Introducing the ADLC: The Agent Development Life Cycle
In this episode of "How Many CTOs Does It Take?" podcast, hosts Scott Porad and Brad Hefta-Gaub" open with Super Bowl reactions and a meme about non-fans describing plays, then pivot to the ai.com halftime ad, where heavy spending reportedly led to a site crash for a Claude wrapper. They discuss MCP servers and "Rent A Human," noting gig marketplace challenges like trust, KYC, and fake profiles, and compare MCP hype to Facebook integration in 2010. Brad highlights AI agents using MCP servers for observability as a major debugging unlock, prompting nostalgic ancedotes from Scott about Microsoft Log Parser and Ajax. In their discussion about the importance of branding tools, Scott and Brad coin "ADLC" (Agent Development Lifecycle) for integrating agents into existing SDLC/CICD with human merges. After a brief tangent about O'Reilly cover animals, Brad argues this approach beats skeptics and vibe coders while changing roles, pressuring product ideation, and reducing the need for separate QA via automated and browser-driven testing.
Resources:
- How Many CTOs Pod: https://howmanyctospod.com
- Scott Porad: https://www.linkedin.com/in/scottporad/
- Brad Hefta-Gaub: https://www.linkedin.com/in/bradheftagaub/
#TechPodcast #EngineeringPodcast #DevTalks #PodcastForDevs #HowManyCTOs #Podcast #CTOs #CTOPodcast #ChiefTechnologyOfficer #Technology #Engineering #SoftwareDevelopment #SoftwareEngineering #TechLeadership #EngineeringLeadership #EngineeringCulture #TechDebates #AI #AIRevolution #AIAssistedProgramming #AINews #SDLC #ADLC #MCP #Ajax #QA #QATesting #MCPservers
Brad: I think the people that are basically integrating agents into an already robust software development lifecycle,
those are the teams that I think are probably being the most successful.
Scott: Hey, Brad.
Brad: How's it going,
Scott?
Scott: Good.
How are you doing?
Brad: I'm doing great. Should we record a podcast?
Scott: Let's do it.
Brad: Welcome everybody to another episode of the How Many CTOs Does It Take podcast.
I'm Brad Hefta-Gaub and my co-host is...
Scott: Scott Porad.
Brad: Hey Scott,
how's it going?
Scott: Well,
I mean, I got a couple of things I want to talk to you about, Brad.
Brad: Yeah.
Scott: The first is I always introduce you as my trusty co-host, but then you just enter me as your co-host.
And so I'm just wondering what's going on there.
Brad: Are you feeling triggered over that?
I apologize,
sir.
I consider you...
Scott: Oh,
Brad: I
Scott: I'm just just
Brad: trust those.
Scott: hassling you, Brad.
Before we talk shop,
can we talk sports ball?
Brad: Let's talk sports ball.
I mean, this is going to get released like a month and a half after the Super Bowl,
but we should still talk sports ball.
Scott: Okay, did you care at all about the Super Bowl?
Brad: I watched it. It was awesome.
Scott: Yeah, it was super fun.
Brad: It was so fun.
Scott: It was really fun.
Brad: I could not figure out why anybody thought the Patriots had any chance.
Scott: I don't know either.
Brad: I mean,
I just didn't think they had a chance.
And it was pretty obvious from the first seconds of the game that they had no chance.
So there you go.
Scott: I would agree.
We won't stay on this subject long.
I just say I really enjoy football.
I find it a very intellectually enjoyable game.
Brad: All right.
Scott: People say it's a chess match and it's kind of like chess.
Brad: I think that the way the Seahawks play the game.
Scott: Yeah.
Brad: I agree.
I have you seen that? Have you seen? OK,
well, as long as we're talking like sports ball, we can talk about like other memes that are out there about us. Have you seen the meme yet?
Have you seen the meme where apparently a whole bunch of people did this where they had their their this tended to be like,
you know, male female couples. But it was like the big sports fan was forced to the football fan was forced to look away from the screen.
while their partner described play-by-play of what was happening and as soon as i spent as soon as i spent 10 seconds watching one of those the algorithm pushed like 100 others at me and they were just absolutely hilarious to watch
Scott: I'm going to have to take a look for
Brad: that. it was hilarious
Scott: I can tell you how that would go in my house.
Brad: Yeah.
Scott: I just don't think it would go very well.
Brad: Well,
that was kind of the...
Scott: Well,
I think the problem I have is not that Lisa wouldn't be able to describe what was happening,
was that she would find the whole exercise so...
exasperating that she wouldn't want to do it in the first she's like really you want me to do what
Brad: these were clearly like the in the case in the ones i saw the partners were very much interested in the idea it was mostly the sports fan exasperated at the non-technical the non-sports fans descriptions of what they were seeing you know like like you know they didn't know how to read like first and 10 or you know second and
and five or something
Scott: Oh,
Brad: like that and so and so they were and they were like well the like one of them was calling them the what was it like the bluebirds versus the pilgrims that was what they were called oh a pill oh the pill oh the pilgrim guy just got oh oh that looked like that hurt what happened what happened
Scott: this reminds me of two other things,
one specific and one general.
Very specific little pet peeve I have.
If any of the 39 listeners out here have any relationship with NFL football production or college football, any form of football production,
you know, it says on the screen the score and then it says like first and 10.
Brad: Mm-hmm.
Scott: And I wish it just said where the ball is at.
Brad: Right. Well, yeah.
Scott: first and 10 on the 40 like think of how much more information you would get just with that and you could do it with plus i would do it like like first and 10 plus 40 means you're on the other guys for your first and 10 minus 40 means you're on your 40 plus minus and that those three characters plus and then two digits plus or minus would
Create a tremendous amount of more information for about what is happening in the football game.
Brad: Right.
That's
Scott: So,
Brad: true.
Scott: Well, yeah,
Brad: that was another thing was that, you know, how the on the yard lines, there's like 40 and they have hash marks pointing in the direction of the of the direction towards the goal on that side,
Scott: yeah,
Brad: you know, so a lot of these people that were not football fans were describing it. They're on the negative 40. They're on the negative 30 or something. And the fan was like, you know, the person that was not able to see the screen was like, there's no negative 40 yard line,
Scott: right.
Brad: you know, or whatever.
But anyway,
Scott: Okay, here's
Brad: it was
Scott: my big
Brad: it was pretty funny.
Scott: picture.
Between the Super Bowl and the Olympics, which either started that weekend or the next year or something,
NBC,
that's who produced both of them, their capacity, these are massive productions with,
I mean, we talk about software development here,
right?
But just to talk about the technology operation behind.
Brad: These things
Scott: The orchestration
Brad: happen.
Scott: is just super impressive to me.
Brad: Yeah.
Scott: I actually,
I would love to learn about it just to understand how they do it.
Brad: How about the AI.com halftime commercial where they spent what I heard was they spent 70 million on the domain they spent on the domain name AI.com they spent another 10 millions on that ad and the whole website crashed and it's basically a open claw wrapper for everyday people that's what that app is open claw they don't exist
existed like three weeks ago.
Scott: Wow.
The truth is, I didn't see it.
I was locked in on the game and for a lot of the commercials,
I kind of had to get up and go.
get some snacks and like let my energy level come down so that I didn't have a heart attack
because we I was I went to a friend's house they were kind of having a little Super Bowl party and then there was where the tv was and over in the other area they had food so I'd get up and go get some chips and dip and stuff
Brad: Well, I'm trying to tie it back to the How Many CTOs podcast.
So AI.com.
is basically just a wrapper it's a wrapper on open claw or yeah you know old claude bot whatever it's called these days you know molt bot claude bot open claw it's just a wrapper on open claw and they just dumped all this money into a domain name and a super bowl commercial and the site crashes and you know people are disappointed
Scott: And what do you think that they didn't, like,
clearly they did not engineer.
That system correctly
Brad: I mean, I have to imagine this was like a super – I mean the whole concept that you're just going to pick a website that somehow wraps OpenClaw,
like OpenClaw didn't exist a month ago.
So they had to like invent this whole business model and try to launch this thing in a month. That's a huge undertaking.
Scott: But couldn't you just like put it in somewhere in AWS with some sort of auto scaling on there
Brad: Oh,
Scott: and it magically
Brad: yeah,
yeah,
Scott: works?
Brad: sure.
It depends.
I mean, there's still work to be done. But yeah,
right.
Scott: But couldn't you just get AI.com to do that work for you?
Brad: That's how it should work, right?
Scott: Couldn't you just go to Lovable and
Brad: And have love. Maybe it was a lovable app. That'd
Scott: you
Brad: be funny.
Scott: make me this thing?
like isn't that how it works okay here's something crazy i know some people at taskrabbit listen to this podcast and i know that one of them who listens to this podcast is going to be listening to this thing right right now i've now got people at taskrabbit i shouldn't say i've now got people it's not me we have now got people at test whatever there are now people in the world using lovable to make effectively powerpoint presentations
Brad: Did you hear about the, did I already say that?
You say this,
there's an MCP server that somebody built,
right?
MCP servers being the thing that allows the AI agents to go do these requests.
Someone has built an MCP server called Rent a Human.
Scott: Oh, have I heard about it?
Look,
what business am I in right now?
Brad: Rent a Human.
Scott: I mean, yes.
What is your impressions of ventriloquism?
Brad: Oh, man,
I have so many thoughts.
Where do you want me to start? Like,
well,
Scott: Anywhere you want.
You brought it up.
Brad: I mean, I think it's like there's a version of this that's like.
the beginning of the apocalypse right like
Scott: Right, this is the beginning of the matrix
Brad: the robots exactly the robots are basically renting the humans to do stuff for them you know there you go that's one thought i have i think that another another thought i have is that to your point taskrabbit is a totally valid it's a gig economy it's a totally valid business model and and there's plenty of things that people want
other you know there's value in being said like i can do a task for you whether that be helping you put a bookshelf together or you know some kind of like a courier service across town right like all those things can kind of be done by these um you know gig economy
Scott: It's
Brad: type systems yeah
Scott: not just TaskRabbit. I mean, there's Upwork and, you
Brad: fiverr
Scott: know, all the fiber.
Brad: yeah
Scott: There's a million of them.
Brad: totally
Scott: So this one, I mean, we've looked at it, obviously that came out and that blew through TaskRabbit.
corporate like a storm like everyone knew about that in an instant and
Brad: Well, has TaskRabbit built an MCP server?
Or is that what you said something once about you were describing to your CEO what an MCP server was?
Scott: let me come back to that in a second because it's an interesting topic uh rent a human is a database of profiles that you can access via mcp server it's not clear that they have a monetization model they'll figure it out they'll take a cut of the transaction they're not supposed
It doesn't appear,
what I'm saying is it doesn't appear that they facilitate the transaction at the moment.
Brad: Interesting.
And I'm assuming that I'm just going to put my little,
you know, I'm going to wear my business hat a little bit, you know, not my tech hat,
right?
Once you start actually trying to facilitate those transactions,
you have a whole bunch of hard work to do around know your customer and, you know, validating that both sides of that marketplace are both are valid,
right?
So
Scott: Yeah,
Brad: that's obviously business intelligence.
That's a business strategy backlog that is much harder to solve than just I wrote some code to have a database of a bunch of profiles.
Scott: Yeah, you know, when we think about, it's not just TaskRabbit, you know, TaskRabbit is the third gig economy marketplace I've worked on.
And fundamentally, we think about our competitive advantages,
as at least I think about it as your brand,
because in a gig economy,
in any economy marketplace,
there needs to be trust.
And so if people think, you know, Craigslist doesn't have a lot of trust, they even got, you know, signs in there, you know, watch out for these scams, don't
Brad: Yep.
Scott: send money,
right?
And so your brand is really important.
The quality of your supply,
like which influences your brand, meaning like are the workers in your marketplace, the people offering services,
the humans,
Brad: Are they
Scott: do
Brad: are they have quality?
Can they actually deliver on what you claim they're doing,
whether that's a dog walker or a car driver?
Scott: Yeah, let's talk about car drivers. I mean, for a while, Uber had a problem with assault,
you know, and so on and so forth. And they probably had a very few number of incidents.
sense but it's just a spread you know oh my god watch out your uber driver is going to kill you and now they've got all sorts of tools that they have in their app and so forth to give people safety so the your your well that actually leads to the third piece and then your technology right and so to get a bunch of people to sign up and put a profile out there to say i'm available for work is clearly easy you and i could vibe coat that by the end of this podcast
Brad: Right
Scott: Right.
Brad: I have an agent working on it right now.
Scott: Yeah.
And they say they have half a million profiles in there or something like that. And I got to assume a ton of those are fake because that's a big problem with these again.
Brad: Yeah, sure.
Scott: So, yeah, they haven't gotten to the hard part of the business yet. The software is the easy part of the business. In fact, this is kind of a segue into the CTO thing.
For a lot of our businesses,
most of the businesses I've worked in.
The problem has not been the technology.
We have had the know-how, the ability to put together the technology.
We haven't had the know-how of what the customers wanted and how to execute it in a way that did things that...
delivered on the needs people had now one exception actually was kind of where you and i were together which was health care co i don't know if you've ever said i guess we could say it's out of business 98.6 we
Brad: Right.
Scott: didn't we didn't have the technology that was they were they that was cutting edge trying to figure out how to do essentially ai for doctor visits and you know if we were doing open
98.6 today would be a different game,
right? Because of the advancements in LLMs.
Brad: For sure. Well,
but by the way, that's another good example, though,
where, you know, if the sort of through line there is the technology is one piece, but the rest of the business is a whole other piece. The technology doesn't matter without the business. And in the case of 98.6, in the case of health tech,
you know, one of the challenges that very much exists is that health related services are.
are very tightly regulated by the Food and Drug Administration.
And so, you know, I remember at 98.6, one of the challenges that we faced there was that if we built the tech and it was doing too much.
then it it goes into the realm of essentially becoming a medical device and once it becomes a medical device then all of a sudden you have to go through lots of additional like bureaucratic uh burden right and so the the tech piece of it is was hard but it was not the whole problem like the the
Scott: That's right.
Brad: there's other things
Scott: So you asked the question,
are we building an MCP server?
And that is, I literally was in a meeting earlier today talking about that.
MCP server is to the 2026 what Facebook integration was
Brad: Yep.
Scott: to 2010.
Brad: Yep. Valid.
Scott: which is like well what do you mean by that exactly and
Brad: Yeah.
Scott: we were having a discussion like well what who is the customer for this server right is are we exposing this externally to people is this for so it kind of it's a riff on the same thing which is like yes the engineering team can build you an mcp server we can do that what what kind of mc what what do you want to do with this stuff
Brad: Yeah. What actions do you want that server to have?
Scott: Yeah, and we were able to get to that answer,
but there's actually, you know, a half dozen things that could do for
Brad: Yeah.
Scott: different user segments.
Brad: Yeah,
that's interesting.
So on another AI news,
here's one for you.
So one of the teams that I work with, they finally found some MCP servers for the tooling that they use regularly that has been a big unlock for them.
So, you know, a simple example of this is like we have an MCP server for our observability platform,
right?
Well, that all of a sudden was an unlock when they were trying to fix a bug that was some obscure bug that only happened to some random customer.
Boom,
let the agent do the deep dive into the observability,
find the evidence and come back with an answer.
Scott: I just unless somebody was watching the video of this they couldn't see me just put my hand on my forehead I was just having oh my goodness why didn't I ever think of that moment how many times have you spent digging through logs trying to correlate one thing with this thing oh I think look these things happen they're only a few milliseconds apart it must must have been the same thing oh I've spent hours of my life doing that
Brad: the agents are so good at this like you can give them you and well the agents already like you could just dump a log file on them
them and ask them to do some analysis and they can come back with like mermaid diagrams showing exactly like all the events that thread together that cause the problem but now just like don't don't even bother downloading the log file just just point them at the observability mcp server and let it go find let it go do the searches that it needs let it let it because it can go look at the code to see what observability signals theoretically exist in the code and then it can go and just hit the observability back end
again and come back with like all the various you know actually the evidence for you so like that's a huge unlock um that one of the teams i'm working with just just unlocked for themselves and just like they're like you know pigs and shit basically they're so happy
Scott: makes me want to sing right now okay i i have two i have two thoughts to share here these are these are not that serious the first is do you remember a tool from back in the olden times provided for free by microsoft called
About the log parser.
Brad: I have vague memory of that,
yes.
Scott: It was a simple command line tool that allowed you to query structured text files.
So you could have some, you didn't tell what structure it was, whether it was tab delimited or comma delimited, or if it had its own structure,
you could give it a template.
But mostly they were just comma CSV files.
And so you could give it, you could just write a SQL like syntax to run queries against log files and you could join, you could do joins across different files too,
a super.
Super cool.
This is like 2000.
in 1998 to 2002 kind of era but when i remember i mean i was a pro at that thing in terms of spelunking through server logs
Brad: Yeah,
Scott: to
Brad: totally.
Scott: try to figure out what was going okay but in the same era this thing happened this guy i worked with i really liked his name was alan taylor alan if you're listening to this been a long time hello alan by the way went on he ended up leaving the company i was working out of him and his wife moved to this is pre-remote working course they moved from seattle to boston
where he went to work for the Boston Globe and the Boston Globe had this feature called the big picture which was basically just effectively one blog post a day featuring one AP one beautiful or interesting AP photograph and
Brad: Oh.
Scott: Alan did this like this was a skunk works project that he was a computer program where he did this because his observation was
We had all these AP photos coming through the Boston Globe all the time and I kind of knew about this blog technology So I kind of like just to kind of just put one up every day because they're cool You know like here's an eagle catching a fish or here's a forest fire or whatever That became a big thing of to be a big a
Brad: Yeah.
Scott: big piece of their newspaper Anyway,
I walk into Alan's office one day and he's got this book on his desk You know one of the O'Reilly books with the animals on front I don't remember what the animal was but it said Ajax on it
And I was like, what's Ajax? He explains it to me.
We all know what Ajax was. It's basically a web page being able to make an XML request and render stuff on the page.
This other guy I work with named Mark Puckett, who as far as I know is still in the Seattle area.
Brad: Mark, if you're listening,
Scott: And
Brad: howdy.
Scott: Mark, if you're listening,
howdy. It's been too long.
Mark,
consider Mark a fellow friend in the software world.
He had just...
months earlier he was working on some app and just made a thing where he was just like oh look i figured out how to make the this little thing what was it called the x and x http request
Brad: Yeah,
Scott: look
Brad: XHR.
Scott: i could go request it he called it the target here that's because he just created this web this little internal tool to this web page that could go get data and render just called it a target here and when i saw alan's book and he explained that to me i was like hey we already did that
And the lesson that taught me, and I think you're getting to it with the observability stuff that you're talking about now that we know now,
when we figure out these clever new things,
we got to brand them.
Brad: Yeah.
everyone
Scott: if mark had called the target here something sexy like ajax and went out and put a blog post up about it and say hey i've got this new
Brad: would
Scott: way
Brad: have been yeah
Scott: of doing thing he would have he would have been the ajax guy instead of whoever invented ajax but he just did it and this was i just got clever i figured out this thing let's move on with life so you got to brand your observability splunking you got to give that name but
Brad: I'm sure that we're not the first one to do it but I hear you
Scott: we're in an era both i guess the point is just like the web
The Web in 2000, where
Brad: Yeah, we're moving.
Scott: we're
Brad: Everything's moving so fast.
Scott: in an era where things are moving fast and evolving fast and people are going to be figuring out things just out of necessity and cleverness.
They're going to be figuring out the next Ajax's,
Brad: Yeah,
Scott: right?
Brad: I think you're right. Well,
here's here's sort of pull on that thread a little bit further. So the same team, another one of the developers and this is one of the developers who I'd say was like.
Like kind of early on, he was a little bit AI skeptical,
but he's a full convert now.
And he was just telling the team like he set up a signal MCP server.
And so basically when he's when his when his coding agent is running along,
if it runs into a problem,
it sends him a signal message.
Scott: Mm
Brad: And-hmm. then on his phone on Signal,
he's able to basically just respond to the agent and tell it what to do.
And then the agent gets back to work.
to work and keeps working.
Scott: So why signal and not slack?
Brad: that's what that was the api that he chose he could have been he could have done it in slack as well right so it's like he liked that particular form factor and he didn't go full open claw so he's not running like open claw like giving it credentials he just said oh look there's already this mcp server so when the when the agent has a question and when it needs to do something it can ask me a question and it does it it asks me over here and boom i keep going and yeah totally he could have used he could have used um slack he could have used
messages whatever he used signal because that was that was the tooling
Scott: Even think about MCP. Who came out with MCP? Is that a Google thing?
Brad: uh i think that was actually um probably
Scott: Anthropic?
Brad: I think it was OpenAI that first suggested
Scott: we
Brad: it.
Scott: gotta gotta look that up anthropic you
Brad: Okay,
there you go.
Sorry,
Anthropic, I gave unfair credit to your competitor.
Scott: think that anthropic spent a lot of time internally thinking about this blah blah blah or do you think it was just like some guy who was like we need a i'm just gonna hey i made an api so that this thing could talk to this other thing oh yeah what do you call it i call it i don't know it's a model idea
By giving a context, I guess it's the model context API.
Well, it's not an eight, but it's not, you know, whatever.
Brad: Right.
Scott: That's how it got invented.
Brad: So a recent episode that you that you didn't join in on,
but Philip Rosedale was on and he and I were talking about this and he made the observation that MCP, of course, was the master control program in Tron, the thing that basically was the fascist government or fascist entity.
that took over everything in Tron.
Scott: Yeah, so maybe that's how they got it. But I think the same point is those guys did something and were smart enough. The point I'm trying to make is they were smart enough to brand it.
Yeah.
Brad: Yeah.
Scott: Put it out there.
Brad: Yeah,
totally.
Scott: I think there are competing protocols for this, but somehow the world's imagination Well, got your
Brad: that's
Scott: MCP.
Brad: a classic. Yeah.
Well,
it's a classic like early to your point,
right?
Like early on, it's like, are we doing like, you know, plugins to Internet browsers used to be like the thing, right? It was Netscape had their plug in API and Internet Explorer just used.
you know calm because they already had you know eye out run known and all that stuff right and there was competition for who's going to do plugins of course plugins don't you know chrome chrome uses you know nobody really does plugins to be in pages anymore
Scott: Extensions, they're now extensions,
Brad: now
Scott: right?
Brad: they're extensions um uh another one that is interesting in this space is again one of the team that's like embraced this ai model where they all treat the ai agents just like
like team members that work for them and they they have to follow the same cicd pipeline that that developers do human developers do one of the developers was saying to me uh in this you know mid midway through the sprint he's just like yeah i want my agents working 24 hours a day i want them working on i want them and it's like let's have them read off our jira board let's have them pick cards off the jira board let's have them you know
do the work let's have them submit to pr let's have the other agents do the code review nobody's allowed to merge until a human looks at it and merges it but let's have them follow our same cicd using the same practices we do same same sdlc that we have but the tools for them to do that are available exactly
Scott: That's what you're doing. You got agents working 24 hours a day.
Brad: I don't know that he's fully gotten it to that point. I don't have agents working 24 hours a day,
but I definitely am.
I definitely have agents doing all of the pieces of the SDLC and I just kind of monitor them as a manager.
And that's what all of the developers, that's what all the developers on the team are doing.
And the more that we can get them to glue,
essentially follow the script of like,
you know, hey,
you open a PR,
you.
you make your change you um you have to lint it you have to run the unit test you have to add new unit tests you have to make sure it compiles you have to submit a pr once you submit a pr request a code review by the other code review agent what you know come back five minutes later look at the comments on that address those comments when you're all ready with that uh let me know like that like that's essentially where we're getting to
Scott: So what's your brand name for this?
Brad: Well,
I think that it's just agents is all they call it. This is how the top performing teams are working.
This is how they work.
Scott: So give a name to how
Brad: they I work. don't have a name. I'm not a marketing genius.
What's the name for this thing?
It's called agent SDLC,
agent development lifecycle.
There you go.
The ADLC.
Scott: ADLC,
that's it.
We've just invented the ADL-CBREP.
Brad: ADLC.
That's the title of this episode,
the ADLC.
Scott: You think I'm joking?
Brad: No, you're not.
I know you're not. I know you're not joking.
Our producer will make sure to title this episode the ADLC.
Scott: The ADL-C is just a layer on top of this new thing called SDD.
And what is SDD?
Brad: spec driven design spec
Scott: And
Brad: driven development
Scott: what is it? What is it really?
Brad: it's just writing a spec is all it is just a bunch of text files right saying write a spec first and then break it down in tasks and then work on the tasks so but
Scott: Yeah.
Brad: you're right you're right it's coined it's coined now as sdd and of course when someone says oh i'm following sdd principles and it's like wait what and then you gotta go look it up and you look it up and you go oh
Okay,
yeah,
that's what we were all doing,
but you just gave it a name.
Scott: Just like the targeteer.
Brad: Just like the target tier versus Ajax.
Scott: So let's ask this question.
If we want a DLC to become a thing so we can be the ones that write the O'Reilly book,
before I interrupt,
before I ask this question, I want to interrupt myself.
When we put out our O'Reilly book with an animal on the front,
Brad: What's the animal going to be?
Scott: what's the animal going to be?
Brad: I haven't all the animals been used by this point.
Scott: Let's ask AI if
Brad: Any animals
Scott: you do it right
Brad: left
Scott: here.
Brad: on O'Reilly books?
Scott: There must be animals left for O'Reilly books.
Brad: I asked Claude, what animals have not been used for an O'Reilly book cover?
Scott: Well,
according to Google's I assume Gemini, there are many animals left.
Brad: I mean, there's so many animals in the world, you would think there ought to be several.
How about raccoons? Have raccoons been used?
Wow,
they've used over 1,200.
Scott: I said I want an animal related to scheme.
Okay, here it is,
the Arctic Fox.
Oh wait,
it's already been used.
Brad: This has gone off the rails completely.
Scott: How about the mountain goat?
Brad: That's not been used yet?
Scott: I mean,
I don't think so. It's hard for me to say.
Or Penguin. I want Penguin.
Oh
Brad: Penguins have been used
Scott: no, Penguin's been used.
Brad: a hundred times.
Penguins of course have been used.
Scott: Okay, so the question is, this is a practical question and we need a marketer on it. We need to get a marketer on here.
If we want ADLC to become a thing,
what do we do?
Brad: We should have a podcast where we talk about it.
got it and we should we got to have we got to have a white paper or a blog post as well we got to basically describe it yes
Scott: No. Yes,
all those are true.
However,
we also need a GitHub repo with
Brad: yes
Scott: a readme and a command line tool that you can download to do a DLC.
Brad: you're right
Scott: because that's i don't know the way i learned about spec-driven development was to go into spec kit and like google or someone's produced this thing called spec kit and
Brad: Yep.
Scott: i was like what's what's sdd i gotta learn about this and someone said spec kit and i was like okay and that's where i learned i was like oh i see it's just a process for doing this so
Brad: Yeah, totally.
Well, there you go.
Scott: how do we yeah that's it we're going to make an adlc who
Brad: ADLC GitHub repo.
There you go.
Scott: can type it faster
Is it
Brad: We
Scott: going to be
Brad: should
Scott: a new get profile
Brad: off this podcast call and start working on that right away.
Scott: We need to sit on this podcast so no one else does it.
Brad: Well, definitely should not release this podcast that we have the GitHub repo up. That's for sure.
Scott: Yeah, I think you're right.
Brad: Yeah,
Scott: I
Brad: that'll
Scott: think
Brad: just make
Scott: you're right.
Brad: the Super Bowl joke that much funnier.
Here's what I think. Okay, so let's try to pivot this back to something.
some interesting conversation for our
Scott: I think this is legitimate.
It's probably
Brad: no
Scott: been no discussed.
Brad: hold on a second hold on a second i'm
Scott: Here's
Brad: i'm let's
Scott: what's interesting for
Brad: keep talking about
Scott: us.
Brad: adlc hold on let's keep talking about adlc for a second but i just want to like i want to pull on this thread a little bit so i here's here's my here's my current working hypothesis my working hypothesis is that there are three camps of developers using ai the first camp
is the i'm going to just label them the skeptics right so
Scott: Mm-hmm.
Brad: they're dabbling they're maybe using it for some minimal kind of like prompt you know kind of you know kind of like automatic text completion you know like a a sort of somewhat fancier version of like auto complete type stuff or and they're really like hesitant to sort of like really dive into letting the agent write any code
Okay.
Scott: Mm-hmm.
Brad: On the way other end of the spectrum is the vibe coders, right?
So they're just like full vibe coding it where they're just like throwing out random ideas and just letting the bot run and come back with it.
And they're not really looking at it from an engineering mindset perspective.
And then there's this third camp, which I'm saying,
I think is the people that are the winners.
You know, when we talked a couple episodes ago about like the K-shaped economy of using AI agents.
I think the people that are basically integrating agents into an already robust software development lifecycle.
Those are the teams that I think are probably being the most successful.
So first of all, what do you think of my hypothesis and are there other types of users of AI agents out there?
Scott: Yeah, I think those are the three that I can think of. I think there's variations of depth in each of those areas.
And I think the third area,
which is.
Really,
the through line to me between STD spectrum and development and ADLC,
which is the agent development lifecycle,
is that what we're seeing now, 12 months later,
right?
We had the explosion a year ago, 12 months later,
what we're seeing is professional software engineers turning this into practice.
Brad: Mm
Scott: Right.
Brad: -hmm.
Scott: Saying,
okay,
we've got this powerful new tool and it needs, we know that the way to build software is through.
structure and guardrails and i don't want to use the word process that's not quite the right word for what i'm trying to express but that's what the reason we have sdls is we know that having a pattern that you follow over and over again allows you to produce better software because you can iterate on the pattern and so what we're seeing is the community trying to figure out okay we got the lifesaver is like wild you know it's like the hose where it's gone too fast and it says waters
are springing over we're trying to get it under control through these professional software development practices so we can aim it exactly where we want it to go and so it produces what we want so we're kind of trying to take a very wild non-deterministic type of thing and turn it into something that is predictable I said wild and non-deterministic I should say wild and unpredictable
predictable and turn it into something predictable because for our profession that's how we do a good job is that predictable patterns that we follow over and over again that we can iterate on and improve and refine we were talking earlier we were goofing around earlier in this podcast i was talking about log parser right now i used to go splunk around in logs well now you've got tools like data dog or whatever your whole systems for this has been refined and refined and refined over the
for the years and so we're entering another phase of that of this adoption cycle we're saying okay how do we take this thing and turn it into a professional tool and ultimately i think you're that you talk about the k-shape economy and the winners yeah the folks in that last bucket
are going to crush the vibe coders by 100 to 1 and they're going to crush the skeptics by 100 to 1 because they're not only going to be producing they're going to be producing a vibe coding volumes with professional software quality and be really effective but
Brad: Yeah.
Scott: i'd say the last thing i'd say i even recognize that i'm saying this my voice has taken a more serious turn it changes the job
Brad: It does.
Absolutely.
Scott: It changes the job.
It is not a coding job anymore.
Brad: You know, 100% agree.
And actually,
I have two other thoughts about this that we haven't yet touched on that I think are interesting.
And maybe this dovetails nicely into your it changes the job.
Two of the organizations that I'm working with,
the developers are now moving so fast that the product owners actually are having a hard time keeping up with coming up with ideas.
Scott: I believe that.
Brad: And so I think that's really interesting.
months ago you and I talked about the implications of that and the fact that that might mean you know you need fewer developers or and you I think you pose the question what or maybe you need more
product managers or you need more ideas so what i think is interesting is that's one of the ways that i think it changes the job and i'm seeing this in i'm seeing as a matter of fact i'm seeing this in in the organizations i'm working with where effectively the developers are now coming to the table with product ideas because now they if they are already sort of if they're already sort of like entrenched in the product understand the product really well they're coming up with ideas
and they can basically bring those ideas to fruition incredibly fast now right and offer them essentially to product owners to say what if we did this and i already have it working right like Yeah. in the last basically in the last two weeks i have seen very small engineering teams take on tasks that the product owners thought would take them weeks to months to do and the developers have the high quality developers strong developers using the agent development
development lifecycle philosophy,
have cranked out these month-long projects in a matter of days.
Scott: Yeah. So, you know, it's interesting here because.
To some degree, I have two thoughts in response.
I like where you're going with this.
To some degree,
coding was only a small portion of the job.
Brad: Yes.
Scott: It was a lot. At the lower levels,
when you're beginning your career, it's coding, right? You're getting a ticket, you're coding it. But as you progress,
it became a lot more about understanding the requirements.
Well, if you want to push it all the way, understanding the customer needs,
understanding the requirements in the design.
right of how to create a scalable system and then coding was the last piece if you did those first pieces well the coding often was just sort of downhill sledding and
Brad: Yep.
Scott: it's just making the coding piece away but the other thought that comes to mind is agile software development and scrum and what you're really saying is i don't know what you're really saying about what you're inspiring me to think is gosh maybe
ADLC is really bringing the promise.
of agile software development the promise of scrum to true fruition because really you can say to an engineer hey here's an idea we have and they can say great i need a two-week sprint to go put together a proof of concept let's come back and look at how it is and then for your at least for your online services like i've always been involved with then it is what does your experimentation framework look like so that you can get something up
you know in front of customers and get real input on it sooner and then that in turn feeds in a little bit to your culture is your culture willing to do stuff like that
To
Brad: Yeah.
Scott: say, I don't know this this engineer and this product manager came up with this idea and now we're going to go put it in front of our customers to see what happens.
Pretty fascinating. This is actually pretty fascinating in terms of what it means for this business.
Brad: Yeah.
That was my first point. And then the other thread you have to pull on is, again,
the teams that I see that are moving super fast,
they don't have QA teams.
And they and
Scott: Forget about it.
Brad: their own and the developers own the QA.
Right.
So it's all about like the accountability,
like there is so much more accountability on the lines of code that you make in that pull request are like as the engineer that worked with these AI agents as part of the agent development lifecycle like.
agents need to be doing as much testing as possible and you need to be able to validate that the testing that the agents are doing really does manifest correctly and so that's the other unlock that at least one of the teams i've seen has has solved with um mcp servers is there's these mcp servers for doing um basically like running a headless chrome or running a headed chrome which like will fire up your web server log in do all kinds of things and like literally i was just testing this out the other day
Today, I just asked the MTP server to just like log into the application. That's all I told it. I told it email and a password.
I told it to log in and it doesn't need any more instructions than that.
Right.
And it knows how to go do it.
Scott: Yeah.
The last time I had QA was 2009.
Brad: Yeah.
Scott: The company I'm at now,
TaskRabbit, we have some QA engineers because of some legacy reasons and code,
but I've found for the longest time having your engineers do QA or peer QA across each other is a hundred times more effective.
Brad: Well,
and that's the thing, right?
Like in this world where the engineering can move as fast, so fast,
it can move so fast, you don't have time to.
play the game of telephone to talk to some department somewhere like like no you need to be doing it like live in the seat
Scott: Yeah, I agree with that for sure.
And there's something to be said for.
It forces some other really healthy behaviors,
right?
You get your healthy test automation behaviors.
You also get your healthy observability and monitoring behaviors and you get your healthy rollback behaviors.
Brad: Yep. And if you've got like, and again, it sort of like encourages you to do things like have a really good, you know,
development environments that you can run locally,
right?
Like have a solution for something like, you know, ephemeral environments or something like that.
You know, let the, you know, if you could tell the agent.
make a change to know what all of its automated tests that it should run and pass that it needs to update those and also if you can tell it hey look if you once you have your pr up and running you will get access to an ephemeral environment go run like end-to-end tests against it using this like using your own description of the the test right you just added a new feature to have a you know
mic mute button okay go test that like you just added the feature go test it in in the full end-to-end solution see that it worked correctly
Scott: Or you use the ADLC.
Let's talk about how you would test that.
Brad: yeah
Scott: Let's make a plan for testing it.
yeah and then you know and then there's some other parts of your deployment story might change in terms of to again to test you have canary deploys blue green all that kind of stuff right those become more valuable 100
Brad: Yeah.
And so, again,
like I think what you're basically,
I think that we're in violent agreement on this, which
Scott: i'm
Brad: is, yeah.
Scott: not going to disagree with you
Brad: So we'll summarize it for the listeners.
If you've already got.
Excellent engineering best practices around a software development lifecycle.
Have your agents start using that software development lifecycle.
That's
Scott: Mm
Brad: the
Scott: -hmm.
Brad: agent software,
the agent development lifecycle,
ADLC.
And that means you've got clean processes for how you evaluate the whole,
you know, the product ask to the implementation plan,
to the code itself,
to the validation of the code,
to the verification of the code through some audit cycles like whatever you're if you've got SOC compliance,
like what are your what are your rules around validating that the code matches your, you know, your software security.
requirements.
Like all of that stuff, hold the agents just as accountable as you would the human engineers.
Make
Scott: Yeah.
Brad: them follow that same cycle.
Scott: You know, it's interesting. I had a conversation today with our head of design who wants her designers to be able to make, you know, front end type changes.
And she was kind of trying to figure out the process, like would it have to get code reviewed by a lot? I said, look.
I'm totally supportive of this. Done it with designers before.
I think it's great.
But you're going to operate just like the rest of our engineers.
We'll get your designers set up with a logo development environment.
They'll follow our process. They can go to the engineering troubleshooting channel if they got questions,
just like everyone else does, and be happy to have you submit code pull requests.
It's the same with the agents.
Brad: Yeah.
Yeah,
Scott: Follow
Brad: I,
Scott: the
Brad: yeah,
Scott: process.
Brad: and, you know, one of my more leaning in.
product owner visionary founder type people he's basically like boom asking agents to go build stuff using the you know kind of the github we use github for they use github that that team uses github and and it's like uh yeah submit a ticket ask ask co-pilot to start working on it you know github's added these chat mechanisms but he's just like yeah i will i all the tooling exists so it can't go wrong you know and if it
if and if it doesn't pass lint then he says you don't pass lint you got to fix that you know and if it doesn't pass the unit test hey it's not passing the test you got to go fix that you know and it's just like yeah he's got the guardrails he's got the guardrails of all of our existing uh you know cicd and sdlc and so he just lets an agent go work in that playground and he's confident that you know he'll get what he wants at the end so there
Scott: Well,
Brad: you go
Scott: Brian.
The ADLC.
Brad: The ADLC.
Scott: I love it.
Brad: Yeah.
Scott: This has been a fun Wednesday.
Brad: Yeah.
Scott: We covered a lot of ground.
Brad: We did.
Scott: We started with nothing and we ended up with the future of software engineering,
the ADLC.
Brad: The ADLC.
All right. Good talking to you, Scott.
Scott: Good talking to you, brother.
Brad: This episode of How Many CTOs Does It Take podcast was edited by Safran Epigal.
Our theme song was AI generated by Brad.
Please follow us on all the social media platforms at at how many CTOs pod.
You can find us on YouTube,
LinkedIn,
Instagram,
Facebook,
Threads, X,
TikTok,
and Patreon, or at our website at howmanyCTOsPod.com.
And remember to rate,
review,
and subscribe on the podcast platform of your choice.
Thanks for listening.