot of work to get it to where it was, but you know, I I looked at the pull I looked at the it was a kernel level pull request. I looked at the pull request, you know, Greptile, which is an automated AI review tool, gave it a five out of five, said, Hey man, this thing's awesome, it's ready to go.
That person tested tested their code, it worked for them. A couple other people had tested it, it worked for them. So I kind of walked in with the expectation that, you know, this is ready to rock and roll. I just just out of habit, right? I just hit that little changes side and you know, pop the hood, so to speak, and look underneath and take a quick look around, make sure things seem in order. And
And I started seeing stuff that I didn't particularly care for. I was like, Okay, well, you know, maybe there's just some leftover remnants here. We can kick those out, not a big deal.
And then I realized there was a 10,000 line of code patch file. And what had happened is it seems like somewhere along the way, a whole bunch of patches just got smashed down into one big one, which is just a recipe for disaster. Like it can
Thank you.
it contained things that impacted that tends to hide things, right? When you're looking at 10,000 lines of code, it's real easy to hide something in there. And and after running it through some reveal tools and other things, like we've
found a number of things that were in there that would have had sweeping impacts. So I was like, hey, I gotta, I gotta close this one. Like we need we need to do some work on this. but I think that illustrates kind of the same point, right? I had people messaging me saying, hey, this is ready to go. And by all accounts, by all visual metrics, it was good to go. But it really wasn't.
And I would have paid dearly for it later. It's just a matter of when. and I think that's a it's an important, like, cautionary tale and also reminder for the value of maintaining both maintaining domain expertise and achieving domain expertise. Because right now, with all of these tools and with AI, the people I worry about the most are juniors.
Mm-hmm.
I don't know as a junior anything.
what the future looks like. I still find that there's a lot of value in achieving domain expertise at where and when that's possible. so even, you know, within our team, we have some junior t folks and, you know, part of even one of the ones that was a junior dev is I made him write code without AI for the first bit that he was working because I was like, I want you to know how this stuff works.
Eventually I don't need you to know how to write it, right? I don't know how to write a fucking if statement in Rust by off the top of my head. But I can tell you when you should use one. And that's what's important that was what was important to me is like understand the tools you have at your disposal and how those tools work together before you just go wild using them. Because I do fear that if you if you never evolve past sort of that like
that vibe coding pathway where you really just don't understand how anything works. I think there does become like a this like glass ceiling of what you can achieve.
At least right
Well, and
now.
y you said junior anything, and I'll give you a completely different example that has nothing to do with code at all, which is annual planning. We're getting we're going through our own annual planning exercises right now. Well, domain expertise, I've done this twenty times from an annual planning perspective, both for Oodle, for clients and for past Fortune five hundred organizations that I've worked for. I have a high level of domain expertise as to how and what goes into annual planning.
And exercises that you can do to get the most out of your your leadership team or your team in general of what you should watch out for, how you organize those thoughts, how you put them into something that it resembles a framework, how you then prioritize that framework, how you then take that framework and put it into an actual an action plan, how you create revenue plans and and and and and and right. That level of expertise is is something that you can prompt engineer your way through AI without ever understanding, but it doesn't.
mean you're going to come anywhere close to the desired outcome because it's it's going to suggest as part of that's part of this journey things that just don't fit. The things that that are going to add a add a tremendous amount of of of garbage to the equation. Ten thousand lines of code in in the world of development, but unnecessary documentation, unne unnecessary exercises and unnecessary XYZ for something else. Or not enough.
Or not enough. I think if
you know, if we asked I could ask Claude right now to give me an annual plan for Oodle. And I'll bet it'll come up, it'll give me something. It's never gonna
Mm-hmm.
not. But it'll be complete dog shit. Because it's not gonna be based on f our financials, it's not gonna be based on our client base, it's not gonna be based on historics, it's not gonna be based on any of those activities that you just mentioned. Like
There's a lot of inputs there that what you're describing is how to get the inputs, right? And then how to feed some of those inputs to agents, perhaps, and say, you know, here's historically some of our our our client billings. And maybe it spots something and it goes, Hey, did you know that this client always spends more money with you in the fall?
Mm-hmm.
You're like, I'll be damned, I didn't know that.
Maybe I could have spotted that. Maybe we've kind of felt that. But you know when you get a few hundred lines of of data that's off by a little bit, eventually it all kind of blurs together, and you know, you can't find some of those small trends. or maybe it lines those up with emails, and it's like, did you realize every time you email this client, they spend more money with you? You know, I'll be damned, I didn't know that.
So, you know, here I thought they were spending more money in the fall. It turns out I just happened to email them every fall. Maybe I should email them every quarter. Right? Yeah.
Once a week.
Eventually there's diminishing returns there.
Yeah.
But I d I think the the cautionary tale is that domain expertise isn't going anywhere. And I do think you're you're right. I think people that are coming in, and I see it even in, you know, call it mid-20s, that I've interacted with and worked with, their very first response to anything is to go and and prompt. And there's nothing necessarily wrong with that if you're also retaining the knowledge along the way. Whereas
Hard to do though.
It's really, really hard
It's extremely hard.
not to let those skills atrophy and not to delegate your brain to agents, especially if it's something that you're new to.
Yep. It's yeah, it's I I too do not know what the future holds for anyone that is junior at something and has not acquired that level of of domain expertise. I'm hopeful. I'm hopeful that the next evolution becomes sort of like cell phones and the internet became to dictionaries where
Well, I think that we
have I think that we have the tools, it's gonna be how we apply them, right? You know, there's there's some in terms of like we were just talking about, you know, how things get committed to your memory have completely changed with using agents. So as I think about, you know, something as simple as taking notes. It used to be the logic was if you take notes by hand versus type them, you remember them better because you have that.
that hand eye coordination ask piece that comes with it.
That's all well and good. now we have agents taking notes for us, so nobody's even typing the motherfuckers anymore. So there's a further disconnect from that. Now the exchange is perhaps they're more accurate. Perhaps I'm able to pay closer attention in the meeting now, versus trying to split my attention between taking notes and being in the meeting.
So I actually recall more because I was paying better attention. And then the notes just help jog my memory. there's also the distribution of the notes. So one of the things like I've used Grok for is I have kind of agents per client that help me just keep an eye on everything. so if I'm getting ready to go into a meeting, it'll know and it'll
It'll
Give me kind of a brief of what's going on and say, hey, you know, here's what was discussed in last week's meeting. Here's what's going on that I've been able to find. Here's some stuff that's due that you might want to be aware of. and that's something I never would have had before. I stopped taking written notes years ago. A, because my handwriting is atrocious, B, I can type faster than I can than I can write. And C, it's not referenceable. I found that I had.
You know, I had mole skins that had perfect layouts and table of contents and I'd keep track of where everything was. And then I'd never read it again because I can't command F in a f fucking moleskin.
Yeah.
You're you're not wrong. And there there's we could go into the evolution of brain development and all these different things, right? Where where those things apply to, you know, kids versus adults versus, you know, older adults, etcetera. But I I do think that there's
In in a period of a heightened amount of change, there will be skills that atrophy over time. We're we're picking on a couple of them. Note-taking. the ones that I worry most about are the ones that are incredibly difficult to replace. So critical thinking. If you let critical thinking be outsourced to agents, that's something that is incredibly difficult to replace.
And and very, very dangerous, right? We you actually I mean with all agent pr all agent work, it should be a trust but verify activity. And if we lose our ability or if anyone loses their ability to have that critical thinking layer and it's just trusting the robots, that's that's not a good thing for for yourself or for humanity.
No, I mean those those are the situations where they make weird decisions. You know, we we talked about this with like code review and those sorts of things, I think, a couple of weeks ago. And, you know, i everybody looks at that as like a binary equation, but it's really not. And even now, like I've you know, I have plenty of processes where I have agent dire driven reviews and they can merge pull requests and you know, if certain criteria are met and then but I still have
the the the overlord layer of it where there's still some aspect of all right tell me what you did. Tell me what you did. Tell me why you made those decisions. Tell me what those looked like and every so often I'll poke into one here and there and see maybe it didn't make a mistake. Maybe it didn't. Maybe it merged something I wouldn't have merged. and it's for the good. Like it it found something that I wouldn't even have thought about and fixed something.
Those are all interesting interactions and I think that, you know, that really points to what we've been talking about the whole time is that I think it's necessary for everybody to have a level of comfort with shifting and changing and just sort of riding the new waves that come in and see what you can find out of it, because I think you will often be surprised it's not whatever the sales pitch is, right?
I can I can say that definitively. Whatever the sales pitch is for whatever the tool is, is probably not true. if they're pitching that you can get rid of your team and replace them with agents, bullshit. now what you can do is you can give your team superpowers. And I think that's what our approach has been within Oodle is our view of AI has been the entire time, with any tools, AR or not, our goal
is to give our team superpowers and make everything better, faster, easier, so that they can free up brain cycles to spend on the actual outputs that we're delivering. And I think that's you know truly what everybody should be focused on.
I agree. And I I'll put it slightly differently. I think if pre AI you were a person that was uncomfortable in gray spaces, in gray areas where you're just not exactly sure what's the what step is next, exactly where do I go next and and kind of needed needed that very defined roadmap, you're gonna have a bad time because it's way worse than it and it's it's not gonna get any better anytime soon.
If you were a type of person that thrives in gray areas where you're always finding new ways to do things, you circumvent the rules because you can see a better path. You reorganize things because it's that's just a fun thing for you to do. that is a person that is gonna thrive in in an agentic area, an agentic world. And I think we're seeing that even amongst our teams and and amongst clients and how they've interacted with different things, the ones that that are embracing
the opportunity to live in a bit more of a gray area are the ones that are winning. And the ones that are are either afraid to move forward and make mistakes, or they are have the inability to define what their own next step looks like and need to be told what the next step looks like, those are the ones that are struggling the most.
I agree. Till next time.
Till next time.