Prompt and Circumstance

PROMPT 19 -
===

[00:00:28] Tom Adams: And, uh, we are here, gentlemen. Good to see you. How are you today?

[00:00:33] Mike Richardson: We're back. Woo!

[00:00:34] Tom Adams: We are back and, and it's been a while since our last confession, and so I think, uh, we should start by just sort of sharing where we've been and what craziness has happened since our last conversation, which was pre-World Cup finish. Not that I wanna go into the World Cup, but at least we need to go back and just catch people up on where we've been in the last, I don't know, four to five weeks.

[00:00:57] Mike Richardson: But before we do that, hang on, hang on, hang on. You just used the word confession. Now, back in your history, you used to be a minister, correct?

[00:01:04] Tom Adams: Well, way, way back 35 years ago, so let, let's not get stuck there 'cause there was no confessions. There's no confessions in my world. No, not, not at all. They

[00:01:16] Mark Redgrave: Tom, we sh- we could speak more, you know? Now I know that. Like some spiritual guidance and moral thingy w- could be most

[00:01:24] Tom Adams: Well, we all need it at this point, so

Mike, let's start with you. What, what's, uh, what's transpiring in your world in the last three to five weeks? Yeah. You know, uh, you know, my, my hair's on fire still, as you guys know. I'm in the middle of selling a house and renting a house and buying a house and all that kind of stuff. But in the midst of all of that, continue to do my peer groups and, and my boards and all of those kinds of things. And, you know, always delighted to hear sort of after the fact that connections are being made, um, you know, beneath the radar scope.

[00:01:58] Mike Richardson: And I now know of three members who have engaged with Cadre AI, you know, the speaker that we've had in the past, Riley Strickland. Um, uh, and, uh, you know, all so far so good, and they've been through phase one, and they've sort of re-upped, you know, for phase two. And, and, and so that just-- that's just heartwarming, you know, to see people getting it sufficient to start placing some bets and rolling the dice and, and getting moving on something.

And so, yeah, I'm, I'm, I'm excited to see that gradually taking shape.

[00:02:32] Tom Adams: Very cool. Mark, what about you? What's happening in your world?

[00:02:36] Mark Redgrave: Yeah, we, um, so are we allowed to talk about the England-France, like, loser final thing? No, we

won't talk about that, okay

[00:02:44] Tom Adams: well you can. Yeah, because

[00:02:46] Mike Richardson: Six four.

[00:02:47] Mark Redgrave: Ah, what a game. Like if you, if you, if anyone listening still doesn't like really know about soccer and l- wonder whether it's worth, like just go to YouTube and find the England-France like playoff final from the World Cup. You won't see a more entertaining game of soccer anywhere. So we'll leave it at that, but it was fantastic.

So, and at me

[00:03:07] Tom Adams: to bring a wrap to our last conversation, uh, which was all about, uh, World Cup, um, AI stuff, so

[00:03:15] Mark Redgrave: Yeah. So it was a bit that, so that was, uh, that was quite motivating. You know, honestly, like we've... This quiet period, right? So, um, now I think what's, what, like, what's surprising is the quiet period hasn't been very quiet. I mean, it's, it's summer holidays. Um, you know, like traditionally we would, and I, I've definitely personally like taken some time in the last, over the last sort of three or four weeks just to like redo websites, like tidied some stuff up, which is awesome 'cause we all need to be able to do that.

Um, you know, created some more, uh, you know, just some more assets and things that I need in my business, uh, played with some new tools. So that's actually been really good. Like I've enjoyed it. But we, you know, like, uh, it really doesn't feel like we've had the summer lull that we normally have. And some of my clients, um, like we've had, uh, in the media and advertising, it's like we've literally had the biggest July, August ever.

So it's like, so, so, so it's kind of, I found that a little bit surprising. I don't know if you guys have felt the same, but it's like everything's just kind of powered on a bit more than normal. But anyway, that's, that's what I've felt.

[00:04:18] Tom Adams: Yeah. Well, I, uh, well, I've had a, a crazy month since we last talked. I, I went through the, um, the hospice and death of my dad

[00:04:28] Mike Richardson: Oh, yeah, yeah, yeah. Ah

[00:04:29] Tom Adams: and, uh, that, that was challenging in its own way. Um, I, my, you know, my coaching practice, like you, Mark, has, has been really busy. Uh, and then, you know, I've told you guys about my new AI startup, which, uh, we got literally this week, we got our first, uh, pilot, official pilot where, um, we're working.

And some of the older software that I built, it has a new pilot as of today. They just all logged in today to my HR voice notes, um, tool. Um, uh, the second client on that one. Um, so yeah, some interesting fun stuff is happening. So I don't know, and I'm trying to be on my boat because we only get 100... You know, you guys are in Southern California.

I'm in Buffalo area, and we only get 100 days, so I'm taking advantage. Last night out on the boat, tonight out on the boat. I leave my office, and I

[00:05:21] Mike Richardson: 100 days of

[00:05:22] Tom Adams: street and get them.

[00:05:23] Mike Richardson: 100 days of summer, or rather, do you mean 100 days of not being snowed in? Is that what,

[00:05:27] Tom Adams: Right. 100 days of not, whatever that is, but 100 days where you can, you can be on a boat.

So I'm on the boat as much as I can possibly be. So,

[00:05:36] Mark Redgrave: Tom, to the degree, to the degree you can share, tell us about the pilot, to the degree you can

[00:05:40] Tom Adams: Yeah, no, the pilot is, uh... So, uh, we have a, uh, built a software called Simparai, which is conversational intelligence, which I've talked about before. Uh, it does lead triage with, uh, multiple, um, agents at play at the same time in the conversation to actually reduce the cost and complexity of onboarding a new client or deciding whether it's a good new client to take on.

And, uh, it's actually an insurance company that's, uh, looking at it and to decide if it's a good fit for them in terms of their practice. Uh, they don't want personal, personal lines. They don't want that. They want business to business kind of stuff. And, um, they spend an inordinate amount of time chasing people who have filled in forms, and they're gonna process through our, our toolkit as the way to get the, to the result.

So, uh, got our first pilot, which is kind of fun. So

[00:06:33] Mark Redgrave: That's awesome though. And what, Tom, what does... So, so in terms of like what good looks like, like how, how do they, how will they know whether this is value creating?

[00:06:41] Tom Adams: If, uh, it completely reduces any chasing of leads that come through their front door, 'cause that's what they do now. They get a lead through their front door, say on their website, they fill in a form

[00:06:52] Mark Redgrave: is, which is pay, organic or paid, right? But it's

[00:06:54] Tom Adams: Yep, organic or paid. It comes in the front door, and then a lot of times there's nothing in the form, so they don't know what to do with it.

They don't know what kind of a, a opportunity it is. So somebody has to pick up the phone and call, and then if they don't get them the first time, they have to do a second call, right? Then they've gotta go sort of chasing the opportunity that already came in the front door. And so what, what they're gonna do is go through our conversational platform, have this conversation, uh, which is, uh, literally four to five minutes, and that actually triages the opportunity and figures out is this a good lead?

Is this... Does this fit their profile?

[00:07:32] Mark Redgrave: So Tom, there'll probably be some people listening where like, like this is a classic sales funnel for them, right? So, so what you're describing is there's, there's like, there's, there's paid or organic or there's like, there's like inbound, and that process you're describing is like a, is like a se- is a lead qualification process.

So you'd call that like from a marketing qualified lead to a sales qualified lead. That's what you, that's what you might traditionally call it, right? So, and you're saying that, that what you're building is doing the sales qualifying bit,

[00:08:03] Tom Adams: Yes

[00:08:04] Mark Redgrave: Okay. Cool.

[00:08:06] Mike Richardson: Yeah.

By the way, Tom, you will,

[00:08:08] Tom Adams: a ch- not as a chatbot. It's not a chatbot, it's actually a conversation. It looks like you're in a meeting, and you go into a meeting and two or three people are asking questions, um, and those, those are being sort of supported by a multi-agentic structure to support the whole process and to be able to audit it and be able to add all kinds of intelligence on the back

[00:08:28] Mark Redgrave: And it's pulling publicly accessible information about the cl- the, the business into that conversation? Okay. Ooh, man, I like that, Tommy. I like it

[00:08:37] Mike Richardson: Tom, you'll be sending me a commission check every time you use the word triage, is that correct?

[00:08:43] Tom Adams: Right. Right. I will. I mean, because

[00:08:45] Mike Richardson: Okay, good. Just, just, just checking. Keep it-- keep in mind that my address is changing, so be sure it doesn't get lost in the mail, okay?

[00:08:51] Tom Adams: Right, right. It, uh... I will send

[00:08:53] Mark Redgrave: won't get, Mike, it won't get lost in the mail 'cause it won't be in the mail.

[00:08:57] Mike Richardson: correct. Yes, yes. You'd be lost in the ether.

[00:09:00] Mark Redgrave: That's very cool though, Tom. Like, I love that because this, this is like, so, so, you know, like, um, for a lot of people like that, that, that, that like, and the MQL is, is, is like easier. The, the sales qualified lead piece, like that's a d- that's a very intense human thing and offshored massively, right?

So for a lot of companies it's like this is work that is done Philippines, India, like, you know, because it's like, because it's like, gosh, we've just gotta get through so much. And I, and I feel like there's a lot of chat, um, like out there at the moment about like the, uh, the, the AI and agentic possibilities, um, around like, uh, like catching all these leads that are either coming already in the system that never get qualified correctly, right?

And it's, and it's, and it's such a, it's such a tangible value creator, right? 'Cause you can literally measure it, can't you?

[00:09:50] Tom Adams: Yep.

[00:09:51] Mark Redgrave: what, be- because at the end somebody sells something and like that's al- that's always a really, really like very easy way to determine whether the investment's worth it.

[00:09:59] Tom Adams: Yeah. Yeah. So, so that's fun. So I had this, uh, I read this article, um, uh, last week, and I thought it was something we could talk about today 'cause I think it affects, uh, both ourselves, our clients, uh, the work we're doing in the world, and it was inspired by, uh, this article from Dan Pupius, who's the chief technology officer at The General Partnership, which is a venture capital firm.

Uh, but he built Ramp. He was the CEO of Ramp, which was a financial tool. Uh, he used to be with Google. He built the Google Chat program, so he's got chops in that regard. And, um, and the, the concept comes from the framework that he built, so I'm not taking any credit for this. This is his framework. Um, and it came out in an article recently, um, because it really gives a better way, I think, to talk about some of the stuff we talk about related to, um, the decisions we're making.

So the premise that he wrote about is whenever we make an AI decision, we're making a bet on where the world is actually going related to AI. We might call it a technology decision, what tool we're gonna use, whether to build an agent, whether to automate a workflow. But underneath it is a view that we take of the future, and the four framework...

or the framework names four of those specific bets that I think are really helpful. None of them to me is right or wrong. The question is, uh, which end of the spectrum are we betting on, and how conscious are we? So I'm gonna go through the four pretty quickly, and then we can talk about it. So the four bets he talks about, the first one is token economics.

Uh, is your bet, um, that it's scarce, uh, tokens, uh, are gonna be scarce or abundant? So does the cost of running AI become high enough to constrain us or fall towards zero and give us unlimited capability? Because if it comes cheaper, experimentation and always-on AI experiences become more practical, but cheap intelligence also makes it easier for competitors to play.

So that's the first one. The second one is model self-sufficiency. Um, and model self-sufficiency is, um, do you need more scaffolding around things, or does the model handle things natively? So the question really is: would our product, our workflow, our service, our value proposition still be needed if ChatGPT or, or Claude has dramatically better capabilities going forward?

So some

[00:12:24] Mark Redgrave: are not one-word answers, Tom, you know that. All

right.

[00:12:30] Tom Adams: I wanna talk about the bets, right? Well, the next one I think though is not an one-word answer. Platform structure, um, locked in or commoditized. So are we locked into one AI provider? And I know the two of you and I have deferred a lot of our attention to Claude recently. Or can we switch without breaking important parts of our business where all of our stuff is housed?

Going deep with one provider offers speed and differentiated capabilities, but staying flexible reduces dependency, and so that's a bet that we're all making. I know I'm making it with the software I'm building or the tools I'm building. And then the final one is trust and governance, permissiveness or constraint.

A- and so the question is, do we move fast now and deal with governance later, or do we invest in oversight and accountability upfront? Governance is more than just formal regulation. Uh, now, I, I've just heard recently that if you don't have an audit trail completely in place, if you're using AI, you are actually l- uh, liable in a court of law if you don't have some kind of audit capability related to that.

Uh, and that's coming out of the big consulting firms right now. So, um, what do you do-- what are we thinking about that? Because it, it... And this, I think, reflects more towards builders, but if we're doing that and, and we're talking about that, and we're using AI with our clients or encouraging them to go directions, those are what we're thinking about.

So

[00:14:00] Mike Richardson: Beautiful.

[00:14:01] Tom Adams: those are the four axes, and I, I, I feel like it's a really good perspective to work with. So when you hear that every AI strategy is really set of bets about the future, what bets do you think that I've just named there are leaders making today, uh, from your perspective?

[00:14:18] Mike Richardson: before we get into that meat, 'cause I've, I've, I've s- um, re-speed read the article, that is the meat in the sandwich. But let's talk about the sandwich of the opening context and the closing context in which he frames that meat. And I think you're gonna really like this, Mark.

Um, I'm sure you read the article as best you could as well. He , starts out by saying, "Don't ask yourself the question, what do we think about the future? What's our prediction about the future?" What he's really getting at is how can we be ready for a range of futures? And yes, to some degree we're gonna place bets, but to some degree we're gonna hedge our bets.

We're not gonna be prematurely pouring cement around things that could radically shift, and now we've gotta chisel up that cement and pour new cement. And so that's for me, when I, when I read the sort of opening part of the sandwich. And then the closing part of the sandwich, he basically, um, asks, uh, three-- four things, Mark, that I, I know resonates so well with your language.

He, he, he asks, "What is our implicit bet?" is the first question. Secondly, he asks, "What would have to be true for that bet to pay off?" Thirdly, "What signals would tell us we're wrong? What early warning signals?" And then fourthly, "How quickly could we adapt if we, if, if we have a critical mass of evidence to say that we're wrong?"

So I just wanted to jump in there,

Tom, 'cause I

[00:15:59] Tom Adams: really helpful. Yes

[00:16:00] Mike Richardson: that framing really puts those four bets into, into context.

[00:16:04] Mark Redgrave: which by the way, Mike, in terms of the closing bit, is classic agility

thinking.

[00:16:09] Mike Richardson: Yeah. Uh,

[00:16:10] Mark Redgrave: It's like, isn't it? It's like, it's like, and this is why the, the more everything changes, the more it stays the same. It's like there are fundamental principles here of like, o- of like agility and, uh, tracking and measuring and being able to determine when to move and the ability to pivot.

Like those, those are, those are, those are

[00:16:28] Mike Richardson: Yeah, I mean, you could to some degree,

[00:16:29] Mark Redgrave: talked about that a lot

[00:16:30] Mike Richardson: you could to some degree retitle his article into Agile AI, right? How do you take an approach to AI with embedded agility to be able to adapt, adapt, adapt, you know, as quickly and cheaply

[00:16:46] Mark Redgrave: So, so, so

[00:16:47] Mike Richardson: reversibly as you need to

[00:16:50] Mark Redgrave: let's-- Those four things, like, uh, like let-- Do you wanna tackle them one by one or do you wanna, should we just go at it

[00:16:55] Tom Adams: Yeah, no, let's tackle the one- Let's start with token economics, which

[00:16:58] Mike Richardson: and am I allowed, am I allowed to ask dumb questions as we go along? Am I allowed to

[00:17:02] Tom Adams: please do. Scarce or abundant? Token economics, scarce or abundant?

[00:17:07] Mark Redgrave: Okay, so I don't, I don't know, right? I, I mean, you'd have to, you'd have to argue that like, that, that, that... I mean, the economics is so complicated right now because of like, because of fundamental ingredients around power. You know, it's like this is, this is why it's so difficult. And when you realize that, you know, when, when you've got people saying, "Hey, we need 12 nuclear power stations in the next, in the next two years to even power this stuff, and it takes us 10 years to build a nuclear power station," you kind of go, "Oh, well, that's might be a problem."

So, so I, I kind of, I kind of feel like, and that's way above my pay grade. I don't even know how to think about that problem, right? Um, I do know though that like I, I read something really interesting the other day, which was like we-- with these models that are coming out, right? There, there's, there's like the latest great...

We're like, lots of people are chasing the latest, greatest. And I re- I, I listened to this great podcast and this guy said, "Look, like for, for 95% of what we need, you do not need the latest, greatest model," right? So like this concept of like load balancing your requests, I thought was like really like, like, I, I guess obvious, but like, yeah, that's gonna be a thing real quick, right?

So like when we hear these stories of enterprises and they're like, "We blew through $7 million worth of tokens in the month," it's like we're gonna need to, we're gonna need to be able to, to channel requests to different models. And I assume, guys, that like the aged models, okay, if they're still supported, will be significantly cheaper.

So, so, so this is always the case. Like it's, it's no different to, you know, we use 1% of the features on our iPhone, right? But we've all got iPhone 17s. It's, it's, I, I, I kind of feel like that's a thing. So I guess that plays in this space. Or you, you'd have to say tokens will be abundant. I would have to make that call.

But the f- but, but like we're gonna be mu- we're gonna need to get much smarter at how we use tokens

[00:19:08] Mike Richardson: And, uh, uh, proudly from the slow guy in the room, dumb question number one. Can you please describe for myself and all of those listeners behind me, what do you mean by token economics?

[00:19:22] Tom Adams: Token economics is the ability to... So when you, when, when you load up ChatGPT or Claude or use it via API, um, a token is a unit of measurement for how much you consume or you provide. Like how much Claude or ChatGPT gives you, and how much you ask it, and what the processing requirement of that is. And so, um, for example, if, uh, token economics are high and you've got a million context window, which is what the current context window is if you're paying for a good version of Claude, you get a million tokens context window.

Um, the question is, are you willing to s- spend or are, are those gonna be constrained from a price perspective such that you only use Claude, um, and it costs you, uh, $6 to get information back from Claude when you do that back and forth? So, um, whereas what I see happening is you see some of the Chinese models coming out.

Recently, K-3, um, from Kimi comes out, which is another model that none of us generally use unless you're really into the space, and it's now something that they gave away free. They literally gave it away free, and if you have the bandwidth on a local computer or set of computers to run it, you can-- you got that essentially free.

So $25 from Claude or, uh, even if you use Kimi directly with API, those tokens are only costing you $3 versus $25. So that's, that's kind of the thing here. Tokens are just the unit of measurement in AI

[00:21:07] Mike Richardson: of sort of processing power required to get stuff done

[00:21:11] Mark Redgrave: And, and so, so just to link those two things together, Mike, so let's-- So like the, the latest, greatest models will use more tokens, right? The, uh, the, the, the ge- the, the previous generations will use less, and that will be classic pricing. That'll, that, that, that will, that will follow l- absolute traditional pricing rules

[00:21:28] Mike Richardson: And what I'm a little confused about is do I, as a user of Claude, have visibility to the consumption of tokens that I'm working through? 'Cause only-- I only give it-- ever get told when, uh, you've run out.

[00:21:42] Tom Adams: Right, you do. You have to go into your settings and there in your... You go into settings, into your settings column, there is a, a, there is a slot called usage, and you see how much capacity you have in your given plan,

[00:21:55] Mike Richardson: thank you for the little, uh,

[00:21:56] Tom Adams: Well, I guess, I guess from my perspective, 'cause I'm also building software and using it, building tools and using it, and the AI is actually embedded in the tools that I'm building 'cause it's AI that's actually facilitating things, um, I have to bet on abundance. Uh, my bet in this case is towards abundance, but I'm also building it such that I can switch, right, to the lower price provider.

In fact, I'm building what are called small language models. So like you said, Mark, you don't need all the smarts in the room for this particular task. If you build a small language model, it only does these 22 things. That's all it does. And so that one costs me nothing to run other than the hosting capability, whereas the large language models are where all the costs are incurred.

But I gotta bet that it's all going towards the, the price of tokens are going down over time

[00:22:54] Mike Richardson: Yeah. And, and this for me brings up kind of another general principle, which he sort of speaks about somewhere in the text, and that is to some degree in each of these, you're gonna be on the fence a little bit like you were, Mark, I don't know. But you're gonna lean one way or the other, right? You're going to lean towards abundance, but remain, um, aware of the flip side on a just in case basis, because if I'm wrong, I need to have inbuilt flexibility to be able to adjust cheaply, quickly, and reversibly

[00:23:30] Mark Redgrave: It's-- The economics are really bizarre though, right? Like I'm, I, I don't know why, but I'm sort of fascinated that all of this inve- not all of it, but a lot of this investment, trillions of dollars is sitting off balance sheet. Like the economist in me goes, "That's super weird." It's not illegal. It's not... No one's suggesting it's illegal, but it's really weird.

So, so like the true cost of this is not known, guys. Like, like it is, there is no way that like... 'Cause if the true cost of what they were doing was actually reflected in the balance sheets, all these companies can close the doors tomorrow, and all the directors are probably trading insolvently. So it's like, so, so it's like, which is clearly not the case and it's not illegal, but it's so weird.

And you, and you kinda go, y- you know, you kinda go, I believe, I believe like during this conversation, like w- we'll be abundant. But like the economic-- The tokens will be abundant, but the economics are not clear. Like, like how are these guys gonna make money? Because all they're doing at the moment is losing a massive amount of money, hundreds of billions of dollars of money.

And you kinda go, "Well, that's, is that sustainable?" I mean, I guess if you keep it all off balance sheet, maybe it is. Maybe we can just print the money again. But, but the eco- but the economics are super weird, no? Like if you're, if you're a business owner, like this is, like you, you can't, you can't run your company by these rules.

[00:24:50] Mike Richardson: Right. Right

[00:24:52] Mark Redgrave: But these guys can. So it's like, I, I don't really know where that ends us, Tom, honestly. 'Cause it's like, it like on paper it's like this is completely unsustainable, but it's clearly not going away.

[00:25:03] Tom Adams: Right, right. All right, let's move on. Uh, the next one is model self-sufficiency, and the, the question here is, would our products, our workflows, the toolkit we're using, uh, still be needed if ChatGPT or Claude or Gemini or Grok, 'cause Grok just came out with a wickedly cool product in the last week, um, would...

If, if that toolkit just keeps getting better from the four majors or, you know, the 10 majors, whatever they are, um, large language models, um, would the stuff we're doing make sense anymore? Like, I asked that. I've had to ask that of myself in terms of the tools that I'm building. If they get better, do they need my tool?

Does, does the insurance company, does the accounting firm need my stuff? And that's, that's, I guess, the next question. if ChatGPT or Claude or Grok gets better, do we need the stuff I'm building or the stuff we're doing at some of our companies, the investments we're making in Cadre or whoever we're making Thoughts, Mike?

[00:26:09] Mike Richardson: I was thinking back, you know, to my aerospace days when we would be, um, you know, providing specialist elements of the aviation system of a modern commercial airliner. And we had to compete at several levels, one of which was we were always competing, if you like, of where we fit in the food chain.

Because if someone above us in the food chain, uh, rolled up our functionality into their bigger box, we're now irrelevant, as you've just said, right? So you're, you're kind of saying the same thing, right? If Microsoft or, or you know, some- or Salesforce or whatever, if they start to perhaps acquire and roll in all of this stuff and they become the mother of all AI applications, then everybody else, to some degree, is being- coming at risk of being irrelevant

[00:27:08] Tom Adams: Right. And that's the bet. Like, that's the bet we're all making in some regards.

[00:27:13] Mike Richardson: Yeah. And for me, I think, I think these are great bets. I like them a lot because they are specific versions of bets that we've always had to make. We've always had to make, make buy decision bets.

[00:27:31] Tom Adams: Mm-hmm.

[00:27:31] Mike Richardson: We've always had to make centralize or decentralize decision bets. So really, w- we're familiar with making these kinds of bets.

These obviously are very specific to AI, uh, but I like them a lot.

[00:27:47] Mark Redgrave: I mean, I mean, I can't answer a question, will these models become self-sustaining? Like, it has to be yes. It has to be. I can't see it landing anywhere else. How long? Don't know. Like, you know, like, like, and, and will there be a tremendous opportunity for suppliers and partners to wrap around the ecosystem for years to come?

Yes. But like I, I'm reminded as well, Mike, I loved your example there actually, your, um, your aerospace example. But like, like, uh, like similarly, you know, like we can fly, take off, land planes fully autonomously today, right? But we have pilots in them. Okay? It's like, it's like, it's the same for me. So, so like it, it just because this can be done autonomously, like are we ready?

[00:28:33] Mike Richardson: Well, there's a famous joke, there's a famous joke, Mark. As, as, as, as cockpits became more and more automated, the joke was that eventually cockpits will have one pilot and one dog, and the pilot is there to feed the dog,

[00:28:48] Mark Redgrave: Yeah.

[00:28:48] Mike Richardson: and the dog is there to bite the pilot if he or

[00:28:51] Mark Redgrave: To stop the pilot. Yeah, to stop the pilot. I love that. I love that. I love that. so we, you know, we, we kind of get, get to the point where you say, like, when it comes to governance and when it comes to like, uh, accountability, like are we prepared to defer 100% accountability to digital systems?

I don't know. Like, I don't even know how that works if you're in a heavy regulated, publicly traded company. I don't even know how that works. So there's some very big stuff at play here, right? Which is again, way bit-- Actually, it seems like everything today is beyond my pay grade. But, um, but,

[00:29:24] Mike Richardson: Well,

actually, seeing as you, seeing as you said that, Mark, I-- if you're watching the video, everybody, Mark's label here says Mark Redgrave/Shift. Is that a typo, Mark?

[00:29:36] Mark Redgrave: Yeah, funny. Mate, I'm a marketer. I'm a marketer. I take every opportunity. I'll be-- I'm gonna wear a Shift branded T-shirt next

time. Just

[00:29:44] Tom Adams: as you should. As you

[00:29:45] Mark Redgrave: you watch. Wait till I get my

[00:29:47] Mike Richardson: I don't think he, I don't think he got my joke yet, but he will

[00:29:49] Mark Redgrave: No,

[00:29:50] Mike Richardson: on. It's a delayed, it's a delayed

[00:29:51] Mark Redgrave: your joke. It's not a keyboard, it's not a keyboard mistake, Mike,

[00:29:55] Tom Adams: Shift, shift happens. Um,

[00:29:58] Mark Redgrave: Upshifts Creek

[00:29:59] Tom Adams: uh, from, from my perspective, I, I believe that the models will keep getting better and better, but I think the, the thing that I'm seeing is integration sometimes between all the real-world messiness of stuff that companies have. Um, for example, I'm looking at this, uh, this insurance client and realizing, uh, to integrate AI, they have to fundamentally change process maps, right?

How things happen. Um, and it-- And so-- And then connecting the dots to all these things. You can go on Groq, uh, ChatGPT, Claude now, and connect everything. The problem is, once you connect it, does it all flow in the way you want it to flow, or is it still frustrating? And I think sometimes, I mean, my lean is towards it still needs some kind of scaffolding.

It's still gonna need some kind of support, uh, because it keeps changing so rapidly that you not just need the scaffolding, you're gonna need the ongoing maintenance of the scaffolding just to make sure this works over the long haul. So, uh, it doesn't necessarily answer the bet, um, is it fully native integration?

But at the same time, uh, I'm betting on the fact that there is some degree of scaffolding needed in the future for all of

[00:31:13] Mike Richardson: actually it's interesting because one of my members the other day said, "Look, I've got this friend who's also a CEO, but on the side they do all this AI stuff, and she happens to be here right now. Can she join us in the one-to-one?" And the question they asked was, "You know, we think we can do all this stuff, but then we talked to an AI consultant and they said, 'Don't do it, because when it breaks, it's gonna be a real problem.'"

And so your point about you're still gonna need some level of scaffolding that not only has to be built, it has to be maintained, and when it breaks, which it will, it has to be fixed. And if it's become mission critical and you are now dependent upon it, and you perhaps, you know, discontinued other things you used to use that are not there anymore as a fallback, you're now in trouble

[00:32:01] Tom Adams: Correct.

[00:32:01] Mark Redgrave: Yeah. And, and, and, and another thing connecting to exactly the same point, Tom, I read, uh, this morning on an article on LinkedIn, and they're like, "Guys, like whatever rules and prompts you put into your business like, like, um, six months ago, like way out of date. Gone. Like complete pointless." So, so it's like, so, so when you talk about maintenance and it's like, yeah, like we- we're all doing this with good intention, but it's like, like this is changing so unbelievably fast.

So, so like what are we gonna do about that? Because you did, "Hey, Mr. Customer, you did a great thing like making sure that everyone was using these prompts." It's like they're completely irrelevant today.

[00:32:38] Mike Richardson: Yeah

[00:32:38] Tom Adams: Yeah, like Fable right now, Fable, like the latest, which is Claude's sort of biggest model that they kind of half pulled because of the government stuff. Um, right now in terms of development and coding work, they're saying pull out all your prompts. All you say is like pull out all of your rules and your rule

[00:32:56] Mark Redgrave: Yeah, 'cause it's just constraining us

[00:32:58] Tom Adams: It's constraining us now, so literally just pull everything out and tell it what you need, and it does it now. And I'm going, I got, I got dec- like I got full booklets in there about how you're supposed to operate, and now it's saying get rid of them

[00:33:11] Mark Redgrave: we sat on this podcast nine months ago and said, "Hey, how you structure your prompts is really important. Don't bundle it." Right? Like now, literally now they're saying bundle it.

[00:33:24] Tom Adams: Yep,

[00:33:25] Mark Redgrave: Yeah, like literally that's the advice. It's like, "Just whack it all in. We've got

[00:33:29] Tom Adams: Yep, we got it. All right, number three, platform structure. So this, this affects us personally and I think companies. Um, the question is, are you locked into one AI provider, or could you switch without breaking the important superstructure of your business, your life, your stuff? And I know, uh, we've all kind of moved Claude, um, centric in many ways, and at the same time you go, "Uh," like, what happens now if something in Claude...

'Cause Claude's latest round for a lot of people has been messy. It's like, uh, what is gone on with Claude? It's, it's become kind of a pain in the butt recently. So I guess the question is, what are you betting on there? Sh- like, do you lock yourself in or do you commoditize and think about how can I move this stuff and what happens when I need to?

I remember Mark sent me a message and said, "I'm trying to leave, uh, uh, ChatGPT to go over to Claude. How do I get over there and bring all of this history that I've built?"

[00:34:29] Mike Richardson: I'll go first on this one. I, I-- 'cause I'll use your language, Mark. I just can't imagine a future in which we don't remain locked in, um, because we're locked in today. If I wanna move from Microsoft to Google or to Apple, it's a, a major headache to do that. I mean, my wife constantly comes to me, she's an Apple person, she's constantly coming to me saying, "I can't find this file that I just saved."

And I just throw my hands up saying, "I can't help you. I don't even know where to start. If it was on my Microsoft machine, I could find it in five seconds, but where it is on your machine, I have no idea."

[00:35:07] Tom Adams: Right. We, we have a rule in our house which is my wife and I have to have the exact same phone because I can't help her unless she has the same phone as me. Otherwise, there's no- it's completely illegal to have different phones in our house

[00:35:22] Mark Redgrave: so I, and I, I think we've, we've spoken about this, the three of us before, but it's like I'm reminded of s- of Salesforce and Oracle, right? And I'm reminded of like when you force switching, right, like the market becomes closed to you.

So Salesforce massively accelerated their business when they decided to take a position of like, you don't need to remove Oracle to use Salesforce. Play nice. Yeah. Like, if we all play nice in the sandpit, we can all do what we need to do. That's how this must go. I cannot see a future where integration of any tool is fluid and immediate, because for me, like, uh, that's the only way that everyone's gonna get what they want.

Now, it doesn't mean everyone will have access to all the tools, but I think there's gonna be this massive integration layer built which allows people to stand up, like pretty much, uh, you know... What, what does Salesforce used to say? It's like clicks, not code, right? That was their whole thing, like configuration.

Don't, don't write code, configure. And it's like I see an analogous situation to that about to happen. There will be these platforms built where you can click in what you need. Microsoft are already doing this because in, uh, in Microsoft's products, you can use Claude, you can, you know, like there's a number of, there's a number of options.

So I believe this is what's going to happen because it's the only way the market will truly keep up.

[00:36:51] Mike Richardson: And what I tend to find is yes, and at the top end of that, your advanced users typically have pushed the envelope on some of the more sophisticated things that they're able to get out of the system, which when you try to move over, that top layer of stuff breaks because, because that is the sort of, you know, more sophisticated stuff.

And they may be in the minority of, of, you know, most users are not, are not using the last, you know, the last mile of, of functionality, but the advanced users are, and, and that's the stuff that breaks most readily, and they're shouting and screaming, you know, "Hey, I used to be able to do this and I can't do it anymore."

[00:37:34] Tom Adams: Yeah, and, and, and I, on the other hand, um, I'm already experiencing what Mark talked about because I don't use ChatGPT or Claude. I don't use their apps. Everything's on my desktops, right? So I have it in my computer, and then I bring the power of it to my... not to my desktop app, into the bowels of my computer.

And so I'm able to run ChatGPT, Claude, Codex, all of the tools on anything that I'm working on at any time. So, like, I don't let Claude, um, judge itself. I only let Grok or ChatGPT or Codex now judge what I'm doing with Claude. So I'm using all of those inter- intermittently. So I'm commoditizing. I'm working from a commoditized perspective, and I don't use the...

I don't use their top layer, which is the app that everybody uses. I use their, I use their

capability without getting stuck in the, in the tool

[00:38:35] Mike Richardson: And I think Tom is solving for anti-gravity as well while he goes along. He's a, he's a very advanced user of all this

stuff

[00:38:42] Tom Adams: but I- I'm, I'm not trying to say that. I'm just trying to say that's where I'm making my bet. My bet is, so like in, in my Simpera product, my ch- my conversational intelligence, um, Google has a tool that allows you to pick any of the models to use, and literally you plug the model into the toolkit and it's really quite impressive.

So Goo- Google's doing some interesting work, um, in the platforms versus they, they may not be leading in the actual models, but they're winning on platform 'cause their platform is sensational

[00:39:15] Mark Redgrave: somewhere in here, and I'm sure, I'm sure this already exists or is being worked on, but like surely there's a platform here, right? That's acting as this master integrator that's saying, "Right, here's the AI workload from this enterprise, from this company," right? This stuff is like this, we need to, we need to flow this.

It's, it's like, um, it's soft, it's easy. Flow it to this low-cost model. Right now that looks like ChatGPT 3 point... I don't know, I'm making this

[00:39:43] Tom Adams: Right.

[00:39:43] Mark Redgrave: It's like, but like that load balancing and distribution of load to the right model, that's the thing. Wrap it in enterprise security. That's like a, that's a middle layer that feels like someone is going to own that, and the economics are gonna be if, if I'm fl- like I get paid every time we flow, we flow workload to ChatGPT, we get paid, we g- we clip the ticket every time.

Like there's a, there's a, um, there's a platform there somewhere. But it, and it, and, and the, and, and what it does is it gives you like a view of enterprise security and governance, yeah, at that platform layer. Somebody must be building that

[00:40:18] Tom Adams: Yeah. Well, there's something similar, uh, but it, it takes a little bit more work, and Stripe is noted to be looking at buying it. It's called OpenRouter, and OpenRouter gives you access to every model through a, through a single API key, and you can actually build in your prompt structure which models you wanna use for particular things, and it just, it just pings them up.

And you, you pay more for that, but you only have a single interface to every single model available in the universe,

[00:40:47] Mark Redgrave: That's what I'm talking about. And it-- and you shouldn't even have to tell it, it

[00:40:50] Mike Richardson: Let's do that. Let's do that then. Let's do

[00:40:52] Tom Adams: Let's do that. Yeah. All right, final one, 'cause, uh, I know Mike has, you know, Mike's, Mike's got a lot of stuff on his plate, so we can't mess with Mike's time here.

Final one is, is governance, right? Is, uh, I think a lot of times now we're like Wild West in AI, and I, I think we always know that eventually things move towards governance, especially with corporate, with, uh, enterprise level. So the bet is, um, are you willing to be permissive now and then deal with governance later, or do governance now and be per...

And, and then, um, be permissive now and then build it over time? So I, I mean, I don't... I, I, I'm just interested in your thoughts on that one

[00:41:30] Mark Redgrave: Oh, capitalism says, capitalism says we'll deal with it later.

[00:41:35] Mike Richardson: Exactly. Wild West, um, you know what, and then when it gets too painful, uh, to continue that way, then we'll finally, you know, do the, do the legwork we've gotta do. Um,

but,

[00:41:49] Tom Adams: so weirdly, I'm playing differently. I'm building governance in right now. Um, even though I've played a lot over the last couple years, uh, in, in my, in my primary, uh, new software, I'm building my AI software. I'm building governance in right from the baseline, so it already has full audit capability built in.

I can audit every conversation, and I can audit why the large language model decided to say what it said in my conversation

[00:42:19] Mark Redgrave: I

[00:42:19] Mike Richardson: a smart move. I think that's a great move. Yeah

[00:42:22] Mark Redgrave: Yeah, I misinterpreted the question, Tom, actually, because I was talking at the high level platform level, right? Because it's like there's this, you know, there's this like ethical and whatever conversation with the platforms about governance and control. I was thinking at that level.

At, at the actual product services solutions, I agree with you, but like build it in now 'cause I... Like clients are saying this to me. Yeah, clients are saying like, "What do we do about..." Especially in HR and stuff, "What do we do about governance?" Because like, have I now got a load of risks emerging by people using this stuff and creating outcomes that are, you know, like, like is there risk in that for me?

I think that's a thing and I think we need to be dealing with that now. But when it comes to like, are these platforms gonna be governing themselves? No, they will not.

Until, until

like, yeah, it all turns to shit and then someone's gonna have to do something

[00:43:14] Mike Richardson: And by the way booking a new speaker for my groups who talks about the cybersecurity risks of AI, right? Where all of a sudden, yeah, you're, you're blazing a trail. It's the Wild West. You're doing all this AI stuff, but you're just opening, opening the back door for all kinds of cybersecurity threats and risks

[00:43:31] Tom Adams: Yeah. Well, I just, I thought this was an interesting set of sort of, um, bets

or continuums that, that I think, like you said, Mike, earlier are something we've always dealt with. But for the sake of this conversation, I think for the sake of people who are listening, uh, I think it's an important set of conversations to have, even with all the unknowns to these, these four categories.

And I'll make sure in the show notes we link to the article 'cause I think that's helpful.

[00:44:00] Mike Richardson: And in many ways, in many ways, just to sort of go full circle, Mark said it earlier, really this is agility. That's what agility is, right? Agility, when you really boil it down, is about a flow of decision-making and of course the actions that follow those decisions, and the questions and thinking that goes into those decisions.

And every decision is a bet. And, and how do we unbundle a big decision into a progression of smaller bets so that we don't have to bet the farm right out of the gate, where we're gonna make the s- the smallest bet we can and hedge our bets on the rest until we have no more choice, where we're either losing traction because we're not placing a bet or the world has shifted and, and the landscape has become clear and we now need to decide one or the other in a binary way.

It's this or that. It can't be an and anymore. We've now got to get down off the fence and decide are we going that way or that way?

[00:45:03] Mark Redgrave: Yeah, I th-- Mike, I think, I think that's absolutely right. And, you know, like I love the point you just made about like small bets. Stop making big bets, yeah? Like make small ones. And if you don't like where you are, take a step forward and look again. You know, it's like, it's like just classic stuff, right?

Don't-- certainly if you're a business leader, like, like don't f- like, and you're overwhelmed with these big decisions, you should absolutely be thinking about what small decisions can we make that give us more insight, move us forward, put us in a different position to be at a different perspective, right?

And that's classic agility. Don't get locked into things right now, right? 'Cause it's, it's moving too fast. You must remain with, you know, optionality and you need to preserve agility

[00:45:50] Mike Richardson: Yeah. And, and just to put a cherry on top of that, of course, one of the most famous authors of all time, Jim Collins, in his book, not Gre- Good to Great, his fir- his... Not his first book, but you know, the, the, the first of the series of books, and then he wrote a book called Great by Choice. And in that book he said, "Never, ever fire uncalibrated cannonballs. Fire bullets first." Fire a progression of small bullets, get calibrated to what works and what doesn't, and only once you're calibrated do you then sort of work on the bigger thing, the 2.0, 3.0 thing of firing a cannonball, um, w- without having to sort of bet the farm.

[00:46:38] Mark Redgrave: Yep

[00:46:38] Tom Adams: Lovely. Well, gentlemen, this has been fabulous as always. I love our, I love our conversations, and thank you for, um, sharing them. And, uh, yeah, we'll catch up next time

[00:46:48] Mike Richardson: Yeah. Get that typo, get that typo fixed, Mark, please for next time

[00:46:51] Mark Redgrave: Uh, yeah, will do, mate. Yeah, thanks.