KOPYA NG SALIN 🔗
Anne: So, hello and welcome to this week's episode of Asynchronous and Unreliable, a weekly podcast where we discuss the most interesting ideas and concepts in tech. I'm your host, Anne Curry, author of Building Green Software from O'Reilly and the science fiction Panopticon series. And today I'm going to do something a little bit different. I mentioned before that my regular guest, John Berger, is also happens to be my husband and is generally cached locally and today for the first time we're actually going to do a face-to-face episode so you'll have to forgive us for for struggling a little bit with this because we're neither of us actual TV presenters. So welcome John welcome back to the show. Thanks always good to be back. So I've got two two motives for having you on today. One is that I have I've spoken a few times recently about I've had guests on, particularly Yan Cheng Chen, talking on about feedback loops. And I know that that is a big that's something that you're really interested in. You wrote a lot about it in your book, what happens when you're not in the room. And the other thing I want you to talk about today is the the e there's the Kindle version of your book and when it's coming out. So the the physical version. viewers will be able to see and I'm holding up the visc physical version. that the physical version came out a couple of couple of about a month ago now, two months ago. Yeah, can I be living off September? so the Kindle version is gonna come out on the first of December. Absolutely. Which is good. I'm looking forward to that. And there will be an introductory price. so it'll also be cheap, which is nice. Which makes sense, doesn't it? Because a It's always it's always slightly galling when the Kindle version of a book is the same price as a physical version of a book. And it always feels like it should be a lot cheaper because it's more expensive, But so so we're gonna talk a little bit about what happens when you're not in the room. But what I really want to be talking about is feedback loops. So Yan Qing was trying, was working on trying to to make her AI chatbots understand and implement feedback loops. And she and in the the episode that she recorded with me, she had not yet managed to make that work. But but she's already told me she has now been been able to make it work. So with with the more modern versions of the AI. So I'll I'll get her back on to talk about that. But it did occur to me while I was talking to her that I really needed to get her to explain what a feedback loop was and why it was a good idea, where did it come from and why it was a useful management tool when she was trying to train AIs to be better at management. So but I didn't ask her But I know I can ask you because you wrote all about it in your book. So tell me, what is a feedback book feedback loop? you have something that you are trying to improve, or something that you're trying to stabilize, where you want to be able to take you want
Load full transcript
to be able to continually look at the output of what you're achieving and say. That's right, that's not right. How can I make it better? So you have an input, the the loop starts, something something kicks off. You were you were doing some work or you were doing something mechanical, you have the output, and if you don't have a loop, input goes to output and that's that. Yeah, you never look at it again. But if you want a loop, you look at that output and go, ooh, well, is that what I wanted? And that's that you that that's normally called the signal. It's like What is it that I'm seeing in this output that is different from ideal, for example, or can be improved? You take that signal, you turn it into feedback, you feed that back into the process at the start of the loop, and then you change what you do so that you are more successful the next time you go through the loop, and then you just keep going through the loop. So, originally these loops came from, or these were really were first described in the context of mechanical governors. So steam engines, for example, had these governors. I don't know whether I can describe this, I'll give it a go, where you would have these two steel balls, which would spin round. You'll see these in in pictures of early machines. And they would spin round and they would be attached in a in in a diamond with the steel balls at the at the the left and right side of the diamond whizzing round. And as the machine went faster and faster, centrifugal force would push out those balls so that the diamond would squash down and that would Drive a piston or something like that, or a lever, which slowed the machine down. And then as it slowed down, the balls would come in and such that the diamond would elongate more vertically. Yeah. And that's feedback. So that the feedback from the speed of which the thing is going round controls the speed at which. Yeah. The engine is running. But which is a really good example and I and I like it. That that is an example of something where a feedback loop has already been automated into your system. But with management, what you're talking about in your books is not about a situation where there's already a feedback loop that's been automated into your system. It's about for for most companies you don't have that yet. The whole point is to get to and I don't think that Yan Ching is so much wisdom. The more automatic the better, and what Yan Chen wants is fully automatic. but but loops are used in in all sorts of ways and and they are used automatically in systems as well. You know, if have a system which is automatically scaling up and scaling down based on workload, then you've got something monitoring the workload and work monitoring the load on the system, which says, well, wait a minute, I'm a bit overloaded, spin up another machine, spin up another VM, or I'm now actually pretty flat. That's a loop. Yeah. So in the tech industry, autoscaling effectively. But isn't the the interesting thing about the tech industry? So in some ways we've gone off on one because all of these very detailed examples of feedback loops Are there's loads of counterexamples as well. They're not they're not great examples because so, for example, auto-scaling, I talk about that a lot in in Building Green Software. and one of my other guests on the podcast talks about host talks about it quite a lot as well, Holly Cummins. One of the the one of the problems with autoscaling is it's not a very good feed, it's not a feedback loop. It's only one way. people tend to scale up, but they never scale down because scaling down is much it's it's much safer to over provision than to under provision. So people do this scaling, the automated part of the feedback of this needs to be bigger, but they don't complete the loop with the and now this can be smaller again. So it I I would say we've we've leaped ahead a little bit here, I think, which is fine. We were allowed to leap ahead. Well well I mean what you're saying is it's possible to implement loops badly. And it is. And that's an example. If you you only ever get move in one direction, and you're never looking at that signal which says, I've got vastly more deployed resources than I need. If you're not looking at that, well that's a that's either you're missing another loop, like maybe there's another loop that needs to run, or you've poorly implemented that loop. And it is very easy to poorly implement loops, and you probably will get it wrong when you first do it. Yeah. One of the things that you want to do. Is have a loop to study your loops. Yeah. Now here again we're gap galloping massively ahead because I'm gonna go roll it all the way back a little bit and say so a loop is like you do something, you look to see whether it works or not, you make a change and then you do it again and see whether it got better or worse. Fundamentally, is that the kind of thing that a loop is? there are actually a couple of different types of loop that I talk about in the book. So that's a that is a classic feedback. Exactly that. and I do have I do have quite a few loops and quite a lot of bits in the book about loops because I I do like loops. When when I first told people I was writing a book on management, several of them said, is it about loops? It's like it is about loops, but it's not only about loops, it's about other things too. So in the in the context of management, a lot of people will have heard of the UDA loop or the PDSA loop, which are Which are the PDSA loop clearly is a feedback loop, so that's plan, do, study, and act. So you you plan to do something, you then do that thing, and if you if you didn't go further, yeah, then then there's no loop. But you study you look at that thing that you've created and say, is that what I wanted? Is it for example? So these were first created by Edward Deming as a quality as part of sort of measuring and improving quality. So he'd look at what he'd just done.
Okay, is that is that car up to quality? Is that thing that my machine has just built at the quality level I want? Let's do some testing of that, preferably automatically. And if it isn't, you find that in your in in your your study and then you feed back and and an act or an adjust such that the next time you go around the loop it is of a higher quality. So so Deming was a basically an early time and motiony type you know organizational study manufacturing studying pioneer and he was he was he did a lot of his work during the first world war. Second World War Second sorry, Second World War. and it was about how to automate get get your get your tanks and your aeroplanes out faster and to a higher quality. Yeah, but initially, well actually initially most of his research really was done during the war. Most of his application of it, the point where he really became famous and and people started saying, well, wait a minute, this guy seems to know what he's talking about, was really immediately after the Second World War, particularly as part of rebuilding Japan and getting their industry on track. And that's where he really became famous. Yeah. And that that led directly to the total production system. Yeah, yeah yeah, which of course the famous Toyota production system, just in time, all all of the all of that management thinking that really massive that totally revolutionized the way that things are produced at scale. Yeah. But but it needed his loop to be able to control the quality such that you could operate at scale and change things without breaking. And that was the that was his so it was his PDSA loop. Plan do study study act acts or adjust. Some people say adjust. So say fundamentally John, we so it talks a load about PDSA loops, but they are not the most famous loop at the moment. So in the tech industry at least, and I don't know about whether that's that's this is a more widely held position. The the famous one is not Demming's PDSA loops, it's Colonel Boyd's ODA loops. So tell tell us a little bit about UDA loops and how are they the same and how are they different? You cover both of them in your book. Yeah, so Colonel Boyd was a US Air Force fighter pilot in Vietnam War. And coming out of that, he came up with UDA loop, which probably is the most famous loop. And UDA stands for observe, orient, decide, and act. So you start by observing what's going on, you then orient yourself. So observe is the is the objective, dispassionate, this is what I see. Orion is your subjective interpretation of that. Okay, in my context, what does that mean? This is what I see, what do I mean? I really like the difference between those two in in UTA. That's why I personally love UDA. the diff you you separating those things out. Then you decide, okay, based on what I'm seeing, what I think about it, this is what I want to do. then you act and then you see what what impact that's had. And he that the idea when he created it was that if you are in a dog fight, then you want to go around that loop as as quickly as possible. And the pilot that goes round that loop fastest wins the dog fight because they're making the right decision in that context and they're doing it very quickly. I have No idea whether that's actually useful when you are in a dogfight, but I think it's incredibly valuable in a business context and it's certainly used much more widely than the tech industry, which is probably why it's it is quite famous. And and for me, the biggest value is the difference between that observe and orient, because it means that we you can go through that loop with other people. Because generally speaking, we're not in dog fights when we're going through that loop. We're not under as much time pressure. So being able to separate out that observe from the orient means that you can go through that loop with the rest of your team, with other people who care about getting the better result. And if you all come up with different orange, so you all see the same thing, you will agree that yes, that is what we observed, but different people have different interpretations of what that means. That's super valuable in terms of fleshing out a wedding. Why do we see different things? Why, why, why are we interpreting this differently? And using that as a tool of alignment can be really helpful. Again, then you then you decide, then you are separating those things out rather than just going, let's do a thing, let's do a thing, let's do a thing. Is can be very powerful. it's almost not it's not this is not entirely true, it's almost like one of the values of doing this in business is the exact opposite of what he came up with. He came up with a loop that you're supposed to go around as quickly as is humanly possible, but the value is almost that it slows you down instead of just going, let's do this, let's do this, let's do this, and it worked, let's do this, let's do this. You go, okay, what have we actually seen? What do we think happened? What are we going to do about it? Okay, let's do it. And then go around that with different people being able to say, wait a minute, as you go around the different phases of that loop. that is very interesting. That is very interesting. So so we've now talked about PDSA loops and we talked about UDA loops. at the beginning you were talking about governors on which was an example of of feedback or physical feedback on that completely re again completely revelate. Revolutionized another another manufacturing industry, which was steam engines. That was a long time before Deming. So you had you had governors, and then you know, 150 years later, whatever, you had PDSA from Deming. It was all about well, how do we do better at producing planes and tanks that were guns that work? and and his His approach there was saying, Well look what we had been doing and a lot a lot of with with loops you go, yeah no, it's a no brainer. Of course you'd look to see what the results of your of your
systems were and then and then make a decision about how to make whether that was good enough or how to make them better. But it wasn't obvious, wasn't it? Well that that wasn't really the way people thought about manufacturing back then. you know, going back to to to Ford and really probably the most famous person in in terms of revolutionising the way large-scale manufacturing was done with a car, was you'd put an enormous amount of effort into setting up your production line and you'd really, really try to ensure up front, before any cars rolled off the line, That the production line would create the quality that you wanted. And then at some point later, you'd create a new production line, probably for a new model of car, and it would be different. Whereas what Deming was trying to do was to look at Each thing that rolled off the production line in in an ideal word, you're not, you're not actually, he wasn't able to look at each thing because you didn't have that level of automation available to you when you're building when you're building physical things. So he'd look at every thousand item that that came off the line, run it through the the the tests to say, is it within tolerances? And if it isn't, then you do something about that. And a whole bunch of other interesting things come out of that, your hand-on cords and. Usage production systems are all great. So you've got that ability to check as you're going along: is the machine doing what you want to do? Can we improve it? Can we can we create better tolerances? And the whole the the attitude change that comes from everyone in that factory who's thinking, well, we're not just turning the handle, it's not just the machine's been set up, all of the buffins have done their job, and now we just have to keep making sure that we feed things in at this end and take them out the other end. It's Job everyone in the factory to maintain and improve that quality continually throughout the manufacturing life cycle of that machine. Yeah. And if you can improve the machine as you go, people's lives get even better and the output gets better. I and one of the really interesting things is how hard a cultural shift that is. I mean. So you and I started in tech industry 30 years ago. And at that time it was j it was the pre-deming way. We know we did waterfall, we did loads and loads of thinking up front. We just we just all you know all our brains were going like crazy, planning and planning planning. I I kinda like that. I really like that. it's even now, it's quite recent that people have moved into a more of a and it I would say it's not even agile. It's it's not, it's it's that kind of iterative, fast flow kind of a way of approaching things. But it's it's it's iter iteration. Doing lots of small things, iterating, CICD is the is the key is the key tool really here. you need you need something like that. It's you know, it's all about testing. For for Deming it was it was about quality. How do you how
do you get that up? CI C D is a key tool there. It it's not quite the same because well it's very much not quite the same. Because in physical manufacturing of these things, what you were doing was building a machine once to build to that would go away and generate you a million of these other things. It's like great, go and build me a million cars, go and build me a million washers or whatever. You know, it's like so. It was worth putting a huge amount of effort in up front. But you could easily see why people thought that it was all about the upfront effort rather than the on And it's not that they're saying, don't bother doing any planning, don't bother putting the effort into creating it up front. It's just that don't think that's the end of it. That's one of the things I like most about all of the loops that we're going to talk about is that they don't end. Software development doesn't end until no one cares about your product anymore. Yeah. Right? The point where you ship your product, that's that's really the start of it. That's when your customers first get to go, wait a minute, that's exactly What I wanted apart from this, or it's not at all what I wanted. And you're going to spend any successful software product spends vastly longer in maintenance and support and actually engaging with customers than it does with the initial development. People spend all of their time worrying about getting the initial development to be great. But if you think about it as a loop, and and some things like you know, test-driven development, behavior-driven development, and all the Or the various DDs try to move a little bit in that direction. But but if you're thinking about what you're doing as a loop, which you're continually improving your product, that's that's very different. But you're only still only building really one product. You're not going right and now the code works, but spin up a million copies of it. I mean you're sort of doing that, but that's not the machine that's doing that. And if you are in an environment where things are changing quickly, or are actually really just changing at all, in fact, then those loops help you enormously because you don't just go, we've done a thing, finish. You go, we've done a thing, now we do another thing, now we do another thing, now we do another thing. And you've got the tooling in place, you've got your CICD, but you've got your attitude of your team in place, which is we're constantly changing, we're constantly improving. It's not new. It's not a wait a minute, what do you mean I've got to do this? It's like the expectation is of that constant improvement. And and that's why I I've always thought loops are not just about that bit of software development, which is we we have a requirement here, we have a product that meets that requirement. But but everything else about the way you work, how are you developing your people, how are your how are your teams operating? How are they happy? Are they productive? Are they doing what they think they should be doing? How is your how what's your customer support process like? Everything that's part of a business can be improved. And a loop is a brilliant way of doing that. Yeah. I mean it's it's interesting. It's like you talk about it and you think, God, it's a no-brainer to do this. It's so much better and completely revolutionized everything. It's like, But I think it's really well worth remind constantly reminding ourselves that things that everybody wants to or wants to re revert to the single big shop delivery. You know, waterfall, everyone wants to go down the waterfall really in the end. Or something that even if they don't, they they build things that are like smaller waterfalls rather than Rather than that continuous feedback. So so so in The Gold by Eli Rock Goldratt. So Eli Goldratt was a writer in the eighties, wrote this absolute blockbusting management book. If if such a thing such thing existed then, I don't think it will ever exist again. Blockbuster management book called The The Goal, which was about introducing Toyota production system, that kind of deming, ironically, created by an American, then moved to ch to Japan. refined there and then Eli Goldrat's attempting it to bring it back to to the US. and I I enjoyed the the the the follow-up to it to the I think it was the the probably the final version of his book before he died which was somewhat in the 90s saying that the interesting thing is even in Japan people don't all follow the toys production system they want to So everybody in Japan knows all about it, knows it's great, wants to do it, but but hardly anyone actually can successfully pull it off. It's such a complete mindset shift away from like you're clever enough to think all of this up front, which all engineers think now. whether manufacturing engineers or whether they're software engineers, they all think they're clever enough to think it all through up front. And I'm just about you, both of us are terrible for that kind of thing. We we we both think we're clever enough to think it all through up front. But you have to accept that it's not really about you being clever or not. It's about s you know, there's just so many unpredictable things. I mean well, the the world is very unpredictable, which is one of the reasons why. Why it's hard, but it's actually got nothing to do with how clever you are, is the point. It's like the the fallacy is not people thought they were cleverer than they are. The fallacy is that that's even relevant, right? why why are why are the loops useful? Because they change your attitude, as I've just said, because they allow you to change. Right, even if you're smart enough to think in advance of exactly how this motor car should be built, exactly how this piece of software should be built, the idea that you can predict every future demand on that thing or change in whether it's a change in What your customers want, it's a change in what
your exec team wants, whether it's your change in what the government allows you to do, or the the the idea that your product will will and should never change after that, or that you can predict all of them, is crazy. So if they're gonna change and you're gonna create this brilliant product that you've thought really hard about and done an absolutely knock-off job of creating creating the thing in the first place, brilliant. Then a million people buy it, still brilliant. Then the world changes. And do you want someone else to come along and do a slightly better job because it's more aware of the context? Yeah. Or do you want to be able to keep moving quickly to stay ahead of all of those requirements? But it's not even just that that makes loops really powerful. One of the reasons why loops are in this book is because. Loops allow the manager to get out of the room. When you once you've created that loop and people are capable of operating it, they can improve things without you needing to be there with your giant brain telling people what to do. They can see in the moment while you're asleep, while you're on holiday, or while there are a thousand other things going on. They can see what needs to be improved. They can see, well, wait a minute. We're not achieving what good looks like. How do we do that? Or we are achieving what good looks like. How do we achieve what great looks like? And running through those loops allows them to do that, irrespective of what you're doing. One of the reasons for that is because you probably you you have to have spent some time at that point having a discussion about what does good look like. And just getting that out in the open and that being a shared Vision is really the key to allowing people to then operate on their own and go fast and use their own expertise and awareness and understanding of the situation, irrespective of whether you have that or not. Yeah, yeah, yeah. Yeah, I mean this also, of course, reminds me why Yan Ching is really keen on defining loops for AI agents because. she wants to go to sleep and have the AI agent loop around and iterate on the towards a goal, a defined goal, while she's not asleep or she's w or for let's face it, operating on a completely different timescale to the time scale of her AI agents. Absolutely. And she wants them to be able to improve things that themselves autonomously, like you would want a good employee to do. And there there was a there was a there was a common fallacy not that long ago, which we can very much laugh about now, that it was all about prompt engineering, getting AI to do the right thing was all about If you just crafted the perfect prompt, it would then go away and do that thing. Yeah. No. What does good look like? And if you can define that and allow an agent to iterate, or a team of agents, to iterate against what good looks like, then they can do that. If you've got to write the perfect prompt up front, the only way you're going to make that successful is another loop outside of that, which goes, your right prompt wasn't good enough. How about it? How about that's that's and then then it's you doing all of the work. Yeah. And your bottleneck. You've created a bottleneck, you've not actually delegated. Yeah. So this has been, I think this has been an excellent episode. We talk about quite lot of stuff. So I want you to push your book again. So here is the paperback copy of what happens when you are not in the room To viewers, to listeners, it's holding up a copy And that is available to purchase already on Amazon and if you want the Kindle version that comes out on the first of December at a special introductory price. And hopefully I will catch you again on next week's episode of Asynchronous and Unreliable. I I do I do run into quite a lot of physical people who who listen to this or watch this podcast. So and I I usually get all my feedback from them. So that my feedback loop is going to the pub or for a coffee with somebody says, I watched your podcast, and quite often they're end up being guests. so hopefully those people will let me know whether or not this is a good format or this is not a good format. Yeah. But thank you very much for listening to Asynchronous Unreliable and I will catch you again next week. Thank you very much.