In this episode of AI x DevOps, Rohit sits down with Sailesh Panchal, CTO of TSB, a UK bank with around five million customers. Sailesh has spent over three decades leading and observing transformations: from building distributed systems in the mid-90s, to CTO at Travelex, to helping Lloyds Banking Group on its journey to cloud, to building his own fintech in the cloud and leading engineering transformations across several banks.
The conversation is about why transformations fail before they start: most organizations don't actually know themselves. Sailesh walks through Transformation Intelligence, his approach to writing an organization down (its decisions, authority, jobs and values) in a structured grammar that both people and AI can read, how he used AI to mine hundreds of decisions out of existing documents, and MTDR, the open-source standard he has published for recording them. Then he asks: once AI makes execution instant, what is left for the organization to do?
⏱ Chapters:
00:00 - Introduction and Guest Background
03:48 - Rethinking Digital Transformation
12:18 - Transformation Intelligence and Organizational Context
24:13 - Mining Decisions with AI and Open Source
34:58 - Engineering Leadership in the AI Era
43:04 - Zero-Time Deployment and the Future of Organizations
Here are some of our favourite takeaways:
If you're leading a transformation and suspect the hard part isn't the technology, this one's for you.
🔗 MTDR, Sailesh's open standard for Transformation Decision Records:
https://github.com/saileshpanchal/mtdr
✍️ Sailesh's writing on digital transformation: https://saileshpanchal.github.io/digital-transformation-advisory
🎧 Listen on Spotify: https://open.spotify.com/show/70tmILfTgaEy7o9ya1sJzC
🎧 Listen on Apple Podcasts: https://podcasts.apple.com/podcast/ai-x-devops-by-facets-cloud/id1813326497
#AIxDevOps #DigitalTransformation #AIinEngineering #PlatformEngineering #DevOps #EngineeringLeadership #AgenticAI #OpenSource #CTO #FinancialServices
Engineering teams are under pressure to move faster, do more with less, and stay ahead of an increasingly complex stack. AI is becoming a key piece of that equation — not just as a tool, but as a shift in how DevOps is done.
At Facets.cloud, we’re building infrastructure orchestration for the AI era. And with AI x DevOps Podcast, we’re creating the space for honest, technical, forward-looking conversations about that shift - from early experiments to long-term visions.
This podcast is about sharing what’s real: what’s working, what’s not, and what’s next. Whether you’re building internal copilots, streamlining CI/CD with AI, or rethinking developer experience — we want to learn from your story.
Sailesh (00:01)
It's the fact that the organization doesn't know itself when it begins the transformations.
Rohit (00:05)
Oftentimes, the tech wasn't the hard part at all, right? It was changing the processes, knowing about the processes and knowing about the people, the roles, the responsibilities. The whole context is what has usually been a bigger challenge than the tech itself.
Sailesh (00:18)
We can make engineering faster, but the company won't move any quicker. So we've got to understand that transformation is a holistic activity for any organization.
Rohit (00:28)
Most organizations are just getting a Claude subscription and hoping that magically they would have bigger, better quality faster shipping and all of that.
Sailesh (00:35)
Engineers aren't adopting it. They don't want to adopt it. They don't believe it. They all played with it a year, two years, three years ago, and then said, "Oh, it hallucinates. The code isn't any good. It's, you know, it's rubbish, blah, blah, blah."
Rohit (00:48)
Have an experimental glass tube where AI writes code, provisions, ships, makes it live, maintains. Build that pipeline, then your organization is transformed.
Sailesh (00:58)
If you could deploy your entire organization instantly, what is disaster recovery? What happens to your infrastructure operations? What happens to your change approval process? What does operational resilience mean?
Rohit (01:12)
Hey everyone, welcome back to another episode of AI in DevOps podcast. I'm Rohit, your host and one of the co-founders at Facets.cloud. So on this show, we sit down with leaders from across the industry and figure out what AI actually means on how we build, ship, and operate software. It's less about hype and more about real shifts that are happening on the ground. And on that note, I'm genuinely excited about our guest today, Sailesh Panchal. He's the CTO of TSB, a bank with around 5 million customers. But that one-liner doesn't do justice to his profile because he has been in the industry for over three decades and most of it undertaking transformations, driving them as well as observing them closely. And this is something that I'm genuinely passionate about myself and trying to understand how transformations work. And Sailesh in our earlier conversations has surfaced really interesting observations and opinions of his regarding digital transformation. So today if we want to go deep on transformation itself. And I'd welcome Sailesh to the show. And thank you, Sailesh, for joining us today.
Sailesh (02:29)
Yeah, hi Rohit, pleasure to connect again online. Yeah, my background, I'm CTO of TSB Bank at the moment. I started my coding journey in the mid-90s, building distributed architectures. Ironically agile before it was as a term, was adopted back then and then we... Carried on in Travelex where I was the CTO there building global teams, doing agile development and starting the transformation journey. Lloyd's Banking Group, which is the largest UK retail bank, helped them on their journey cloud and looking at agile feature teams, value streams and those kind of constructs and then built my own fintech in cloud. So this is where the whole serious journey of understanding how cloud deployment and the challenges of building multi-Cloud solutions came through, so regulated in the UK and in the European markets and then led big engineering team transformations in a number of banks and culminating in TSB and then having done a number of mergers and acquisitions, again, fairly big transformation activities and understanding the challenges that come with that.
Rohit (03:47)
Awesome. Yeah. Yeah. And then, Sailesh, you also have your own personal interest in digital transformation. You have some work that you have put out there regarding a framework for digital transformations and how that should be undertaken as well, right?
Sailesh (04:04)
Yeah, I think the reflection I've had over the last year or so has been I've seen a lot of transformations and if we look at transformation, people tend to put it under the biggest possible banner of the whole bank or the whole organization and really what kind of organization it is. But the insight I had really is like every change in an organization, every consequential change in an organization is an actual transformation and people and organizations are doing multiple transformations at the same time. But the discipline underpinning those transformations needs to be consistent. And you really need to understand what you're doing. I Mean, your product itself, the way you're thinking about discovery, declarative infrastructure, having predefined blueprints, and then you can automatically execute and reconcile these kind of fundamentals and some of the organisational primitives that fall behind that really why I've started thinking is what is transformation? And how would an organization start thinking about transformation? As we know, the challenge isn't sometimes the transformation itself, it's the fact that the organization doesn't know itself when it begins the transformation. So when it works with partners, the first thing it says is I want to transform and then the partner has to do the excavation work to say, what am I transforming? And that can be quite painful because it's a journey of discovery sometimes for both the buyer and the seller and it makes it very difficult for organizations to get the outcome they really want.
Rohit (05:57)
Exactly. So I think that's just gospel to me because, we have been advocating for platform engineering and trying to get platform engineering principles implemented in organizations. And oftentimes the tech wasn't the hard part at all, right? It was changing the processes, winning over the people. This is where, and of course, discovery, like you mentioned rightly, about the processes and knowing about the people, the roles, responsibilities. That itself would and Decisions, the hierarchy of the decisions, why was it made? The whole context is what has usually been a bigger challenge than, the tech itself. So what does actually reimagining an org from ground up? So I know that you have built the framework From an organization perspective, if you are looking to understand yourself, do the baselining first before you undertake a transformation. So what does that mean?
Sailesh (07:02)
So fundamentally, I think there are some primitives involved, right? And my goal isn't just to have, we understand the organization, because most organizations will tell you, we understand ourselves. You go to any senior leader, and they will say, I understand it. The question is, can you develop a shared understanding of the organization? Okay, where all participants can openly share that idea. And in today's world, can you document it in such a way as an agent can read it? Okay. So at the moment, decisions are in multiple board packs. Okay. So they're written in many documents, many manuals, many policy documents. It might be your security policies. It might be lots of other places. And you're expecting to come in and understand all of those decisions of the organization, both historically, currently enacted, and so you can understand why the organization looks the way it does because it's a history of the decisions it's made. And some of those decisions are internal, so they're board level strategy decisions, some of them are policy documents that they're public. I think everyone's received RFPs where you're trying to infer what does the client want because not all the decisions are obvious. That's step number one. So you've got to understand your decisions. You've then got to understand the authority. So who can make the decision? Who is the ultimate decision maker? And I think everyone's been through that trying to really navigate an organization because most people in this organization are operating under delegated authority. Now the interesting piece for me is authority and delegated authority and how you express it are central to the whole AI model. So if an organization understands authority and understands who's operating under that authority, they've got a chance now what that means is you've then got the capability to start expressing I'll call it inheritance so what I and again this is you're getting an early insight into papers that have yet to be written it's quite a struggle trying to get that down the concept of inheritance so as a person what have you inherited and what are you? So what capabilities have you got? So there's a SFIA model that describes your skill levels at various capabilities and what jobs do you do? So people will be familiar with jobs to be done as a framework for understanding internal processes everyone operates with a job description. So if you take the most basic job description of an individual, it tells you who do you work for, what is your job, what skills do they expect you to have, all of that. Now, very interestingly, that also applies to AI and the agents. And so at that point, if you know the job to be done, then that applies to the AI agent. Now, the other revelation is that the existing estate has hundreds, thousands of applications. Do you know what job each one of those applications does in your estate? So you might know it's TIME analysis. So whether you want to terminate it, invest in it, exit it, but you may not know what work it does and how those jobs interact with the jobs around it. So that when you try and lift and shift it into cloud, you can't because it's got too many interconnections and with any transformation, you're actually changing the job. So if you're doing a like for like, that's one thing. But if you're actually trying to transform it, then that's another set of problems. And then the final primitive, I'd call it, is value. So the value is, the business case. Why do I want to do the transformation? What do we value? Operation? Some of it may not have a monetary value, so it might be operational resilience. It might be... A Reputation. So these things are generally not documented in a very clear way. So in your world, when you come in and say, oh, we're going to move you to cloud, if you don't understand the people involved, i.e. How many people have they got in infrastructure deployment today, what skills do they have, what jobs do they do, and the jobs can be go to CAB. With a signed ticket, okay, and get approval, etc. So if they don't have all of that written down, it's near impossible to automate that process and get it through governance. Because again, governance today is a human process we'll go into... The challenge today is deterministic versus probabilistic. Unfortunately, organizations are run 90% on a probabilistic model. Why? Because the biggest probabilistic engine is the human. Right? And so you might be able to say we've got many applications and they're completely deterministic. The reality is it depends on a random day who you get in a governance board as to what ideas come in their head. And then they go, oh, I remembered the security policy that I wrote three years ago that wasn't clear and you can't go ahead. Yeah. Yeah. So this is this is the challenge, right? So I think what we need to do is to think about knowing yourself so it's the whole Sun Tzu thing you know, before you start the battle you've really got to know who you are
Rohit (13:13)
Makes sense. And it's not just a documentation problem either, right? Like, for any organization, it's not just about documenting who you are what my processes are, what are the responsibilities, things today. It's that the change management, any change that goes into that, any new decisions, any change of roles, responsibilities, also needs to go through the same pipeline. Otherwise you're playing catch up again.
Sailesh (13:35)
Yeah, so the way I've framed it is there's transformation intelligence, which is the act of knowing yourself in a structured way. And I've taken this approach by, I think most people will be familiar with use cases and when use cases came out, they were random ideas of requirements and then people came up, you know, the given when then structure. And so it became sensible for people to have a structured grammar within that process. People were used to writing it down. So I call it transformation intelligence. So how can you write this stuff down in a structured grammar, which makes it comprehensible and reviewable by everyone? And more interestingly, when you do it as a structured grammar, AI can look at it. Okay. Then the next phase for me is what I call intelligent transformation. So how do you project a future state of transformation? As a delta against your current state. Okay, so if you have your current state, you can then say, oh, I want to transform. Oh, that means these applications are decommissioned, these jobs get transferred to this. Or There are 50 people involved in doing these pieces of work. They're not going to be done. Oh, the authority doesn't change. You know, the senior manager is still accountable for the revenue from this line of business. Okay, so you understand what your future state is and you know, and if the transformation results in a you know, new product, new feature line, then guess what? You've got new authority structures. Okay, and so you need to model that. Now, the challenge is most organizations are performing multiple transformations at a time. So it's not you know, everyone will say, oh, we're doing digital transformation. Not really. Okay, so you're actually doing multiple transformation simultaneously. It might be you've got an HR program going on, you've got you know, branding campaign going on, you're changing you know, your, your journey to cloud, but in fact, it's not the journey to cloud, it's like multiple application estates. And multiple stakeholders are all getting involved, but they just didn't know it was happening. And then throughout the process, the challenge is always value return. So, you know, AI has demonstrated and cloud did before it. Okay, so we all went through the cloud journey. People didn't understand why cloud was going to be useful. Okay, and how you articulate it and what jobs should be put into cloud or shouldn't. Right, so it didn't disappoint. You know, cloud is great, yeah, but you've got to use cloud for what it's meant to be great at. Okay, so if you lift and shift your technology into cloud, the economics won't work because your platform and your application hasn't changed the job it's doing. You've You've just lifted and shifted it and guess what? A third party will charge you more for hardware than if you buy it and you capitalize it on your own. But if you're not actually deploying it quicker, faster, easier, cheaper, then there's no point. In it. Yeah, so you've got to think about, are you going to take advantage of the evergreening capability of cloud? Right, and in today's world of Mythos threats and everything, all organizations are going to go, well, we need to be able to patch, upgrade, refresh underlying hardware, security protocols, API structures really quickly. So Can you do that on prem? So the value proposition of a transformation may be changing dramatically. Now it doesn't mean you can't be fast on prem, because you've got OpenShift and loads of technologies and you can do things on prem versus cloud. But then understanding the nature of that is critical, right? And DevOps has taught us the same kind of lessons where, you know, it wasn't a tooling change. Yeah, it was never a tooling change. It was an organizational redesign disguised as an engineering practice. Yeah, yeah. It was that kind of model. The deployment velocity was meaningless. If you couldn't change your product teams to understand if the organization wasn't ready for fast deployment and going into production, you know, AB testing, you know, blue-green environments, all that kind of stuff. Then all that happened is you hid behind feature flags until the organization caught up. So the feature still took six months to go live, even though you'd done 80% of the deployment in three weeks. Right, so I think, you know, we can make engineering faster, but the company won't move any quicker. So we've got to understand that transformation is a holistic activity. In the organization.
Rohit (18:44)
Right. It Makes a lot of sense. And I think you have put a lot of structure to, you know, otherwise what would be considered as probably a project management problem statement, right? Like undertaking a transformation and probably a lot of the context in people's heads and the execution gets pos just because not everybody is aware of the lineage of the decisions, the lineage of the other things that are going on. In the organization and the end goal. So I think I've seen that quite a bit. And I can also, you know, sort of draw a parallel in a miniature. So a while back, at least when pre, let's say, Claude, when work that I could get done from my laptop, was something that I could manage, I used to have a system loosely based out of, you know, David Allen's get getting things done, that sort of a framework, right? So I had an inbox, projects, tasks, you know, that sort of a thing. Then all of a sudden Claude came around. I just started, you know, spinning up sessions and doing a lot of random work here and there. And slowly I started losing the discipline. So which is when I thought I'd develop a small binary for myself to, you know, tag every session against what is it actually for? What is it, the hows of it, the decision making part of it. So I just did it from my mental piece of mind, because David Allen says that if you just write things down out of the top of your head, that itself gives you a lot of peace of mind. Like it could be a piece of paper where, you know, things down on what is what, what decision has been made, it sort of makes sense. But what I eventually found out is that now all of a sudden my different AI agents, Claude, all is using this binary. It started becoming aware of all of each other's work. It became smart to look at other people's tasks, it started looking at projects, definitions, things like that. And all of a sudden it seems to bring a lot more value than you know, just task management. And I think this is this what you're suggesting is, is more of an organizational level, not just a task, even decisions in any change log, if it goes through a central discoverable system.
Sailesh (20:55)
Yeah, I think a hundred percent. But I think what you're, the fundamental problem is context, right? So everyone knows context is the problem, right here. So you make better decisions with better context. I think everyone's gone through obsidian, second brain, right? And then suddenly discovered all those notes they've got in hundreds of documents and picked paper over time needed to be written down somewhere and you can put it into a latest phone, you know, online AI searchable tool and you personally gain massive value from that. And I've been on the same journey as everyone else, right? So for the last eight, nine years, hundreds of notes, thousands, tens of thousands, and then suddenly being able to discover that is amazingly useful. Now you can scale that, right? So the advanced So whether you're an organization of one, right, with your own personal accountability or an organization of 200,000, the problem is exactly the same. And this is a problem that organizations have battled since the big the pyramids. Okay. The understanding how to give common context and view terms of delivery has been you know, the foundations of Agile. The SAFe activity where you've got PI planning, you've got everyone involved you know, in a single set of ceremonies, people are all told what, what commitments they're making, what decisions have been made, what values the organization values by the senior leader. They've committed to doing something, they've got stuff that's uncommitted, they've got a backlog, right? And if you're in a very mature version of that organization, the business are directly involved so they can understand the trade off. So yes, we do need to spend time on the technical debt this month or this quarter. Right. We do have these production operation issues where we've got 50 people dealing with these failures. If we make this change, an engineer can fix it at a source and we can change our business process. And we can remove that. Now that level of understanding and discipline, so people always said agile has got no discipline. My view is it has way more discipline than handing someone a 50 page PRD document with a random committed date on it and saying, right, you're executing against that date. Everyone feels comfortable that project document has got a date in a year's time, but you're completely and wholly dependent on that single document having the entire context of the whole organization and the analysis. Now, the risk we all know is that no one ever had one document, right? The organization was always executing 20 of them and none of those knew about the others. So how is it conceivably possible that any one document was going to say here's the requirements I have for a project and you're going to land it on this date in a year's time in an organization whose shape I have no idea about.
Rohit (24:13)
Makes sense. So Sailesh, how do we go about addressing this problem? So how I mean, we have I think, framed the problem, we have frame what is needed, but how do we put it in practice? Like, do you have some examples of how you have implemented it or things like that? Yeah.
Sailesh (24:31)
So it's actually, so I've implemented pieces of this in TSB. I'll say I am developing it as a scientific method. Okay, so the challenge with all of these things, it's easy to come up with the hypothesis. You need to test it across both your own organization and across multiple industries for me. Okay, so it doesn't work, you know, it's got to be a fundamental. Truth Because every industry is moving to cloud, for example, right? So it's not a single challenge in a single organization. And it's remarkably easy to do in today's AI world. Once you've got a structured grammar, you can use Copilot, you can use Claude to mine existing documents and say, okay, mine all my existing documents and find all my decisions. Okay? And then it will... Create structured text and you know I've done that in my organization and you get a document and it'll come out like a policy document and it'll have 300 decisions in it and then you can sit there and review them and say Oh, I now agree with it in a very structured way that that decision is the actual decision Oh, and then you'll go right it's good it's time bound it's got a limitation It's got all of these other constraints which may or may not have been apparent in the steering pack where it was made because the steering pack is sometimes a point hot solution. Right. So in steering committees, you may, you can't sometimes have the entire context and you rely on your stakeholders understanding the vision. For their decision that they make in the steering committee, but the person capturing the minutes can't put that entirety of the context down. Okay, and so they'll say, right, we've decided to exit that function. But that's all you've got, right? You haven't got the graph. And this is the important thing that happens next, is you put everything like this into a graph, because that's where it naturally sits. And that winds up being an immutable graph, a mutual graph. And then you get the lovely word projection. Okay, so you can then start projecting from that controls, and then start projecting future states of the organization. And even better, when you get to your future state, you'll be able to look back and say, well, what did we project this was going to be? And did we achieve it? Right. So many, I think we've all been in many projects and programs where the business case that you started with isn't necessarily realized in the way you expect by the time the project finishes, not just at the minute moment it goes live. But two, three years after it's been running because the accountable exec's no longer there if the business case didn't stack up. And, you know, if it was wildly successful, someone else is taking the credit. Yeah, so it's that long-term structure. And then when you start doing things like understanding that all of your organization's procurement, is a series of decisions. And then do you understand your partnerships? So do they understand, you know, so how many people have bought applications and only use 10% or 20% of their application capability? Right, so they bought it for one thing, it's got a lot more capability, but they don't understand that capability. And then they'll go and buy another tool, or they won't work with the vendor to fully exploit. The Full capabilities of their platform because they can't understand how those capabilities would interact with their future projection. So you're constantly presumably saying, well, if you use this properly, it'll speed this bit up and you can you know, redeploy these people. But because they've got no way of computationally or mentally thinking of that, the change to their organization, they will hold back on that transformation of adopting that capability because it just sounds too complicated and they don't feel they've got the authority they need to execute it.
Facets (ad) (28:51)
Sorry. A quick pause before we continue. I want to share something we're building at Facets.cloud that ties closely to today's conversation. Your AI is a brilliant engineer. It can architect an entire system in seconds, but it has no hands, no cloud, no database, nowhere to ship. So what do you get? A beautiful front end, running on fake data. Sure, you could host it yourself. Maybe even a Postgres sitting wide open on the internet, one password away from getting hacked. Scoop gives your AI real cloud, secured the way a seasoned DevOps engineer would set it up. No cloud account to procure. You get it directly and no AI credits to buy. You build with the harness you already have, cloud code, codecs, whatever you use. But hey, don't take my word for it. Here's Laura. She's a parent and she actually built one.
Laura (ad) (29:45)
Hi, so I was drowning in my kids' school PDFs, a new one every single day in the school's app. So I just asked Claude to fix it.
Facets (ad) (29:54)
And with Scoop, Claude architected the whole thing. A secret store for her login, a scheduled job that fetches the PDFs, a private database, a blob store for the files, a front end on a CDN, and sign in. And now, Vibyor High section 2D parents are happy. And now, she can roll it out to parents in every grade. And if it ever turns into a real venture, eject to your own cloud. That's coming soon. Until then, run on the cloud we provide for you. Scoop, give your AI the world.
Rohit (30:30)
So I think again, I think the critical point here would be that historical decisions, yes, we could mine, create that catalog, we put it, yes, as you mentioned, in a graph, make it accessible. How do we make sure that the further decisions, further changes go in? Is it just an organization discipline that you need to... Is it a people problem?
Sailesh (30:53)
Well, I mean, it's the same as writing requirements in Jira, right? That's a future projection. It's not very hard. I Mean, people do this all the time, right? It's just that there's no way of finding out what the implications of that future projection are. So it could,
Rohit (31:13)
It could sit alongside your current processes where people are just working through that,
Sailesh (31:18)
I mean, yeah, I mean, this is this is not a, it isn't, Lee so, it's not a change your process. And the advantage of doing it with AI is that potentially you're doing actual minimal work. And in my experience, what it's done is when we've mined the decisions, it's actually significantly improving our decision making in the activity. And then certainly when you doing migration events, you're moving from one platform to another. If you, Once you understand the job that your current platform's doing. Then and when you move to a new platform and a new transformation, you've got a disciplined way to say, oh, actually the new platform doesn't support these capabilities. Right. Right. Yeah. And what are we going to do about it? But that level of discipline, I would suggest, is not there in many organizations where the, you know, I think the common refrain you hear is, oh, it's an application, but I need the SME for the platform because I don't know what it does. Right, I don't understand the business context. That SME knows all the decisions that were made. Is that an acceptable position for leaders to find themselves in? To discharge their accountabilities? I don't think so. 100%. So I think the framework, so in
Rohit (32:38)
The mental model that I have as leader, I mean, gets this, this process implemented where you mine whatever, whatever is your process right now. Like, it could be Jira, GitHub tickets, all of that, get that decision. So I think the framework helps you have a GitHub repo where the framework sits and it could help extract this
Sailesh (33:00)
Open source as well. Yeah, open source as well. So the goal of the exercise is people can contribute, run it. I've used the various latest techniques for agent deployment. So that so to open knowledge framework compliant repo. You'll see inside that repo all the decisions I've made are in there.
Rohit (33:26)
Yeah, it's bootstrapped.
Sailesh (33:27)
It's fully bootstrapped by itself, which again, I think is unique for a repo. The other interesting challenge is people are going to go, oh, it was written with Claude. Okay. So, you know, a lot of people will go, oh, my God, right? All of this content, text, everything inside it is all done with Claude code, right? And the conversation I have. Is it's still my authority, right? I'm still the person or whoever it is running the repo is still accountable for the decision to adopt it and make that. And you can see every decision and there are hundreds now in there, I was I had to sit down and go, okay, there are options here. This is the way I want to do it. And this is where AI and humans are going to really interact on an ongoing basis is if humans cannot make their decisions known and valued, okay, then we are going to lose, right? Because AI decisions, people just say, oh, it's hallucination, it's this. No, it's the fact you weren't accountable and you didn't hold yourself clearly there for the guard rails at every point in the journey, right? It's like, online, you just go, okay, consent, click, click, click, just let me get to the content. Yeah, you're losing that. So I think it's important where people will go engineers who are using AI. And I think it goes to the heart of AI adoption in engineering and in many other facets. I think I suggest that if you're a lead engineer, understanding how good you were as a mentor for other engineers, okay, and how good you were during the code reviews, and how diligent you were during their code reviews and how well you were at explaining why to another person. Why their code wasn't good and what a better solution was and they should have used this algorithm or they forgot about this API and you know, they didn't think about what the next release of Python was going to have in it and you haven't used the features et cetera. Those are absolutely the skills that you need when you're training or utilizing AI in an engineering context. And that's what will distinguish the great engineers. So the distinguished engineer title and all that is really about how good were you at training other engineers and bringing uplifting the skills of your people. So if you weren't very good at that, I don't think you're going to be great at AI because you didn't really embrace the fact that you as the lead engineer, your job was to enable, you know, 10 engineers, 100 engineers, 1000 engineers in your organization. And if you see agents as junior engineers, and actually junior engineers all with a PhD now in mathematics, physics, and every discipline under the world, okay, and you can't let that go and you didn't understand what your role was. Then I think you're going to struggle in the new world, because you're going to think your value is in coding, not in driving the shape of the organization forward. Got it.
Rohit (36:57)
That makes a lot of sense. So I think now that vibe coded, you know, diffs, I mean, PRs are getting merged, even in the Linux kernel. So I think it's high time that, you know, the open source communities themselves could also use this framework, log catalog their decisions and ensure that whoever is contributing is aligned or a dispute.
Sailesh (37:23)
Yeah, so this is where I think one of the again, this because I'm a big advocate of open source. Okay. And is it, you know, because open source tends to be a labor of love for individuals. Right. So an open source project will develop because the main contributor wants to make it develop in a particular way. They will constantly choose what features goes in, goes out. Which contributions are accepted, rejected. There's no construct which makes the open source creator have to demonstrate their value and encode it across multiple lenses. So my concept of value, if you look at it, is a range. So you can understand the trade offs, it might be economic, it might be resilience, it might be, you know, I value the reputational impact. And then when you tag value to a decision, okay, then you can kind of see for this decision what, what values does it align with and it may be multiple. And if you, as you said, I think that my, my goal for this framework is it's applicable to all organizations and all software systems, including open source projects, because they too are an organization and an ecosystem.
Rohit (38:48)
So does it already implement the graph sort of layout for the information or because that is funny picture.
Sailesh (38:56)
So like any good engineer or architect, I, I have built all of that. So as I, so part of the process is as I the without grammar, I'm building the immutable graph on the other side and the projections. It's just not public. Because to me, one... I frame it as that's my that's the commercial advantage. Okay. Right. So like any good freemium open source project, you've got the core capabilities there. And I think the thinking around the graph and future projection, you know, everyone's got a graph technology. Everyone has that today. So you could implement it yourself. Okay, now question is, have you thought everything through quite like you know, like your product, right? It's like, you may have, made, you know, the advantage is what you bring to the common constructs and how you join it together and bring your expertise together on it. Is where the extra value is. So I would suggest that the graph construct I'm creating is a little bit further forward than most people think on a normal basis. Now, again, I'm trying to write this up into a book and I'm not going to share the graph full thinking the or second book. So the first one is just going to be about Transformation Intelligence and then how you think about intelligent transformation afterwards is a much wider, broader, more complex concept. In my history, you know, things like new payments architecture, I, I'm very much into ecosystem level thinking. Okay. And so you'll see some of my recent articles on fraud. Extend this very same thinking process into all the participants of a fraud journey. And what you'll see is things like delegated authority, decision-making values, inheritance, all come up in the same process. So I'm trying to demonstrate. So would you design something like this today? Right. And I think what people will see is that the language of fraud, while it may apply to banking, you could use the same language and principles if you're in a retail organization or if in your situation you're trying to sell product to a client. I think the language and the risks and the articulation is different, but the fundamental, I'll call them systems thinking patterns. Are identical. I'm sure that
Rohit (41:41)
That makes sense, so I think that gives me and so The binary that I had told you about, this small CRI tool that I'd made for my agents, it's called Flow. That is also open source. And I also mined some knowledge base. I call it knowledge base items from each session once it has done. You're explicitly asked to say that, Hey, my task is done. And it minds certain information after each job and saves it in a knowledge base. Right now, it's plain text. I also had some plans to move it into a graph. But there, are the dimensions I had was organization, people, and processes. Right. And I think organization probably now needs to be all the decisions that were made needs to go there as an inspiration that I take from your project and probably I'll incorporate. I think the framework, board, it's called MTDR.
Sailesh (42:34)
Yeah, it's a markdown. Transformation decision record. And inside there is all of the decision records, the records, the authority and inheritance will emerge. I mean, sort of testing mode. So I test all of these inside the bank with many engineers, me and my test. It did that kind of thing. I think the other lesson I'd finally leave you with is in transformations, okay, which a question that doesn't get asked is efficiency of transformation and the challenge of transformation, okay? So if we the concept of zero based financial planning. Okay. And you go, okay, what would happen if you had zero time deployments? So what would happen if you could deploy everything instantly? Okay, what would change in an organization? So, you know, what happens to your traditional environment management? Right, would it work? What would happen to release management? If you could, what happens to disaster recovery? If you could deploy your entire organization instantly, what is disaster recovery? What happens to your infrastructure operations? What happens to your change approval process? What does operational resilience mean? So I think a lot of times people, so when they think about the intelligent transformation, they don't go, okay, what would happen if I didn't need any people? What processes would I need to change? So I mean, I'm sure you find it with your product, right? People do not utilize your product to the full maximum capability because they don't challenge themselves to say, okay, what happens if everything happened instantly at the cost, right, effectively? And what would I need to change in my organization to make that true? I think if people don't think like that, I think they're going to struggle with the adoption of AI. Because the big differential I'd say is, you know, engineering, you can engineer everything instantly now. Yeah. So the real challenge, so this is the battle everyone's had, right? So since the beginning of time, the organized engineering has been the slow part. Yeah. The actual build process has taken months, if not years to execute. So the organisation had plenty of time to make decisions, change its mind, do everything. With Agile it wound up being, okay, you've got two weeks or a PI planning quarter where you've got to make the decisions with discipline because in the next quarter it's actually going to go live. In the next two weeks it's actually going to go live. Right, so you needed to put... It compressed that, you know, force that decision making and authority and the value measurement and the people conversation into a very compressed time scale. So the thought I'd experiment I'd leave you with is what happens with AI when that execution time goes to zero. Yeah. So imagine your agile feature team, your scrum team, your value stream being able to deliver everything instantly. Right. Yeah. So I think what happens to your Organization? What does CIO do?
Rohit (46:06)
Everyone needs to have a North Star. It might be a utopian North Star, but have it and then project your situation, your future along the line.
Sailesh (46:15)
I think the element here is if you adopt AI in a sensible way, you're not adopting it across the whole organization. You'll implement it as a glass tube concept, like a trial specific pilot into production. And that will be instant. That could be instant. OK, so the rest of your organization might be slow. But what you need to be doing is demonstrating that AI can be instant and what does that mean for the organization. So just like you can you know, build a web page, previously took you weeks and weeks, now it's literally 15 minutes. Right? That's instant. So if you imagine all the web development suddenly is all instant in your organization, what happens?
Rohit (47:04)
So what you would do right now is have an experimental glass tube where AI writes code, provisions, ships, makes it live, maintains. Build that pipeline, the rest of the code could stay as There is no point in giving a co-pilot to each one to speed up whatever they're doing on their laptops. Get this done, then your organization is
Sailesh (47:28)
Transformed. But no, what happens is that people can see, right, what does the lead engineer in this instant organization do? What is the role of the DevOps engineer in this new instant model? How does production operations change? And how does CAB need to change? How does architecture need to start thinking in order to make that happen? So in your glass tube, what does risk need to do? How does security need to change? How do you know, everyone, how do they all need to change? Okay, because then you've done it without job titles, without putting you know, people at risk. You've used your most talented individuals to bring them along the journey. And if they're the right people, they'll act as champions and evangelists for the model. Because the hard part about this is, and you'll see this everywhere, is engineers aren't adopting it. Right? They don't want to adopt it. They don't believe it. They all played with it a year, two years, three years ago. And then said, oh, it hallucinates. The code isn't any good. It's, you know, it's rubbish, blah, blah, blah. But then the challenge I have for those engineers is, well, did you do TDD? When you adopted TDD, how much did it change your engineering practice? And how hard was that to adopt, right? So I got 2000 engineers to adopt TDD and it was a battle all the way. But the velocity and the quality improvement was dramatic. Okay, and we know from you know, our own Claude implementations that writing the test cases first And then getting it to implement and then getting it to in a loop, out a loop, graph, blah, blah is the most efficient way of controlled execution. And then because you're treating that code the same way as human-written code, right, because Claude could be a drunken engineer in your organization. Right? So you can go and put the same SonarQube and all of the SAST and DAST testing and code linting checks and security controls in your existing pipeline and apply it to all of the same code. Yeah, so it's no different from a rogue engineer with a bad code review from their senior manager. Makes sense.
Rohit (49:53)
Yeah. Yeah, so that is a great, great. I mean, I think most organizations are right now, you know, just getting a Claude subscription and hoping that, you know, magically they will have bigger, better quality, faster shipping and all of that. But I think the suggestion that you have brought up definitely makes a strong case for you know, having that ambition. I think the ambition is it needs to need to reset. It. I don't think people appreciate where this cannot lead to.
Sailesh (50:24)
Yeah, it's pretty scary when they're spending a trillion dollars of investment, isn't it? It's Yeah, and an awful lot of people are putting their careers on the line for AI. And, you know, that's what I say. I think it's a case of challenging for AI leaders. To try and make sure that their adoption. Of their tooling and processes is successful. And I think, you know, as I said, what I'd suggest is, is, that transformation intelligence of understanding the organization and how it's going to, you know, structure the decisions and reactions is central to successful deployment. Now, whether it's AI or a regular project, I think the model works regardless. I think it's it's the usual, you know, more success comes through more disciplined approach. It's not it's not a, it's not a radical concept.
Rohit (51:19)
I get it. Yeah. And I think, yeah, so I think we're nearing the end of our conversation. So it's, I had more topics to discuss with you, probably the cloud transformation itself, your reflection on how it went and all. But I think it was worthwhile to explore intelligent transformation. Itself because that is truly what we are discussing here. And I think intelligent and disciplined transformation is what we're discussing here. I think great insights have come out from this conversation, Sailesh. Thanks a lot for your time here. I think probably we would connect back for, you know, a follow-up episode sometime later, where we can we can discuss the other topics that we would encourage this time. Yeah, yeah, I think, you know,
Sailesh (52:06)
What I'd leave the listeners with is, you know, this is a journey that we've been through for thousands of years. We're optimizing it now post-industrial revolution. We're now getting to the point of post agile, AI is sharpening our aim for needing to do this. Maximize the investment. You know, what we're saying now is cloud has made infrastructure programmable, right? AI has got the possibility of making the organizational model programmable. The next transformation isn't the digital transformation, it's an organizational reconstruction. Because AI will directly impact people and how they go through this. The competitive advantage isn't not having AI, it's the fact that everyone has got AI. And the advantage is going to be how can you redesign your organization quickly. Okay, and efficiently with the least risk. And I fundamentally say, you know, the whole message of Transformation Intelligence is the future organization needs to understand itself and then redesign itself and increasingly reconstruct itself. In order to succeed and win in this sort of current phase of evolution as we look at organizational evolution and not to treat AI as just another technology like cloud. Because cloud only touched infrastructure, this is changing the entire organization and your people and how they think and how you deliver. And it really means that leaders need to spend time thinking because that's what the only advantage humans have over these, over the AI is really thinking about how you're going to change and what you value as an organization.
Rohit (54:01)
Yeah, I think Sailesh, I'll truly leave the listeners with that message. I wouldn't want to add anything more to that. Thank you for your time, I would love to connect back with you for a full episode sometime later. And yeah, have a nice day.
Sailesh (54:15)
Thank you. Thanks a lot, Rohit. Bye.
Rohit (54:18)
All right, folks. That's a wrap for today's episode of AI in DevOps. I hope you found it fun and thought-provoking. It's okay if left you with more questions than answers. That's kind of the point. We're trying to figure out the role of AI in DevOps. If you have somebody in mind who you think we should invite onto the podcast, and you would like to listen to them, send us their coordinates, we'll invite them wherever you're listening, whether it be Spotify or Apple, whatever platform, just click on the subscribe button so that you get notified whenever we have a new guest.