Aug 18, 2026 · 59 min · 11 segments
This week's conversation is one we believe every MSP needs to hear. AI-generated vulnerability patches have become the obvious next move for software developers drowning in a tsunami of CVEs. It…
Keith HoodletGuestAndrew MorganHostLironHost
Phyllis LeeHostJonathanHostSure.
So you would type stuff and it would tell you, I think this is what you want.
And it said, oh, it's magical.
Then we got into vibe coding where you could just type stuff and talk to it and it would do stuff for you.
And now if you listen to Boris Cherny and I forget the creator of OpenClaw, they talk about this concept of loop engineering because the models have gotten to a point where the how is a lot less important than the what.
And so you go in there and you say, this is the end state that I'm looking for, but it's an engineering end state, right? So you have to provide metrics and be as specific as you want.
It's basically acceptance criteria.
And you say, go and do this and don't come back to me until this thing is done.
Now, that is like the future right or this is what the best people in the world that are doing this are doing right now so what i did was i took a youtube transcript of boris tourney talking about this fed it to claude and said hey this sounds like a really good idea i want to start doing this turn it into a skill for me so then i turn it into a skill and then when i go and i call my loop skill it asks me a bunch of questions it defines what the end state needs to to look like And then it just turns until it achieves the end state for you.
So that's loop engineering
level.
Keith, does that resonate for you? You know, when you think about, you know, coding and how, and I don't know, you know, again, we're going to, I don't want to give away today's session, but when you thought about, you know, developing a research for this.
And so as part of the research, we do include what we call like an iterative approach.
And one of the challenges, and that's teasing a little bit of the research that's forthcoming, is when we looked at this research and the results came out the way that we discovered them, we realized like, oh, no.
This is not a great place to lead the world right now, given how many vulnerabilities are being reported to both open source projects and private industry alike.
So one of the following pieces of research is going to be around how many times through the loop does it take to actually produce what we call a scenario one successful patch outcome.
There may be situations in which that loop could be potentially infinite, as we cover in the case study section of the research itself.
where you never end up in a scenario where you actually get a successful vulnerability patched um but it's you know yet to be seen as to exactly what that loop has to look like what you know percentage of correct instructions uh are required versus maybe just having no instructions uh being better in some instances so um yeah overall we are looking very closely at this as like a what's next because we can't just drop this research and be like, sorry, we're done.
Sure.
So you would type stuff and it would tell you, I think this is what you want.
And it said, oh, it's magical.
Then we got into vibe coding where you could just type stuff and talk to it and it would do stuff for you.
And now if you listen to Boris Cherny and I forget the creator of OpenClaw, they talk about this concept of loop engineering because the models have gotten to a point where the how is a lot less important than the what.
And so you go in there and you say, this is the end state that I'm looking for, but it's an engineering end state, right? So you have to provide metrics and be as specific as you want.
It's basically acceptance criteria.
And you say, go and do this and don't come back to me until this thing is done.
Now, that is like the future right or this is what the best people in the world that are doing this are doing right now so what i did was i took a youtube transcript of boris tourney talking about this fed it to claude and said hey this sounds like a really good idea i want to start doing this turn it into a skill for me so then i turn it into a skill and then when i go and i call my loop skill it asks me a bunch of questions it defines what the end state needs to to look like And then it just turns until it achieves the end state for you.
So that's loop engineering
level.
Keith, does that resonate for you? You know, when you think about, you know, coding and how, and I don't know, you know, again, we're going to, I don't want to give away today's session, but when you thought about, you know, developing a research for this.
And so as part of the research, we do include what we call like an iterative approach.
And one of the challenges, and that's teasing a little bit of the research that's forthcoming, is when we looked at this research and the results came out the way that we discovered them, we realized like, oh, no.
This is not a great place to lead the world right now, given how many vulnerabilities are being reported to both open source projects and private industry alike.
So one of the following pieces of research is going to be around how many times through the loop does it take to actually produce what we call a scenario one successful patch outcome.
There may be situations in which that loop could be potentially infinite, as we cover in the case study section of the research itself.
where you never end up in a scenario where you actually get a successful vulnerability patched um but it's you know yet to be seen as to exactly what that loop has to look like what you know percentage of correct instructions uh are required versus maybe just having no instructions uh being better in some instances so um yeah overall we are looking very closely at this as like a what's next because we can't just drop this research and be like, sorry, we're done.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.