#9 - 21 Apr 2026

Episode 9: AI Deniers, AI Slop & KnowledgeFlow Cracks Salesforce 🍺

Neil's on a non-alcoholic beer again — this time because he's in the doghouse with Mrs Watkins and needs to drive her to a romantic weekend away to patch things up. It's that kind of Friday. Welcome to Episode 9.

Mrs Watkins is an AI denier — and she's not alone Neil tried to convince his wife of the wonders of AI. She said "it's great but it's not for me." Sound familiar? Neil points out this is almost word for word from Richard Susskind's How to Think About AI — a whole category of people who can see the value but won't change their process to fit around it. Then again, Staples' share price apparently collapsed when Mrs Watkins switched from post-it notes to spreadsheets, so perhaps there's hope.

The future of management — courtesy of Nate B. Jones Neil recommends a brilliant piece by Nate B. Jones (his second plug this series, and no, he's not on commission) on what management is actually for. Three roles: routing information to the right place, sense-making in the noise, and accountability. How much of each is AI-able? More than most managers would like to admit.

Trust in a world of AI slop Can you trust a video anymore? A photo? A LinkedIn post? Kieron raises the uncomfortable reality that AI-generated content is everywhere — including on Instagram (those animal rescue videos? Mostly fake). The organisations that will win are those with genuinely trusted brands and curated data sets — like the King's Fund or Stripe. Being a trusted source is now a competitive advantage.

Anthropic's Claude 4 Opus (Mythos) — the model you're not allowed to have Kieron digs into the buzz around Anthropic's most powerful model, apparently so capable it performed zero-day attacks on every major operating system in its first outing. Is it genuinely that dangerous? Or brilliant pre-IPO marketing?

Five security questionnaires, 20 hours, and a lot of AI slop Neil spent most of the week answering overlapping, partially nonsensical security questionnaires from a prospective customer, most of them clearly generated by ChatGPT (the M-dashes are a giveaway). The cobbler's children moment: Leading AI built a KnowledgeFlow security RAG for themselves mid-episode and were answering questions live before the call ended. Twenty hours of pain, sorted in minutes.

Outcomes-based pricing — the next frontier Goldman Sachs says AI companies are moving away from per-seat licensing toward outcomes-based pricing. Kieron has already had the first conversation about it: a college currently pays £20 per student application document check to an outsourced company. KnowledgeFlow can do the same thing for about 30p of AI processing. The maths are not subtle.

Product of the week 🎵 (build to a crescendo) KnowledgeFlow now connects directly to Salesforce via live API calls. No more exporting data, no more waiting for reports, no more paying tens of thousands for bespoke dashboards. Any staff member can now ask questions of their Salesforce data in plain English and get instant answers — across multiple data tables, in real time.

A personal moment — Kieron's dad's care plan Kieron's father moved into a care home in January. The family received a 16-page care plan full of jargon and boxes nobody understood. Kieron put it into KnowledgeFlow, asked "what can I do to help?", and got five clear bullet points back. He sent it to the family WhatsApp. Everyone said thank you. That's what this is actually for.

Plus: house manuals loaded into a RAG, and the ongoing mystery of whether Neil will successfully escape the doghouse by Monday.

Two mates. A bar. Thirty years of business between them. And all they want to talk about is AI.

Pull up a stool — we'll get the beers in.  Even non-alcoholic ones for those who want them.🍺

Transcript

SPEAKER_03 0:00

Should we get on with it? Should we do it? All right, let's get started. Oh, hang on a second. Here we go. Cheers. Oh, you've got a corona as well. Yes, very good.

SPEAKER_00 0:10

Cheers. Maybe you've not got one of those non-alcohol ones again, huh? Shall we? Watching. Is it a non-alcohol one? It is. Let me tell you why.

SPEAKER_03 0:22

Let me tell you why. Um I as after as soon as this as soon as this call um is finished, I have to take Mrs. Watkins away on a uh we special weekend.

SPEAKER_00 0:34

Oh yes.

SPEAKER_03 0:35

Because I'm in the doghouse and um it's gonna cost me a lot of money to get out of the doghouse. So um I've got to drive. Well, there's a couple of things. One uh one I will share with you. We we had a little um discussion. I'm gonna say discussion. We had a discussion, um, and um there's a bit of a problem in uh it turns out Mrs. Watkins is an AI denier, and um I know, which is really awkward given um uh her husband and her son both work in an AI organization. And it's really interesting. So Ben spent ages trying to show her how to use AI. She's like, Oh yeah, I can see how, yeah, it's really yeah, but it's not for me. And uh never catch on. It'll never catch on. And and so I was like, hang on a minute, it can do these things, it can do that thing. Yes, but um, it doesn't do them in the way that I want them to, and I was like, Well, we could do that. She's like, Yes, but it um it doesn't fit my process, and I went, Oh, hang on a minute. Um, have you considered that actually it might be you that needs to change the process to fit AI, or AI is going to change the process for you, or it's gonna speed up. And if other people in the organization are using AI, then and if you haven't adapted, what are the implications for you? And um I can't tell you the exact words that came out of her mouth bar now. Anyway, I'm going away for the weekend, but it made me think quite a bit because I don't know if you remember, there was a really good book that came out from a guy called Richard Suskin, I think it was. Yeah, called How to Think About AI, and and he talked about different categories of people. And there's a category of people which is oh, AI's great, but it's not for me. And she literally said those words. It's I can see how it would help other people, was thought was, and he used the exact same language, these people will say that. And and um, and yeah, we've talked before about uh customers that we've got or people that we've talked to, or managers in organizations that we've talked to, and go, oh yes, I can see how AI would be really good, but not it's not for us, or yeah. Yeah and and um and and Helen's view as well. I've got I've got um, you know, on my spreadsheet, so all my cells are now colour coded, and it's like that's brilliant. I mean, it's not so long since we weaned you off bloody post-it notes and and coloured highlighter pens there. I mean, Staple's share price collapsed when she stopped using them.

SPEAKER_00 2:53

So I did wonder if we if we had to build an AI tool that will write post-it notes and stick them to our wall.

SPEAKER_03 3:02

What we need is a little robot that comes along and then colours it with a yellow highlight and playing and a green highlight. Uh so yeah, I'm in I'm I'm in the doghouse and uh I need to go and uh repair my marriage. So uh just sharing that with the audience. Uh so uh but yeah, that's the reason. Mrs. Watkins is an AI denier and um uh and she will not be convinced otherwise at the moment. So uh let's see, let's see whether a romantic candlelit dinner and a bottle of um uh bubbly will uh will help change your mind. But um I don't have great hopes. I it might even my sales skills aren't that good.

SPEAKER_00 3:38

Might change our mind temporarily, but it's I mean, and the changing the workflow, changing your processes, I mean that's the thing, isn't it? That's really the thing, and at the moment no one's really talking about that. I think I see McKinsey are talking about that more and more, which is great to see, but that is the world where everybody needs to think about really if you were redesigning your organization as an AI first organization from the bottom up today, how would you design it? That's the really interesting stuff because that is also I mean, yes, there's going to be savings in jobs, of course, there will be, but reality is you could make in housing the most amazing housing association, as we touched on before. You could have tenants being able to interact with an app, finding their own stuff, they do their own admin a lot of the time. That's what Amazon's clever thing was, really. I remember in the very early days of the internet, people were talking about Amazon cleverly outsourced to you writing your own shipping label and running your own payment. You said the phone up and someone would have to take it all through and write it down and create the label. You're doing Amazon have it, you doing it all for them. So, but I think so. Some of that world, but also just that you know, the ability just to kind of cut through all of the delays, and let's face it, I mean housing, education, applications for anything are all simply waiting for a bottleneck. Yeah. The theory of constraints, I think, is the uh is the right term, isn't it? But like they're just yeah, waiting to get over this bottleneck. So I think the faster people get their head around that, I think that's when you're really going to see AI making a massive difference.

SPEAKER_03 5:13

And I guess we've seen that in a few customers, haven't we? They they kind of start to wake up to it once they start to use it and they start to see the power. And um we've got a couple of customers where we've just seen that usage go go up and up, which is which is great. Even in one of the um, dare I say AI denying manager type organizations, uh, they don't know that their team are actually using it much more than they actually are. So that that is that is fascinating to me. And um, it kind of links to something uh that I uh listened to, watched um, as you know, we're both fans of uh Nate B. John's, who is our uh is our AI guru.

SPEAKER_00 5:50

Hello Nate, if you're listening.

SPEAKER_03 5:52

I bet he's not he's just listening to him rattling on. Yeah, uh he uh he did a great little session on on Substack. I'll I'll uh when I put the post out, I'll put the link in. But it was about uh the future of management. And uh there's been a lot of talk this week about uh companies flattening their hierarchy, getting rid of layers of management, and uh Mark Zuckerberg creating a virtual CEO to help manage the organization down. And um uh and and what Nate was talking about was there's there's really three jobs to um to management. The first one's routing, so getting information in and making sure it goes to the right place and then gets handled correctly. Uh the second was around sense making, so that kind of uh idea of understanding the the uh signal in the noise, both top down, bottom-up, and internally and externally. And the other and the third uh role of management being um accountability and um making sure that people do what they're supposed to do and uh the things that that need to be done are done. And I I'd not heard of it kind of put in that way before, and and and obviously, pardon it's uh spills around how much of that is AIable and the routing piece clearly can, you know, we've already done the whole kind of in messages in uh for uh for housing associations and others. So, how do we how do we get messages in? Where do they go, who deals with them, what categories, which ones need to be escalated to a human and all of that good stuff. I think we all I was thinking about the sense-making stuff because this is really interesting. This is back to the corporate knowledge stuff where you know if uh if somebody leaves your organization, their corporate knowledge just walks out with them. You there's no way of capturing that unless you've of course you've kept it kept everything in in a rag. And um, and then the final thing around accountability, that uh uh that whole piece about you know, can you outsource accountability to AI and and people saying no you can't, but interestingly enough, that whole uh AI uh uh human in the loop. Um, and we've talked before about uh one of our customers that's that says uh we want AI in the loop, we don't want the human sending anything out unless it's been through the AI to make sure that it's both sense checked, correct, and then the tone is correct. Because actually, uh one of the challenges for some organizations is that the responses that they're giving back to customer inquiries get them into more trouble than than it actually solves, because either something's gone slightly wrong or uh they've mishandled or misunderstood something, and then it just escalates into a full-blown complaint that causes more cost and an issue than it should have been. So yeah, it's been an uh uh it's been quite challenging for me on the old thought processes about organizations uh stuff this week because uh uh not least as I say, because I'm still in the doghouse, but hopefully by next week I'll be out. I'll let you know.

SPEAKER_00 8:47

Well, good luck with this weekend, then I think you're right to try and throw money at the problem. At least at least smooth some of the way, and then a lot of humility. Um, I interesting on your routing point um for there, really uh this week talking to uh one of our large housing association clients about their experience using knowledge flow and kind of seeing what we uh where we can make improvements or indeed add new tools. And one of the things that um is really interesting, the person that's sort of leading this the rollout or the training said she observed that what happens in um, and I'm sure this is very common in lots of organizations, frontline team, customer inquiry team, comms comm centre team, whatever, um, is receiving a perhaps slightly unusual problem uh for a repair, let's say, I'm just sort of example really, um, that they've got to find who is responsible for these repairs. You know, boilers, presumably they know, because they must happen a hundred times a day in uh in big housing associations. Anyway, so you've got to go by who is responsible for this thing. You you ask some colleagues, have a hunt about, find somebody, send them an email. What she said is then three days go past and you get an email back or note back saying, No, not me. They have a five-day service level agreement on responding, and now three days are gone. Um, so all of that leads to uh what RAG can do for that. And in my mind, it is one of the most simple things we could do to have uh what would be a pretty short, clear document in a RAG index that um knows the team's responsibilities, email, phone numbers, whatever you want to have for contact, yeah, and solve that problem instantly. And then, of course, with RAG, you could either just kind of share the problem or just let it decide, let it tell you. Not you know, you don't need to say who's responsible for repairing brickwork, you can just share the email and it goes, Great, I've notified the right person, let's move on. Or better still not even bother in the beginning to have the triage happen in the background and just no human involved in it until it gets to the right human, which is quite interesting.

SPEAKER_03 10:56

It's back to that bottleneck thing, isn't it? It's just it's how do you fix those bottlenecks in organizations, either either necessarily bureaucratic and and sometimes those bottlenecks are in place for a reason because you need to add a bit of friction to the process to make it work. But most more often than not, that that's really not the case. It's just this here's some stuff that needs doing. The reason it doesn't get done is because people are super busy and or they don't know how to um how to prioritize, or they get the prioritization wrong, or they don't have guidance on how to do that. So automating as much of that as possible just seems common sense, really.

SPEAKER_00 11:32

Yeah, indeed.

SPEAKER_03 11:33

But it's not for us, and it doesn't colour code, it doesn't put a colour coded cell into my spreadsheet.

SPEAKER_00 11:39

We're not done from being able to do that. Claude Co-work could do that if you were if you were uh risk of uh well no, what's the word? Uh risk hungry? What risk you have a high risk appetite, I think would be the right. You couldn't you wouldn't want Claude Co touching any of the stuff that Helen does, because who knows what it would be up to. Indeed, we'd lose all of our money. And on that, I mean interesting, two conversations this week I've had about trust. Um uh and I think that's really interesting to think about the future of AI Slop. Um, and I think you're gonna talk about AI Slop shortly, aren't you? But the what what cuts through, you know, we can no longer really trust a video, which is a kind of interesting world to find ourselves in now. You know, I I'm not a massive social media fan, but I, you know, I might spend 10 minutes on Instagram um and have a flick through, and you see these videos, you think, wow, that's amazing. And then you're suddenly thinking, hang on a minute, but it's not obvious, hang on a minute. There's a lot the theme I well, the at least the theme I've seen a lot on Instagram is kind of animal rescue things where they kind of someone personally rescues a duck and it lives with them and grows up with them, and then they release it one day or they don't, or whatever. Um, and uh and it turns out most of those are just made with one of the AI video tools, and you're like, oh, that's really disappointing. But if you in a world where you can't trust video, which is for me was one of the last bits, but you know, Peter Speaks a Thousand Words, quite difficult to kind of uh to fake it too much, to harder than text. So now that's gone. So what can you trust? And I think the answer is brand names. So, you know, as you know, we um the King's Fund, who we do some work with, um, you know, they're well respected in the sector. I think is here is their opportunity now to cut through what you know, you can be on ChatGPT and find out the latest research on hospital management and leadership and strategies. But do you believe it? Do you know if it's true? If it came from the King's Fund, probably you do. And that's uh I think is really interesting for for all organizations to think about is what what if if trust does become important, then what's their position in that and how do they curate a data set ultimately so that they uh own a trusted data set that yeah, you choose to use with AI or not, but just this is our set.

SPEAKER_03 14:00

That trust thing is um another one of Ned's big themes. Um is another plug for Ned B. John's uh uh highly recommend him. He's Fab.

SPEAKER_00 14:10

Have you got a referral link or something?

SPEAKER_03 14:12

I haven't. I know I'll I'll do something, but no, I um and I'm not on commission, just to be clear. Uh uh, but he talks about trust and judgment, you know, in in the in the current age, he talks a lot about those things, uh, and trust. Um so companies like Stripe, for example, um going from strength to strength because they are a trusted organization. So you know, if you're giving your credit card details to Stripe, then you're not giving it to whoever else. And and um uh but that whole kind of knowledge and information, how do you how do you become a trusted, a trusted intermediary um between effectively the data and the customer and so that you're a trusted source? So I think there's a there's a huge amount of work to do on that. And we talk about it a lot. Uh uh, we talked um uh uh uh this week with a an international organization, we've got a few things to say about them, they're really interesting, but uh an inter an international organization um providing services across multiple countries, and um they'd obviously spent quite a lot of time thinking about AI and how it could imp how it could work for them. And um uh but what what I found great in the conversation was just whenever I think I'm getting bored of of Rag AI, that somebody else comes up with another great set of use cases. Because I think it's you get kind of a bit blasy, don't you? It's quite simple to us, it's just it's got to be trusted, it's got to be curated, it's got to be um accurate, it can't hallucinate, it's got to produce the accurate information. As you talked about to them, you know, there's probably 10 variables that we constantly tweak and adjust through the testing period to make sure it gets right. And if something goes wrong, we we get the uh we get the rag whisperers on it in the background to uh to make sure that it has what it's quite it be. And um uh and and so they're making it trust a trusted source of information for the for the for the people and the organizations that are using their services, I think is uh is potentially a really valuable and really interesting use case. I thought it was uh it was uh a great presentation that uh that you did, and I uh I said absolutely nothing all the way through. So uh thanks for making my life easier. Couldn't get a word in edge words. And there's another thing, and let me show you this thing on circle. Let me show you this thing.

SPEAKER_00 16:36

It's so tricky. I've done a lot like that recently, and I kind of very conscious on blasting people with information, and it's so I don't know if it's sensible or not. Ideally, I would have a really well measured and managed way through it. I don't, I'm making it up as I go each time and sort of trying to read the room a little bit and sort of focus on the things I think they're gonna be interested in. But I don't know. I kind of part of me is like, well, it's a good way because they'll all take two or three or four things from it, and they might be different, two or three or four things. Um, so it's helpful for that. And then the other part of me is like, I'm just giving them so much they can't, they don't know what to do next. Yeah, so it's it's very yeah, I struggle over the sales of this stuff. It's very different, you know. Consultants sell it, selling consultancy, you know, ultimately you're kind of selling a recommendation, a bit of a kind of model or theory, methodology or something that to solve a problem. Whereas maybe maybe there is a sort of learning from that in trying to pitch this. Whereas this it tends to be more a kind of look what's possible, let me show off what our platform can do, and um yeah, hopefully some of it sticks. But I think they're interesting. Go on, sorry. No, after you're gonna do that. I was gonna say I think they're interesting that organization because what they they sort of so they're an international funded, they're funded by sort of um uh uh charity money. I was trying to think the right word for that, but I can't find can't can't think of it. But they're funded to then go and do you know various programs in the world and they might sort of go to sort of four or five kind of countries that are s are struggling and do something in teacher education or you know, all kinds of different areas. And the idea that potentially that came with knowledge flow, an AI platform, that they then are in control of the data in the background. That's I've nearly said their name, that's their data in the background that is trusted and driving the right outcomes, helping people to to sort of you know do the right thing by whatever their program ambitions and aims are. I think that's really interesting. And the learning that you could run across the group, you know, if they've got 30 of these things at a time in different, you know, schools, universities, different countries or whatever, you could have them all on the one platform with the right tools built within Knowledge Flow to do whatever those uh need to happen with the right data. Really interesting.

SPEAKER_03 18:54

And you could add in things like the uh sentiment analysis, so you can see when people are frustrated, you can see when they're when they're what they're like and all of that stuff. Uh and the safeguarding piece, of course, which we've we've talked about before, which I think is just super important and um uh not not not made enough of. Well, I was just gonna say it was in the sort of the classical um uh software sales theory literature, it all talks about you know you shouldn't say anything until you've understood their questions and their their problems and you understand how uh you know what they're thinking about these problems, so that you can then adapt your your tool to to them. It's very different with AI, I think, just because it's just so many. It's a bit like the conversation we had with these guys. We were on for what 45 minutes, and at least half of that was then sort of and have you thought about doing this, and and now we could do that, and we could do the other. So it's really it's a it's a much more diff, it's much more kind of organic than than kind of a structured sales um process, I would suggest. So yeah, I uh anyway, I'm not sure you can stick to the script anyway. You you're always off whatever the blue list of talking. Indeed.

SPEAKER_00 20:07

I don't know what I'm gonna say next most of the time. So it's the thought of actually trying to and I'll I'll write notes down before and then completely ignore them. So it's like I just I really just have to just go and show hopefully my passion. And um, and as you know, I am happy th happiest arguably thinking on my feet because I don't do I don't really do a lot of the work outside of that.

SPEAKER_02 20:28

So you need that adrenaline in Russia. Um what are we gonna do now? I do, yeah.

SPEAKER_00 20:34

Yeah, yeah. Shell, I'll tell you about um mythos. Go on. Can we talk about that now? Yeah, let's talk about mythos. So um uh Anthropic's latest model, Mythos, which has been in the news quite prolifically as the the model that's so powerful you're not allowed to have it. And um, I've been reading some stuff around that. That um some scathing remarks saying what a lot of nonsense, they are just lining up for an IPO. Um, and it's really interesting to know. I mean, who knows which side of that maybe anthropic, probably know, but what amazing marketing if it is about the IPO and you're just basically saying no, we've got this amazing model and you're not allowed it, because as you know, there is nothing quite as much as being told you're not allowed something. Like, what is it? Show it to me. Come and have a look.

SPEAKER_03 21:26

Yeah, well, the first thing most people most people want it so they get right. The first thing we're gonna do is we're gonna see if we can break into our own systems, and then the second thing we're gonna do is try and break into everybody else's systems. Actually, and some people do it the other way around, won't they?

SPEAKER_00 21:37

So indeed, but and that's the so apparently for for our audience if they haven't followed the news, apparently um Mythos um will it did in this very first outing do a zero day attack. So that means effectively zero day attacks is a is a serious attack that you have zero days to fix, is where it comes from, apparently. Um So it's you now have to do you've it's already happened, you've got to fix it immediately. And it managed to do that in all of the major operating systems. So that's Apple's, that's Windows, that was Google's, everybody's just managed to do that in minutes. So yeah, if that is true, that is clearly quite concerning.

SPEAKER_03 22:20

But even if it's not, it's very good for the uh marketing, isn't it? I mean, I heard uh I I was at lunchtime, I was I just had the radio on while I was uh making my luncheon uh uh there was something on about the CEO of Barclay saying we need to take account of this is our new world and we need to be thinking about data security. He's absolutely right there, but it's all because of the Mythos thing. So yeah, brilliant, brilliant piece of marketing. If I if only I was as that good at marketing, then uh I wouldn't be drinking uh non-alcoholic beer and having to uh to throw money at my other problems.

SPEAKER_00 22:53

You'd have your driver coming to collect you. Correct, that's right. Very good. Now you've been dealing with um a lot of security questions this week.

SPEAKER_03 23:04

Yeah, speaking of security, yeah, it's a good dad. Beautiful segment. It's uh it's like we planned it here, which clearly we didn't. So it's being obvious. Yes, I have and it's kind of linked to the trust and the AI slot pieces, really, isn't it? It's just like we've got a customer who um basically they want to use knowledge flow, they're very keen to get going. Uh, interestingly enough, uh, they went through multiple demos. We didn't think they were going to go ahead, they chose two, which is great. But um we've had um five questionnaires from three different people in their organization. Three of those questionnaires are related to data security. Lots of the questions overlap. Lots of the questions are clearly AI slot generated from Chat GPT, and you can tell just by a tour of the question and B the uh spelling and the usual um M dashes things that the people just leave in. So they've just they've just gone on to Chat GPT or or other uh LLMs are available and gone, right? I want to make sure that I've got my backside covered with these, you know, what do I need to ask?

SPEAKER_00 24:15

How do I sync this company with that?

SPEAKER_03 24:19

How do I how do I really depress their team? And and seriously, we've we probably have so there's been at least five sets of eyes in it from our organization. We've probably spent 20 hours responding to these questions. And just as an example, in three of the questionnaires, it says, What's your data privacy policy? Uh summarize it and then uh include it, uh include a copy um uh in the in the in the pack that you you send back. And then there's another questionnaire which is specifically around data privacy. It's asking, it's just asking a bunch of questions, which and some of them don't make sense to me. So uh for example, one of them says, uh, what's your data continuity, uh your business continuity, the disaster recovery plan, uh if if your system goes down? And and I cheekily responded saying, We're building this in your Microsoft tenant, it's your disaster recovery plan that you should be looking at, not ours, because once this is up and running, it's over to you. It's you guys who are gonna be you using it and managing it. So um, yeah, just uh that whole kind of uh challenge, I think, that we've got with customers who don't really understand it. And uh it I'm it would have been if I'd have understood how much work it was gonna be at the start, what I might have done was just contact them and say, Can we have a conversation about actually? I know you've got a process, I know you're a bit like Mrs. Watkins, you've got to fill in the bloody form, and it's got to be in grain in this section, it's got to be in red in this section, and you know, but let me just help me understand why you want to know these things because these things have absolutely nothing to do with um uh with the service that we're gonna provide to you. So uh so help help me understand why why you want to know this. So uh I've done it, uh, we'll get it. I'll I'll send the email very shortly with the 20 documents that have been pulled together. Um but yeah, it's just really frustrating. And and um uh next time I think I would look at it and say, please can we have a conversation about this? Because this just doesn't make sense to me. So uh uh I get it that they've got a job to do and they're right to check uh our bona fidees and you know, happy to send them our ISO certificates and our insurance certificates and everything else, but yeah, cracking. It was a it was a painful uh uh it's interesting, isn't it?

SPEAKER_00 26:52

That AI slop creating real work. It's not and you know, there's there's a lot of kind of reading of junk now, isn't there? And sort of plowing through LinkedIn trying to find the useful stuff that's not just a AI post. Um, but yeah, when it's down to this kind of stuff, which is just frustrating and answering questions that are not relevant, and ultimately because of the other side being in a place where they're not knowledgeable enough about what they're doing or what they're buying, and therefore just making it out in belt and braces, get Chat GPT to write it for you. Interesting. But what we should be doing, of course, I believe there's a flake phrase of cobbless children here, isn't there? Because we have the ability to have, as you have said, many times quite with quite some force this week, I think. It's got more forceful as the week's gone on. I think so. I mean, how many internal rags assistants do we have? We must have 30 or 40 ourselves, all built for various things, to help us do our sort of prompt analysis or whatever bid response or whatever. But yeah, we definitely need a rag on just security.

SPEAKER_03 27:51

We've got one. I've had a message this afternoon to say it's it's up and running. Is it?

SPEAKER_00 27:55

Is it in our knowledge flow?

SPEAKER_03 27:57

Uh I haven't seen. I'll be I was too busy responding to the bloody uh other documents. I'm gonna check that out with it.

SPEAKER_00 28:04

I've got our knowledge flow up here at the moment. So I'll uh LAI security, yes. Yeah, is our policy on data uh retention. Let's see what it makes of that. Yeah, there we go. Yeah, no, it's good. It's we've got three ref four sources cited in the referencing. Only for the necessary period. It's deleted in line with our retention policies. Yeah, okay. Oh nice. Good. There you go, you're sorted now. You've been outsourced. You know what?

SPEAKER_03 28:38

I'd be delighted if I was. It's that kind of nonsense that I should really be hoping. So no, if that's good, if the team have already done that, then uh to be fair, I only gave them that um uh I'm gonna say polite instruction at about four o'clock yesterday afternoon. So uh if it's up and running and they've done and got it sorted, then uh good for them.

SPEAKER_00 28:57

So yeah, that's very cool. Good stuff. Um I've got a couple of other bits and uh and products of the week, of course. So uh if our listener can start to think about the uh think about their jingle that they're gonna play, I'm gonna first talk about something else to give you a proper time to consider.

SPEAKER_03 29:14

I was I went too early.

SPEAKER_00 29:16

I think you'll need a fairly dramatic jingle this week. I think something that's quite impactful, sort of building to a crescendo, perhaps. So that's that gives you a clue. Um, so I was reading an article um about Goldman Sachs saying AI companies are now moving more to pricing, our ever-end our problem on pricing, but pricing based on outcomes or based on usage. So usage is has been around for quite a bit. Most of the AI companies have been very similar to software companies, as you know, going per, as you call it, belly buttons, uh, but per seat licensing. Um, that isn't really working very well for lots of people. I saw an interesting piece on recruitment that said you should be recruiting a workflow, not a person, in the future of AI, which is sort of talking about, you know, if you recruit someone for marketing, you might recruit a marketer and you say these are all the competencies and responsibilities I've got for them. Actually, you want to recruit those responsibilities, don't you? That's actually what you want to happen. So with AI, you potentially could think of it more in that regard. But yeah, outcomes pricing. And I think that's really interesting. We had our first conversation about this with one of our college clients who have to do a whole bunch of document checking uh as part of applications, student applications, and they um currently, if the documents aren't already clear and sorted, they pay an outsourced company 20 pounds per one for have them contact that student, ask for something up to date or whatever, and then confirm or not that they have the right documentation. I think we built a product that uh that Donald put together that can do all of that. It effectively checks whatever JPEG or PNG file you've uploaded against whatever you've written in your application. So it checks that things are in date, it can do a little bit of is this actually a real one versus a fake one, but that's quite difficult because you kind of don't know all of the different certificates you might be looking at internationally. Um but it could do all of that and it does it instantly. So literally the minute you put your application in, you press send of your application, it will have checked it and confirmed or not whether you need more evidence. I was suggesting that we could, if they didn't want to pay us to create that solution for the college, they could pay on a per usage, and we could do it for much cheaper. It probably cost about 30p of AI processing, so put a bit of a margin on that. And there you go, not 20 quid, but 19 pounds fifty 22 quid.

SPEAKER_01 31:51

Yeah, that's fair. But it's done by AI, so you'll be able to say that.

SPEAKER_00 31:56

So that's interesting, and then um uh uh product of the week is bla bla bla jingle roll, uh Salesforce APIs into knowledge flow. So we have continually talked about uh knowledge flow being agnostic to any system you're on, so sort of vendor lock-in goes away in that you don't have to get you know you don't have to buy their AI tools, you don't have to get even further embedded with any platform. Um, and this week Donald has built the API catalogue for Salesforce, and we are right now live with a client testing it, and they're their challenge, we touched on it before the 170,000 rows of data problem which we talked about on here. Um, but they have that you know in different ways all the time. They've got many data tables trying to get simple answers for most staff in their organization basically isn't possible because the ones there's a handful that know how to get in and write a little dashboard or a report or whatever, um, and to get them to do your thing, mostly your thing isn't worth it because it's a one-off thing that right now it'd be really helpful to know. So Knowledge Flow will be doing that for them, that it can basically dial in on a live API call so it doesn't need to sort of hold all of their data, it literally will fire an API call in an appropriate one to the right data table or tables to get an answer on any question that they've got at any time of the day. That's brilliant.

SPEAKER_03 33:21

I um a organization that I used to work for, um, they used Salesforce and I and it's quite a while ago now, but I seem to remember the license as being something ridiculous, like 80 quid a month or something. And I think that's it. And and then um because no one could because the the back end is so um because the reporting function is so difficult, getting those cross-table reports out is really hard. Um and they paid quite a lot of money, uh tens of thousands of pounds, to an organization to write a bunch of bespoke processes and reports for them. And uh, but i i in this scenario you can imagine there's going to be some challenges because as soon as uh as soon as people like Salesforce clock on to the fact that people are using API calls to uh manage uh manage their data in a different way, they're gonna be going, oh no, we're not getting the belly button payment, we're gonna want some API call payments. But in the meantime, it means that managers can just instead of and decision makers, instead of having to wait for a report or get somebody else to do it, instant access to your information is just brilliant.

SPEAKER_00 34:32

And that whole discipline of natural language querying, which is what it's termed in AI, I think is I mean, if you just think about any time you've ever had a report from somebody, anytime you any any single report that is a kind of moment in time, here's how things are today, sales, invoice payments, whatever it might be. And you get the very first thing you always do is go, well, why is that like that? And and then you go back to the analyst if you're in a big organization, has to then redo another one. And if you're the CEO, you might get that the same day. If you're anyone else, you're probably never going to get it. Whereas with natural language querying, you can say, Why is that like that? And it will have a crack, and if the data's there, it will give you an answer. Sometimes that's not there. So that's really cool. Um, uh, so well done to Donald and Ibby for uh getting that one uh over the line. Excited to get that going. And I just wanted to share a personal side of knowledge flow. So my father has gone into care um in the beginning of January, he's now in a care home. Um, and um he recently had his uh an assessment and was sent the 16-page care plan, uh, which I looked at and my mum looked at, and my sisters looked at, and we all kind of went, Okay, thank you. What do we do with this? And I and it's complicated and lots of weird language that you don't really understand, and not very well laid out, and lots of different boxes. Anyway, what did I do? Shoved it into Knowledge Flow. And I said, So I laid it in. I mean it was all scans, it was all so it was I had to use OCR in knowledge flow to get it to do it. Um, and then I said to it, um, this is my father's care plan. What can I do to help? And it gave like a five bullet point, here's how you can help. Um, really clear. I sent it, I sent it around our family chat yesterday, and everyone came back going, That's brilliant, thank you so much. It's exactly what we wanted because you're all scratching your head turning this thing almost as like I got it the right way up. I just don't understand any of it. So that was a really good yeah, good use of it. And um I think there's an interesting other use case which um I've done in notebook LM because it's all public data, so why not? Which is to load um house manuals. So I've got like I downloaded like the manual for the microwave, the cooker, the washing machine, the dishwasher, and various other things, and they're all in this one. So if anything in my house, if you say, Okay, what the bulb's gone, what do I what do I need to buy? It would be like, oh that's yeah, this is the B3172 bulb for your oven that you need, and here's how you repair it. And it's like so it's all like in one immediate source of yeah, it's just real basic, but and you can uh also create an infographic, which is quite weird, but it would do an infographic of your house with a quite peculiar. Anyway, there you go.

SPEAKER_03 37:14

Is the biggest is the uh is the biggest um uh electrical item in your house a beer fridge?

SPEAKER_00 37:22

It is uh sadly not. No, it is my uh no, I'm not gonna tell you what it is. I got a frame TV though, which I would uh commend to our audience. It is one that looks like a picture frame when when you turn it off, it puts art up and looks exactly like and no one believes it's a TV until you actually turn it on to a TV. It's a fine item. I'm very happy with it. Excellent, excellent, very good. But it's overly clever and got AI this and that in it, which is all made up AI. It's one of those, it's uh it's probably got a bit of machine learning running there somewhere. It's like claims to have AI screen stuff and AI sound, don't you? Like that's that I don't think so.

SPEAKER_03 38:00

Yeah, that that just sounds like uh AI marketing nonsense. I was gonna swear again. I'm trying to stop swearing on this podcast.

unknown 38:10

Right.

SPEAKER_03 38:10

Trying to clean it up a bit, trying to clean it up a bit after the complaints. Yeah, that's right.

SPEAKER_02 38:16

That's right. Do feel free to complain, by the way. We'd love any feedback. That's right. You can you can twine as much as you like. I'm not sure I can change that much, but yeah. No, indeed. Feel free to complain as much as you like.

SPEAKER_00 38:29

Indeed.

SPEAKER_02 38:30

Right, shall we? Have you got anything else, fella? Have you got anything else before I uh go and um and dig myself out with a poo-poo?

SPEAKER_00 38:36

I think you should get yourself get the car, pack the car, get the dog in the back of the car, and uh head off to the doghouse for the weekend. And um and all that remains is to wish our audience a a very good night. Yes, sleep well.

SPEAKER_01 38:52

Sleep well tonight.

SPEAKER_02 38:54

See you fell, have a good week.

SPEAKER_00 38:56

Have a great weekend. See ya.