Episode 10 · Transcript
Good Enough for Agentic Work
Full transcript of the recorded conversation.
Back to the episode · timestamps jump to the video · lightly machine-transcribed, may contain errors
0:00 Paul
agents can pump out so much content at the end of the day, you're drowning in what we call open loops. if you got a bunch of these guys doing stuff, you're the glue that makes sure they don't get in each other's way. You're the guy that has to authorize
permissions You're the guy that's got to verify it at the end. And if you've got five or ten of these things running, it's kind of like awful. I've got a day's work just to validate some of the stuff that my agents have done during the day. So it's not like easy money, I wish it was, right?
0:37 Sam Rogers
Welcome to Signals and Subtractions. I'm your host, Sam Rogers. How do you know that an AI adoption effort changed anything? Organizations often measure visible activity, right? Policies, committees, training, tool availability, but
The decision requires evidence of changed behaviors or outcomes. My guest today is Paul Gibbons. He's got 30 years advising boards and C-suites on the human side of technology. At PWC, IBM, Google, Microsoft, Deloitte. He's written over 10 books. The Science of Organizational Change was one of the first to bring behavioral science.
To leadership and change. And change myths went after the junk science that the change management profession is still teaching. Two things make Paul really stand out as the right guest for this show. One, he builds production AI systems himself. So he's not just describing this hypothetically from the outside.
Two, as a chairman of KPMG once said to him, he never told us what we wanted to hear. He told us what we needed to hear. And Paul, that's what we love to hear. So welcome. Paul, thanks so much for
being here. Really glad to have you.
1:54 Paul
Good to be here.
Good to be here. And we have so much in common. It's surprising to me given that we've swum in the learning and development and org change waters for twenty years and not bumped into each other until recently.
2:07 Sam Rogers
I knew who you were. You maybe just didn't know who I was. I was I was the little pipsqueak trailing behind the writer of epic tomes
2:15 Paul
Yeah, Thank you.
I've written two books since adopting AI. Adopting AI was in the in the Cretaceous period. It was in April of twenty twenty five it came out.
Like there were velociraptors running around. I wrote a book in November called Brains, Bodies, Mind, which was how the way AI was affecting
medicine, therapy, wellness and all that. and then I wrote a book on the link between poker and business strategy called Polymath Poker which came out in June.
2:42 Sam Rogers
and the science of organizational change?
2:44 Paul
That
that's the one that broke the most new ground
because I was reading a lot of stuff at the time that had been irritating me for a long time, which was the fact that it's not very easy to measure the results of what change management
people do.
and the people that publish research saying change management makes you 60% more likely to succeed. Let me let me say this is an important parentheses because there's a very famous firm in the change management space. I'm gonna save myself a lawyer's bill by not mentioning them by name, but they publish research like this. But the way it's conducted is they ask people like
How's the change management and how successful is the change? And they correlate them. And it's really a surprising to no one that they're successful. Have you ever been in a project where they said, well, the project was really successful, but the change management sucked? Or or the
3:28 Sam Rogers
Right.
3:29 Paul
change management was excellent, but the project failed. Like that doesn't happen. You're sort of asking the same question in a different way, and you find a very high correlation. Well, yeah, wow, surprise, surprise. And in my
3:39 Sam Rogers
Yeah, yeah.
3:40 Paul
own cur consulting career, I did a project for KPMG in the
Two thousands, which was one of the most successful projects I've ever run. And it was kind of a culture change project. And the financial turnaround of the group that I was working with was enormous. And I would like to say, hey, culture change and you know, financial turnaround, and they were the lowest revenue per partner and they went to the highest revenue per partner, and which is true.
4:02 Sam Rogers
Yeah.
4:02 Paul
Something like that. But the CEO wasn't sitting in a hammock while Paul Gibbons was running his culture change program.
He was reorganizing his strategy, it was rehirings and firings and he was doing a lot of stuff 'cause he's a new CEO. And between all the things that we were doing, it was very, very successful. He would say it was instrumental. But he doesn't know any better than anybody else does. we can't draw a causal relationship between what I did and this huge financial turnaround. I'd I'd love to, but you know, just like good.
4:28 Paul
Just the science doesn't stand up.
4:30 Sam Rogers
things, is definitely a team sport. but it's relevant to what I'd like to talk with you a bit more about today, which is specifically
4:37 Paul
Signals.
4:38 Sam Rogers
AI adoption efforts and how we know
4:40 Paul
Yeah.
4:41 Sam Rogers
that that anything did change. I first became aware of your work actually as I was getting the ProSci certification
there were some things that just didn't sit so well with me. And I was looking for voices that had other offerings, other things to say. And that's when I came across yours. and enjoyed your evidence based work. And you know, it seems like such a simple thing let's have some evidence to say something before we say something. but it's surprisingly rare.
5:09 Paul
Yeah, no, in the science of organization I say that no client has ever asked me if I had evidence for what I was saying. Like it's never been asked. And it's funny, I work a lot with Claude now, and Claude's always like, you know, the evidentiary nature of that, the epistemic nature of the you know the I'm like, Claude, nobody gives a right? Nobody asks, nobody cares about how much evidence people have for the things they're doing. I once asked a guy who was an expert in criminology and prison reform.
I said, how much of what would be considered evidence-based practice applies in the world of criminal justice system? He said, Whatever the evidence suggests we do, we do the opposite. a lot of what I based my science of organizational change book on was evidence-based medicine. And you'd think like medicine is pretty sciencey, right? I mean it's like really sciencey, like they died or they didn't die.
They they they cured the disease or what? The cu tumor came back or it didn't didn't come back. Right.
5:55 Sam Rogers
Yeah. It's validatable, yes.
6:00 Paul
And you would have thought that evidence was had been sown through medical science from the get-go, but it wasn't true in the nineteenth century when they added leeches.
I mean medicine became more scientific, but the first paper on evidence based medicine didn't come out until nineteen ninety eight.
And people didn't welcome it with open arms. It's not as if a doctor who went and spent three hundred thousand dollars on his medical education said, my god, thank you. I've been waiting
6:28 Sam Rogers
Ha ha ha.
6:29 Paul
and I've been getting it all wrong and or these great medical institutions welcomed this kind of new paradigm. Now, it's been two and a half decades and evidence based medicine is considered
the highest standard. But I mean I asked a surgeon he ran the surgery section of the hospital. and he said no we don't really adopt that here as well because we think it's against innovation. It's prevents innovation. But I thought that was an interesting take. But this was in the 2010s. So the world
changes slowly and
7:04 Sam Rogers
Yeah.
7:05 Paul
people, human beings, are perfectly capable of ignoring evidence if it's in their interest to do so.
7:13 Sam Rogers
Yeah, I I'd love to say that's an interesting assertion that evidence base restricts the innovation. do you have anything to support that?
7:21 Paul
So I mean, I didn't really want to give him a hard time about that. I
7:24 Sam Rogers
Yeah, yeah, yeah, yeah.
7:25 Paul
I mean there are costs to having high evidentiary standards. But anyway, nobody ever said that about change management. So
7:30 Sam Rogers
Yeah.
7:30 Paul
that's for sure.
7:31 Sam Rogers
Well, well, bringing it back to that and AI adoption kind of stuff there's a lot of people, of course, trying new things, but not really having a solid foundation to know if those things are working. I'd love to hear from you a little bit about any signals that you're watching.
Or things that you're paying attention to that maybe have surprised you more recently for how people are thinking about things, how they're framing it.
7:54 Paul
I gave a talk in Denver
And the guy that was on after me,
He was a plumber and he left school when he was fifteen years old. He dropped out of an American high school. I I'd never
8:01 Sam Rogers
Okay.
8:02 Paul
met a guy, let alone shared a platform with a guy who dropped out of high school when he was fifteen. So he had then joined the military, then he got out of the military and
learned plumbing as a trade. And he stood up a new plumbing business, and the entire architecture of the business runs on AI. And what this guy, a mad props to this guy, has automated and has AI doing the whole bottom end of his business. you think, okay, whatever, it's a plumbing business.
When he talked about it, I thought, Well, I'm not sure I could do that because if you have forty or fifty plumbers working on I don't know how many job sites, getting the equipment to the right place, getting the plumber to the right place, putting the invoice in the right time, making the change of payment the right time, all of the managing cash flow business so you don't run out of dough pay people, like all of that kind of stuff is a great big hairy mess.
I thought, Wow, okay, yeah, I'm glad that's not my job. And he's automated that, getting the right part to the right guy at the right time
And he's integrated with the rest of it. Now now, that's an impressive story by itself. Like if big I don't know, if there are big plumbing like you know, company had one or two thousand or five thousand plumbers on its books. I don't think no such things exist, or something like that. If they had done it, gone and they'd hired and paid, someone five million dollars to stand up the system. I would have said, Wow, that's pretty cool. but this guy did it
himself.
this is not just inspiring for this guy's plumbing business, but the fact a guy with no education.
can do that, I think is extraordinary, not for just for the United States and not just for the well off, but if this is the sort of thing that can lift up the bottom of the pyramid, that to me
9:34 Sam Rogers
Mm.
9:35 Paul
Is like a huge wow. I mean, that's just like you mean some guy in Uganda who's willing to spend his weekends in Claude Code can build something that creates wealth for his family and community and build something of value. I think, whoa, that is cool.
if we can do that for the people who are unwaged or underwage or what they call in economics the precariet.
If we can begin to lift those people then we have a technology that would be remarkable and transformative for the human race.
10:02 Sam Rogers
Yeah, that's a that's a fantastic signal. So in in this case, like you're you're giving the comparable example of like if there was such a business and they were hiring for someone to pull this off, they wouldn't have hired that guy, right? Like he was
10:15 Paul
No.
10:15 Sam Rogers
able to build it from the bottom up, as opposed to taking
devices and talent that works from the top down and distributing down from on high to pull up the masses, being able
10:27 Paul
Yeah.
10:28 Sam Rogers
to build the bootstraps where someone who's motivated enough can just make it happen and they don't know not to make it happen, right?
10:36 Paul
I think that's a signal that was inspirational to me.
Well, the the evidence based world is an interesting juncture to the world of AI, innovation, and adoption. I just wrote a proposal to a big client, and then there's a Spanish poet who said, Caminante no hay caminos se caminos se hacen en andar, which means traveler, there are no paths. Paths are made by walking. And so we're in a world where
For most knowledge workers, the agentic revolution started in February. it was cowork, open claw, Hermes, Codex, those were the things that really allowed AI to do useful work because
In twenty twenty five it could tell you what to do, but it couldn't do anything. And
11:17 Sam Rogers
Yeah.
11:18 Paul
who needs advice, right? I run a small business. I mean, marketing advice is good, but I need execution. I don't need you to tell me, you know, what my SEO strategy would be and write me the 100 page report, which consulting firms charge half a million dollars for. I
11:32 Sam Rogers
Yeah, yeah, yeah.
11:33 Paul
need somebody to do the damn thing.
11:35 Sam Rogers
Yeah. And to do it right. Yeah.
11:35 Paul
And and so finally AI became kind of useful for
the man in the street. only only really this year.
everybody is kinda like figuring out like what ought to be the rules of the game, what are good practices, what are good behaviors, what are good habits, what's are good ways of governance, verification, security. Like we're learning because like none of us were born knowing this right? we're learning as we go.
Who has written a book on how to run a multi-agent OS? Hermes is up doing things and Kimi's up doing things for me right now. It sounds great, but actually it's very hard because I'm the glue that keeps all these agents together. Like when it finishes something, I have to do half an hour or an hour or two hours work to verify, approve, revise it.
And if there's a lot of that happening, I'm drowning. And that happens to a lot of people today, is that
12:25 Sam Rogers
Yeah, yeah, it's moved the bottleneck to the Yep.
12:29 Paul
the a the agents can pump out so much content at the end of the day, you're drowning in what we call open loops. and so if you got a bunch of these guys doing stuff, you're the glue that makes sure they don't get in each other's way. You're the guy that has to authorize and
established permissions and what they're able to do and what they're not able to do. You're the guy that's got to verify it at the end. And if you've got five or ten of these things running, it's kind of like awful. I've got a day's work just to kind of validate some of the stuff that my agents have done during the day. So it's not like easy money, right? I wish it was, right?
13:03 Sam Rogers
Yeah.
13:04 Paul
And my frustrations are like, tear my friggin' hair out here. But everybody's in the same boat. Like everybody's learning how to run. Well, I say everybody. Everybody who's interested in doing so is trying to figure out how to make these multi-agent creations work. and by the way, and not use Fable five point one every time they're doing it and have ten thousand dollars a month in token costs.
how do we use more efficient models when we're having it do something similar? That's a game. Nobody's won that game yet. Some people would say they have.
13:34 Sam Rogers
Well, yeah, it's it's true
that in the in the market nobody really has. And folks like you and me tore our hair out long, long ago. That's a distant distant memory.
13:41 Paul
Well, long ago, long ago.
you treat this as a hypothesis. See what you think about this. Like first of all, I think models are way more than smart enough now for most of what we're doing. Unless you're doing advanced science, advanced mathematics,
for like any knowledge work in business, they've been more than good enough for a long time. Are they getting a little bit better? But it's at the margin, right?
Like the what matters is the the rest, like the harness, right?
14:02 Sam Rogers
Yeah. Yeah.
14:03 Paul
Now our guys at the labs are not dummies. They're aware of that. But this agentic world we say is only six months old or something like that. What's it gonna look like at the end of twenty twenty seven, fifteen months from now? Like there is gonna
14:14 Sam Rogers
Illegible.
14:15 Paul
be someone and already, the currency, the value of a model, because I have a Hermes running, right? And Hermes has a built in model switcher.
And I can tell it which models I want to use. If it wants to use Qwen or Kimi or something that costs one tenth the price or one twentieth the price with some of these things, it can use that. So so how are
14:30 Sam Rogers
Yeah, or just run it locally and it's electrical costs.
14:33 Paul
how are they going to make money out of five Fable Five One? Because Fable Five One is so far beyond most people's use cases in terms of its capabilities, you
14:40 Sam Rogers
Yeah. Yeah.
14:42 Paul
could run it on Kimi for a tenth of the price. So this is the business is like there's I think
There's no money in models. I know that's yeah.
14:49 Sam Rogers
Yeah, I I think
that revelation is starting to hit.
I'm starting to see it.
In discussions that are happening, that like we've got the one that's good enough for us. We're switching to maybe an open source one so that we're not sharing all of our information I guess with 5.1 they've changed it now, or you can manage your data differently. But so far, like with Fable, there's a different data agreement, and many institutions have not been able to do that, you know, for legal reasons.
And and as far as the the subtractions part of this, what I was gonna share is that some of us have been screaming about this for a long time, but that the model itself is now less important as a decision,
build the stack that is good enough for the work you're doing by how it is you divide up the work and more importantly the guardrails that you place around that work, the harness level, that constrains that behavior and makes it trustworthy. That's not the model itself, that's what goes around the model. No different than with people, which is something that we talked about on this show last week.
Like managing agents is much more like managing people than anybody is comfortable with.
15:58 Paul
I've had that same thought. there was a management model that came out and was the first management model I ever read long before I even thought about getting into business. It was called the One Minute Manager. was
16:09 Sam Rogers
yeah yeah yeah.
16:09 Paul
a four-box model, and I and if I remember it correctly, but on one dimension there's motivation, and the other there's competence. And if someone is extremely motivated and extremely competent, you say,
Okay, go do this, let me know when you're done. Right? If someone's not very motivated, but they're very competent when there's certain things you can do. And it's a little bit like that with agentic AI too. it's like it's the same as delegating to a person.
16:32 Sam Rogers
Yes.
16:33 Paul
If you have ten years of shared context working with someone, you could say, Yeah, go do me the deck and give it to me on Friday, right? And they know exactly what you mean and
16:40 Sam Rogers
Yes.
16:41 Paul
exactly what good looks like in a deck because they have all that shared context, right? But if you open up a model like Claude Cowork and you say, make me a deck right now, you know, it'll be rubbish, right? but know nothing about your business and your context, your clients, your clients' needs, your financial cash flows, like the way you like,
16:54 Sam Rogers
Yeah.
16:54 Paul
yeah, like your brand voice,
so it is a lot like managing people in that respect and too. And then once you know it and you have this relationship like really understands all your projects and anything like that, you can get away with and as I do, the sort of lazy man's prompt, like just go away and fix that and tell me when you're done. Right. But if you do that at the outset, you have a big trouble.
17:14 Sam Rogers
Yeah, the
differentiation I always make is the expression of intent should be very intentful. and that makes all the difference in aligning what that motivation is.
I've been writing about aggregated intelligence for years, which isn't
17:29 Sam Rogers
Everybody tells me that it's a typo, right? but really that's what I'm building towards. That's what so many of us are thinking is not people first versus AI first, but really, what is the the maximum amount of intelligence that we can leverage for solving problems? And being able to hone our intent such that people and machines can pull in the same direction and we know how to tell when that's not happening. that's what I'm working on. That's what I've been excited about for a long time. But enough about me.
for the folks listening to bring it back a little bit to AI adoption.
17:57 Paul
Yeah.
17:58 Sam Rogers
and how it is that you know that AI adoption is actually going well.
do you have anything that you would lean in on with someone that you're speaking to, someone that you're advising, around how they would measure the change that they're seeking to create within their organization, specifically around AI?
18:17 Paul
this is undiscovered country, right? So the proposal that went out the door two days ago was some organizations could pay me money to actually work out how they know AI is being used well. Like what are the behaviors? Yeah, variegation, you know, we're well past prompt engineering, but you know, that bit that's after prompt engineering from twenty twenty six, loop engineering, honest engineering, graph engineering, one of the engineering's
You know, like what the behaviors that constitute that? And how do you verify the work product that's generated by AI? Because I honestly like if I have Claude code running when I'm upstairs and there's six PRs waiting for me on
18:49 Sam Rogers
Pull requests, yeah.
18:50 Paul
GitHub, I don't do this anymore. But when I first started, I would just click yeah, yeah, yeah, merge, merge, merge, merge, merge, merge, merge. And and then I I would often run it without permissions because I have attention issue.
19:02 Sam Rogers
And how'd that go?
19:03 Paul
Well, I you know, nothing stupid happened, but I'm not exactly in nuclear weapons here, right? I'm in management consulting, so
19:08 Sam Rogers
Yeah. Yeah, yeah.
19:09 Paul
I've actually had to discipline myself to read stuff like a pull request if you go into Github It's got a bunch of code in it, There'll be language you don't understand and I have to
Force myself to read through it and make sure it makes some sense. I don't claim to be an engineer, but I'm using these tools so much, I just can't give everything the pass. But it's a tendency with human beings. It's like, okay, okay, that looks pretty good. I used to have a friend that said good enough for government work.
it's one of these expressions people say. actually like if you think about it hard, it's actually kind of disgusting. It means, this isn't important enough for me to devote time verifying and seeing and like improving the quality of it. there's so much wrong with that.
That sort of thinking is terrible for for the age of AI. if I write something myself, I know there won't be any ridiculous ups in it because I wrote wrote it myself. I don't know about AI.
my writing process is I write it in mark down. I pass it to Claude. I say, okay, turn that into a a draft. the prompt for something like this might be five or ten pages. Might be a page, right? Whatever, but it's long. Write that into a draft. And then this proposal is a good example. And what came back? I spent the better part of a day rewriting what Claude wrote.
Which was based on my writing. And that was really annoying. so that verification end of it is first of all it's alien to me because it's not the part of the process I like best, but I find myself doing a ton of it. the prose is bloated.
20:33 Sam Rogers
Yeah, so just to give a little insight into how I've been working towards this, the
20:36 Paul
How do you do it? How do you do it? Break it down.
20:39 Sam Rogers
the execution is the last part. So typically when I'm working on something, I will have one AI that I figure out what the spec is going to be and have it craft the test to know if it passed or not, like the test-driven development kind of thing, and hand it.
To another AI, not as a command, but as an invitation, and in giving it to a panel of AIs who can basically say, I can do this best, and here's why. So they compete
21:06 Paul
wow.
21:07 Sam Rogers
for it. And then that one can run the work, does all of the work, fan out all the agents, whatever, within that.
21:14 Paul
The lottery winner.
21:16 Sam Rogers
Yeah, they can, of course, run the validation themselves, but they can't validate it themselves.
gets
validated by the original spec writer. And meanwhile there's a third AI in the mix that is logging everything that happens. And these are all from different model families.
21:31 Paul
So what's the so how are these partners
together? Are you using cloud manage agents or what are you using?
21:35 Sam Rogers
No, no, I'm
I'm using all the AIs all the time. I built my own harness. It's a whole collaboration thing. But
21:41 Sam Rogers
Yeah, it's open source, it's free, Harnessie is the name of it. yes, it it does all of this with local models as well as with the frontier models. Because a lot of times you don't need a lot of intelligence, say for the logging part. Like I can use Gemma 4 or you know, something like that that's
21:56 Paul
Yeah, yeah.
21:56 Sam Rogers
local. And I'm I'm constantly giving as much as possible to the local computer to do the triage and and ingestion part, then refine it to where it's
22:12 Sam Rogers
It's actually not that
deep. So I built the test for it in a day. And then I've iterated it since. It is free and open source. I I'm not trying to, you know, sell you on it. I just wanna
22:23 Paul
No, the last thing I wanna do is add to my tool stack. But no, this is really this is really good. So spectrum development test and development, you have designed the test so you have a kind of lottery kind of like you can bid for this work, who do you think would be better to use it and you do it and then the third agent validate it. I mean when you say it like that, it's not simpler than it is. No.
22:38 Sam Rogers
Yeah, and and so
it's like taking the mixture of experts model kind of thing to the extreme of using different model families and being able to tether them together into a harness where they can't bluff their way out of it. Because you give anything to an agent and I can always say, Yeah, let's not do that and say we did, right? But having some evidence layer.
22:55 Paul
Yeah, they're pretty good at that. And do did you try like anti gravity
or cloud manage agents or sort of these multi agent setups and like 'cause I mean, this is frontier stuff. if you were working in if you were working
23:02 Sam Rogers
Well it
23:04 Paul
in anthropic right now, this is the sort of thing you'd be working on probably.
23:07 Sam Rogers
so here's the thing: Anthropic will never create this. OpenAI will never create this. The reason
23:10 Paul
Why? Why?
23:12 Sam Rogers
is it is integrative and it is collaborative. It is not command and control and it doesn't stay within anybody's product stack. It's an open source approach that welcomes open source and has a place for the frontier. And because it's across the market, they are not gonna shoot themselves in the wallet and ever do anything like this. I'd love to be wrong about that.
23:33 Paul
Well, somebody will.
Someone will make a commercial like with OpenClaw. Somebody will make a commercial version of it.
23:39 Sam Rogers
Well see, OpenClaw wasn't exactly commercial either, right?
23:47 Sam Rogers
this is something where like you and I, I think, coming to things from an organizational mindset, from a change mindset, have a real edge over everybody coming to it from a coding mindset and a developer mindset and
24:02 Paul
Why is that?
24:02 Sam Rogers
those kinds of habits. So I've worked a lot in regulated industry and everything. So I've got a like really strict practice around validation because of my experience doing that. And also working so much with facilitating groups and
training people and all of that. it's just a different mental model for how to work together to create change. and somebody
24:21 Paul
Really?
24:22 Sam Rogers
like you, I would say, don't try to do what the developers do. Do what you do and do it better than they would ever, ever think to do it. And that's a little of what I've done. So I'm kind of coming out of left field. I don't know that I'm a super builder, but I've I've certainly put a lot of
hours in in the last year with my startup.
24:38 Paul
I talked to a lot of builders. You're you're yeah. Yeah, I don't have too many of that. The thing is I would like to believe and there was certainly a time when there was nobody in the change world that could out AI me. Now some of them may have caught up. You may have you may have snuck past me there in your AI building thing. But there I felt like I was the only change guy that was doing anything as remotely sophisticated. I've done some clever stuff, by the way
And then for a technologist, obviously nobody on the engineering side is gonna out human capital me, like not a chance, right? So
25:12 Sam Rogers
Right. Right.
25:14 Paul
now that spot in the middle that dare I say we both occupy, I know a half dozen people.
Maybe, maybe a dozen who can talk credibly about org change and leadership and know what the they're talking about when it comes to AI.
25:27 Sam Rogers
Yeah,
25:28 Paul
that is should be the most valuable coin of the realm right now, because the difficult stuff is the workforce stuff.
But yeah, we're spending all this money on technology.
25:38 Sam Rogers
Yeah.
25:39 Paul
the hard part is this part. I would not have anybody on my team that did not know was really good with AI. Why? Because
they'll be slow. if we need to have a deck ready for the client on Friday, someone AI powered will have it done in a few hours And also, if you don't know enough about what the
Client is trying to do in sufficient detail with a workload redesign or whatever you're trying if you don't understand the first thing about agentic AI how can you do org change? I think I would never ever
work for the change management person unless I was convinced they really knew what they were talking about or were prepared to dig in hard and learn But there aren't many. I know a guy that loves my work. I love him. I think he's great. He thinks I'm great and everything like that. But I'm like, have you tried Obsidian and Claude Code and coworking? He's like, nah, I don't really do that. I'm like, I'm not gonna work with the guy. I don't care how good he is in org change.
26:29 Sam Rogers
Yeah, yeah.
26:30 Paul
So I mean I think it's a rare space. Ought it ought to be very, very valuable, I think. whether it will be or not, we'll we'll soon see.
26:36 Sam Rogers
I agree that it it ought to.
Well, the the other part there that that you just kind of glossed over that I think makes all the difference in the world is the ability to call bullshit on the bullshit. Because there is so much of it, especially at
26:50 Paul
yeah.
26:50 Sam Rogers
this period in history, that is the AI snake oil and things that have always been the change management snake oil, right? when I say,
Managing agents is a lot more like managing people than anybody's comfortable with. What I mean is that you can tell a person to do something, you can have a policy around what people do, but it's really like the thing around them that keeps them in a predictable set of behaviors. It's not asking the person to do something, it's not a request, it's not a set of instructions.
27:19 Paul
Well that's nudge th that's kind
of nudge theory. You've just kind of articulated nudge theory. It's like context and environment is very predictive of behavior. I mean, we we change management. People talk about motivation and they talk about
27:27 Sam Rogers
Exactly, and it's much more predictive. Yeah.
27:31 Paul
purpose and they talk about skills and all that kind of stuff, but context that was the whole contribution
27:34 Sam Rogers
Yeah, but but that's all
27:36 Paul
of the yeah.
27:37 Sam Rogers
that's all frontal brain kind of stuff. And the environment that we're operating within is what makes all of the difference in change management, right? And and no different than with agents, like having a a harness where they can't bullshit their way out of it, of course their agent is going to have agency within that space and they're gonna get very creative about how they interpret those rules, no different than people, right? So the good news about all that is that if you have
ever managed anybody, if you have decent people management skills, you can apply those and get a big leg up over agents. It's not so much technical
28:11 Paul
I I
28:12 Sam Rogers
work.
28:13 Paul
I was I was a good leader, people prefer to talk about leadership rather than talking about management. I think being a good manager is extremely difficult. Now, if you're highly, highly competent, self starting, motivated and driven to learn, I'm probably the right kind of manager for you. Because I'll let you get on with it
Whatever the opposite of micromanaging is, that's fine. my son just graduated from college and he came to do some work with me for a couple of weeks. I sucked. Because I would be like, tell him to do something in the morning and expect him to talk to me at six PM and that the next time. You gotta like check in every hour, like what do you need and support and how's it going? Like I suck at that so much. So
28:48 Sam Rogers
Yeah.
28:48 Paul
So like I'm not a good manager, but that failure mode exists for agents too.
28:53 Sam Rogers
Yes.
28:54 Paul
Because if you're doing entering virgin territory with an agent and you're like, Yeah, go ahead and do this and come back and talk to me at six o'clock, I don't think it's not gonna work very often. Or you're introducing risks r
29:03 Sam Rogers
And you're gonna burn a lot of tokens real quick, yeah.
29:06 Paul
or burn a lot of tokens or you introduce shitloads of risk into the system. Yeah.
29:09 Sam Rogers
Well just in closing, if people wanted to learn more about you and what you do,
29:14 Paul
Paul Gibbons advisory. It's like drinking from a firehouse,
29:17 Sam Rogers
And I'll say for myself, I had plenty of fun engaging with his corpus of content and
29:21 Paul
that's fun.
29:22 Sam Rogers
seeing how it was built and learning from all of your wise words.
29:24 Paul
but you probably understand the architecture.
Most people wouldn't even look there. take it easy, my
29:30 Sam Rogers
thanks so much for joining Signals and Subtractions. we live stream every week, episode out on Friday, newsletter on Sunday. Looking forward to seeing you again real soon. Take care.