#258 - Why Your AI Strategy Is Failing: The AI Paradox of Optimizing Coding Alone - Andrew Haschka

[00:02:02] Introduction

Henry Suryawirawan: Welcome back to another new episode of the Tech Lead Journal podcast. Today I have with me Andrew Haschka. He’s the Field CTO at GitLab. Very excited to have him in the show to talk about what GitLab is doing with its customers with the impact of AI, and how GitLab is kind of like helping them to revolutionize their software development lifecycle with AI. So Andrew, looking forward for this conversation. Thank you for your time today.

Andrew Haschka: Great to be here. Thank you for having me.

[00:02:30] What Are the Key Responsibilities of a Field CTO at GitLab?

Henry Suryawirawan: Alright. To start with, I’m a bit curious about the Field CTO role. Can you maybe explain a little bit what is your day-to-day job? What is the role and responsibilities of a Field CTO at GitLab?

Andrew Haschka: Yeah. So my main focus as Field CTO for Asia Pacific and Japan is to meet customer executives and provide that connection between technology transformation and their business outcomes. So I spend a lot of time working with CTOs, but also CIOs, CISOs, CFOs, and some CEOs across the region. And that would typically cover Australia, New Zealand, across to India, Korea, Japan, Southeast Asia. And I think across those markets there’s a lot of interests around how different industries, different countries, different market segments may be operating. And I usually share some of that context with those leaders when I meet with them as well.

[00:03:26] Why Should Organizations Govern AI Strategy Rather Than Chase the Latest Features?

Henry Suryawirawan: Wow, very exciting that you got a chance to talk to all these business leaders. And especially these days, it’s all, I’m sure, it’s all about AI and transforming organizations with AI. And there are a lot of things in the news. How do you actually keep up with all these changes, while at the same time advising your customers with the right context and with the right solutions?

Andrew Haschka: Yeah, so I think the pace of AI means that I probably spend a lot more time talking about the transformational path for organizations from where they’re at at the moment through to the vision of, you know, more best practice use of agentic flow and agentic AI. I typically help these customers understand the difference between the noise that they’re hearing in the market and things that will typically add value to their businesses. So often, when I go and work with customers from a top down point of view, from a bottoms up point of view as well, developers and practitioners, servicing value to business leaders, it’s really important to understand the outcomes that we’re delivering for customers, the delivery capability, and the risk posture that they’re trying to solve for, rather than just the latest new feature that they may have come across in the media. So I would say, most organizations are looking to improve productivity and operational efficiency in the use of AI. But often, it’s very fragmented and silos, and siloed in the way that they adopt AI. There’s maybe limited context for specific outcomes. You know, there’s such a fast pace of new features, new capabilities, new models in the market. And I think it’s really important to take a step back and talk about, look, what are the measurable outcomes that we can actually improve upon? You know, what are we planning on using agentic AI for that can actually help us achieve some of those board level outcomes, on the annual report for these organizations.

So I think one of the more exciting things, that enterprise leaders are really interested in at the moment, is the point of context with AI. I mean it’s no longer a prompt and response type of engagement that we’re seeing, you know, with agentic AI at the moment. And I see a real evolution there towards multi-step engineering workflows without losing that coherence, with having that persistence, and that full context from business all the way through to customer. I would say that that’s structurally very different to the way that organizations have engaged with AI in the past.

Yeah, one other part on that is around risk. I would say that often organizations are adopting AI at such a rapid pace, but enterprise governance and risk management are maybe secondary thoughts. And so there’s a genuine inflection point now towards elite performers and those who use the latest AI features and widgets. And that those that govern first and optimize their process and culture and the way that they leverage AI are becoming far more successful with an outcome based approach.

[00:06:41] Why Is an End-to-End Agentic Flow More Valuable Than Individual AI Tools?

Henry Suryawirawan: Right. Definitely one of the key that you mentioned is know what the measurable outcomes you wanna do, right? Obviously, there are so many solutions, And in fact, there are plenty of solutions, be it open source or maybe vendors selling their AI solutions. Maybe a little bit of tips, if you have to choose from all these seemingly plenty of vendors, right, how would you pick the good vendor that, you know, differentiates versus all the others, right? What kind of criteria would you assess?

Andrew Haschka: Typically, I think there are models that service specific outcomes in a more beneficial way than maybe other models in the market. Now I think organizations need a level of choice and that’s why GitLab Duo Agent Platform allows these organizations to choose the model that will service that outcome. You know, they can leverage the agents that we have built and integrated within our stack. They can also leverage third-party agents or agents that they have leveraged themselves. But centralizing access, I think is really critical to governance. And I think not just considering one model for one outcome. I think considering the evolution of agentic AI requires a concept of flow where you have that system of record that evolves to a knowledge graph, to reduce context pollution, to go and give all of those agentic tasks the very specific data that they need to improve quality, to reduce hallucinations. But each of those agentic AI engagements require some level of path from the prior task into the next task. And so I think increasingly organizations are building flows. By which they can start with the business ideation and planning phase and go all the way through coding and testing and build, security, compliance, release, operate, and monitor. But have it as a consistent flow architected upfront rather than having to engage individually for each agent to achieve one specific outcome. Have some wait time. And then maybe even hand off to a human-based team.

I guess my advice for most organizations is to consider the end-to-end journey with regards to how they deliver value for the customer rather than just, you know, one coding task. And, you know, one agent, and one request, to go and build or write new code. I think the burden of having siloed and fragmented tasks delivered with AI means that there is a lot of cleanup that happens in testing or in security or in compliance or, you know, before go live in the release phase. And so being holistic I think is actually much more impactful than, you know, the individual engagement with a specific agent I find for organizations.

Henry Suryawirawan: I think thanks for the tips, right? Always think about the end-to-end journey. How do you foresee these AI solutions being injected into your workflow, be it software development or maybe like business operations, right?

[00:09:39] What Is the AI Paradox and How Does Intelligent Orchestration Solve It?

Henry Suryawirawan: So maybe let’s focus today more on the software development side. One of the key thing that we wanna talk about is about this AI paradox. I think GitLab, as your organization observe, right, we have here heard a lot about productivity improvement from the development point of view. Maybe you quote it like, 48%. But it seems like the organizations and software teams are hitting this AI paradox. Maybe let’s start from explaining, what do you mean this AI paradox?

Andrew Haschka: Yeah, so GitLab is re-imagining the software engineering ecosystem as a human and AI collaboration. I think developers will focus on solving complex problems and driving innovation while AI agents will handle more routine repetitive tasks. I think the AI paradox comes down to some problems that I’m seeing in the market where organizations may be launching into using AI to optimize coding, and coding alone. And we do a lot of assessments for organizations to evaluate that flow of work across all of the phases of the SDLC. And we often find that they only have AI adopted potentially in that coding phase.

Now coding, I guess, in terms of the value for the end-to-end software delivery lifecycle, yeah, is much smaller impact than, I guess, perceived from most teams that I’m talking to. Typically organizations spending less than 20% of their time coding, and often when we deliver more code faster, there’s more time that’s spent doing testing, or security validation and coverage, or compliance checks, or validating that what was written is actually going to meet the business outcomes.

And so I think while traditional DevOps has formed more of a closed loop, the process is still very manual in organizations and requires a lot of human intervention at each stage. Whereas what we’re doing in GitLab is, I guess the promise of intelligent orchestration for organizations where agentic AI can speed up these loops by automating them in parallel. Your development teams will operate above the loop, orchestrating and guiding multiple AI agents while agents perform tasks autonomously within the loop. I think the most important three aspects of intelligent orchestration is around the experience layer, which includes chat, but also reasoning across, say, issues based on that knowledge graph from that system of record, you know, from merge requests and pipelines and security findings. You know, I think being able to leverage more foundational agents that solve for some of the wider gaps beyond coding, like planning and having the planner agent or the security analyst agent, have really helped organizations to fast track their end-to-end lifecycle of software delivery.

I’d also say that it’s kind of a list to what I was talking about before. The control layer is super important in terms of governance of AI requires centralized access, access control to certain agents for specific outcomes, and building agentic flows that chains together these agents. And then allows through a native model context protocol or MCP to go and connect to some of those outer loop, tools that may exist within the organizations to date. I think the data layer is probably most critical in terms of solving for that AI paradox. You know, data, when it comes to giving quality AI outcomes can’t be siloed. You typically require a unified context across all repositories, merge request pipelines, security signals. And I think, there is only one platform that I’ve seen in the market with GitLab that can service that knowledge graph covering every phase of the SDLC. I think the result for customers in how they can get higher quality outcomes for AI is having faster access to that context that matters for that agentic based outcome.

And so this is typically what I coach a lot of organizations on how to maybe evolve what they’ve done historically with DevOps. You know, I spent some time with Bendigo and Adelaide Bank recently. Actually they consolidated multiple tools, down to one system of record and they did it within four weeks. It was very quick, their adoption path. And I think for them being able to move huge volumes of projects, 1,500 projects, 30 organizations, 500 users and 500 gigabytes of data to form that system of record is incredibly critical on that more evolutionary journey to adoption of agentic AI.

Henry Suryawirawan: Sounds very interesting what you elaborate just now, right? So maybe for people who don’t know, GitLab doesn’t just provide, you know, like version control, like maybe in the past people associate with version control and CI/CD. I think GitLab now encompasses, you know, so many other solutions as well.

[00:14:47] How Does Shifting Focus to Requirements Quality Transform Software Delivery Outcomes?

Henry Suryawirawan: So maybe I’m quite interested when you say about intelligent orchestrator. Maybe if you can give an example of, you know, a software engineering teams, you know, using GitLab with this orchestrator, what does their day-to-day look like actually? Because all I imagine is like developers, you know, writing code, you know, raising merge requests, you know, merge request pulled and, you know, those CI/CD running. So how does this intelligent orchestrator now change the way software development team are working?

Andrew Haschka: Yeah, I’ve run a number of executive roundtables across the region recently and have spoken to many customers. I would say, you know, Deacon University is one of those customers that I’d spoken to. And one of their objectives, you know, to your point, is to improve developer experience, whilst also simplifying and consolidating that tool chain, the process, the culture by which they operate and improving quality outcomes for their consumers of their software. I think beyond writing code, there’s this concept that many organizations look at, aligned with maybe the requirements that you put into the coding phase, is going to be a precursor to the results that you get out of the coding phase. So to allude to a comment that a customer made recently, they said, well, if we put rubbish in to that coding agent or, you know, that developer phase, then often what we get out of that phase is going to be similarly aligned with the subpar inputs.

And so aligning to quality is hugely important. And that sort of quality, those quality requirements are actually built in the planning phase. And so I think it’s incredibly important to think beyond the more traditional DevOps metrics of speed and stability improvements when it comes to AI, and thinking holistically around how do we deliver quality outcomes for our customer? And I met another customer in Thailand recently, their CTO in the entertainment and retail industry. And he said to me, look, we deliver software once every hour. We have less than one or two percent failures in what we deliver. He didn’t think that agentic AI was going to help him deliver faster or with greater stability.

But what he did think was that, maybe quality was an area that could add benefit to his team and his organization. And thinking about how do we curate a set of requirements that are well documented, have all the levels of context, have that end-to-end closed loop, with access through that knowledge graph, you know, to all of that valuable data in the SDLC, to go and build those requirements upfront. To build adequate testing cases that cover the breadth of requirement that maybe were fed into that coding phase. And then being able to check more holistically around, well, is it compliant for our regulators? Does it proactively meet all of the demands of our security posture requirements for our organization? And before we release, can we validate that what we’ve built was actually what we were trying to build at the start? And so I think an agentic approach and having a flow that comprises that end-to-end lifecycle is critically important for organizations as they evolve beyond just standard software delivery and DevSecOps today.

[00:18:19] How Has GitLab Evolved Beyond CI/CD Into a Full End-to-End Delivery Platform?

Henry Suryawirawan: Yeah, so does it mean that now software development teams will do that from planning, you know, on GitLab? You know, like what are the requirements? Maybe you have a set of template or do you interact with like a chatbots per se, you know. And then it will prompt you back and forth until it gets a clear requirements. And like how do you actually track that requirements up to the deployment and, you know, maybe being used by the users. Maybe just give us a illustration here, maybe for people who seem to be like, knowing GitLab in the past is about, you know, again, like CI/CD and all that.

Andrew Haschka: Yeah, look, I would say that a lot of organizations in their evolution and use of GitLab started with source code management and CI/CD. I think many have now evolved to leverage, in the planning phase, things like enterprise agile planning to consolidate how they do planning with regards to software delivery. They may leverage GitLab Ultimate, which would include the raft of security and compliance features from a shift left point of view as well, inclusive of SaaS and DAST and API scanning and fuzz testing, dependency scanning, software composition analysis, and compliance as policy. You know, all these things are critical in that coverage of software delivery. And so I think, most organizations are looking to solve for consolidation and simplification in one platform rather than having so many different silos in their software delivery journey. I think DevOps is almost 19 years old this year. And the evolution that I see in most organizations is accumulation of a lot of complexity. Too many tools, too many fragmented processes. And really, I’m seeing a lot more of that shift this year in favor of simplification, consolidation, and maybe leveraging GitLab as that one platform that can deliver the end-to-end, from planning, coding, building, testing, security, compliance, release, operate, and monitor.

[00:20:20] What Should Software Teams Prioritize Beyond Coding in the AI Era?

Henry Suryawirawan: Yeah. So again, I think just to highlight what you mentioned earlier, right, coding, probably most of the time only counts for like 20% of what you do in the software development lifecycle. The rest could be, you know, like coordination, you know, planning, testing, security and all that. From your point of view these days, right, with the AI part kind of like solving the coding aspect, you know, I think we all agree that AI can speed up development process. What other things that development teams now have to shift more focus on? You know, you mentioned earlier about testing, security, compliance, all those things. What do you think are the most important things that software development now needs to take care about?

Andrew Haschka: I think it’s definitely individual. And one of the more foundational concepts of software delivery in DevOps has been work visibility and the value stream. And having an understanding around that flow of work, the lead time, the process time, the rework that is involved in delivering software. And for a lot of organizations having a priority list of things that they should move from manual toil through to more automated flows. And agentic flows is typically the path. And some of the bigger gaps I see in having done many of these assessments across organizations is there are very real gaps in the time spent planning. And really a huge amount of value for organizations in having more value added time, and less non-value added time, and less rework involved in the planning phase. Definitely a lot of time needs to be spent in improving a very fragmented way of working in security and compliance.

Often compliance, for organizations that I speak to, is done after deployment, checking with a spreadsheet. And maybe it could be three months or even six months after deployment to go and check whether the application you’ve delivered to your customer is compliant or not. So I think we can definitely be much more proactive around how we do that. And GitLab Ultimate services those needs for organizations too. I would say that organizations are looking to improve their stability and quality outcomes for customers. And so having an agentic approach to ensuring that your testing coverage is complete is really valuable. Being able to have code review that is done in agentic way can maybe add far more value and be delivered in much less time with GitLab Duo Agent Platform as well.

And so I would say that typically the path for organizations should be benchmark your current state. Have a value stream map. Understand that flow of work. Highlight the tasks that take a huge volume of time in your organization, and then work with GitLab to help consolidate those to a more agentic flow way of operating. I would say that it has to be specific to the organization, it has to be measurable, and it has to be business aligned. And think about if we can improve the software delivery speed, stability, and quality, then more holistically in terms of what the customer will get through this approach is gonna be much more valuable for their experience.

Henry Suryawirawan: Yeah. I think, yeah, definitely it should be contextual. But no matter what, I think software development team, like you mentioned, we want to prioritize speed, delivery speed, right? Stability. The quality, also, aspect of the outcome that we deliver. And I think specifically for development teams, right? We have been talking a lot about shifting left, shifting left, right? And probably this time with AI, we really can extremely shift left you know, so many things that we can do earlier now. Especially if you are talking about orchestrating multiple agents, maybe, yeah, we can have more agents in parallel working towards that future, you know, shifting most of the things on the left.

[00:24:14] How Do Organizational Silos Create a Capability Threshold for AI Adoption?

Henry Suryawirawan: So another thing that you say you wanna talk about, right? We have seen a lot of, you know, AI models out there in the market. Some are the big names like Anthropic with Claude, you know, Gemini, Google Gemini and OpenAI ChatGPT and all that. They seem to be having an arms race, you know, with each other, improving their models from time to time. And the pace of change I think is quite frightening. And you say all these change actually, you know, gives an organization actually the capability threshold. So what do you mean by this? Does it mean that organization is not able to absorb those capabilities unless something changes inside?

Andrew Haschka: I would say many organizations that I work with are very excited about the use of the latest capabilities of some of the latest models as you’ve mentioned. Yeah, I do think there is a capability threshold when you operate in a silo. There’s limit of context, there is maybe context pollution that might be flooding some of these models, giving subpar outcomes because we’re not specific around, you know, what we’re trying to achieve, you know, with the request or with with the outcome. I think the latest generation of models, can maintain intent across an entire workflow. Yeah, it can plan a feature, it can build the code, it can run test suite, it can interpret failures, propose fixes, and iterate without losing that thread between steps. So that’s a more architectural shift in the way that we consider leveraging agentic AI today. I think some of the main questions I get asked is, you know, which model should I use for which approach? But I think considering the way that agentic AI in these later models are able to hold that context, we should really be asking, you know, which workflows can I hand off to an orchestrated agent, and how do I govern that effectively. And which tools do I need to go and connect to through model context protocol to service that level of knowledge to those agentic tasks when required.

I would say most organizations, when I speak with them, it’s beyond the use of the agent that is holding them back. I think the generic access to models is not very specific in the outcomes that we are hoping the models will add value. So having enterprise guardrails where you’ve got deployment flexibility, which lets teams choose how to leverage the agents for specific outcomes is incredibly valuable. Being able to choose, I guess, which part of the lifecycle they would like a agentic approach to happen or whether they can have the end-to-end flow defined for them upfront with some inputs based on the specific outcome they’re trying to achieve. I think unifying the entire approach through DevOps, through security, through compliance is critical though. I think in many cases, operating, you know, one model, one agent in a silo with fragmented context is not achieving the outcomes that many practitioners expect, you know, when they start engaging with these models.

[00:27:49] What Practical Strategies Can Organizations Use to Break Down Internal Silos?

Henry Suryawirawan: Yeah. You mentioned quite a few times now about silos. You know, the problem with organizations having a lot of silos. What practically maybe have you advised or have you seen with customers out there breaking these silos, right? Is it like merging two different departments or now they have a, you know, centralized knowledge base or they have some different ways of, you know, breaking these silos. Maybe if you can share a few tips here, that would be great.

Andrew Haschka: Yeah. So I often talk to customers about the concept of inner loop and outer loop. Now I guess we’re all relatively familiar with this in the landscape of software delivery and DevOps. But if you consider GitLab as a end-to-end software development and security platform with agentic AI across every phase of the SDLC, often organizations require a level of context within that inner loop. And the inner loop, is typically going to be servicing that one system of record across the end-to-end SDLC. Now that only comes when you consolidate all of the tools and processes and ways of working into that platform. I think there is definite value in having that outer loop connectivity. You’re leveraging MCP or model context protocol to go and connect with external tools as required. But really I think, you know, the core value that we see for organizations is consolidation, integration into that system of record for that knowledge graph in order to streamline traceability across their workflow.

You know, I was working with a large bank in Singapore recently. And, you know, they had considered that consolidation of their different processes and their tools was the first step in their journey. I also worked with Airwallex recently. They’re a global financial platform. And they leverage GitLab’s AI powered end-to-end DevSecOps platform, typically to focus on expanding their business coverage and meet their customer requests faster. You know. a lot of what their focus on originally was speed. You know, how do they have faster deployments? Now, if you think about optimizing productivity, streamlining the efficiency of the workflow end-to-end from business ideation and planning all the way through to customer, speed and accelerated development comes from consolidation and reducing handoffs between those silos. And so Airwallex managed to get eight times faster deployments by consolidating many of their tools and processes into GitLab. And so I would say that the core premise needs to start, for most organizations, not with, you know, what is the latest feature potentially that you may wanna leverage from some vendor. But number one, do you have a system of record and can you evolve that to a knowledge graph that gives your agentic AI flow that full context that it needs for quality outcomes without those siloed handoffs between each of the phases.

[00:30:58] How Did Ericsson Achieve 50% Faster Deployments and Save 130,000 Hours With GitLab?

Henry Suryawirawan: Nice! And I think you have plenty of customer success stories as well. So one that you quoted before this conversation is about Ericsson. I think the numbers look really fantastic. Maybe can you share about that success story for Ericsson?

Andrew Haschka: Yeah. So Ericsson was very interesting as a case study. They achieved 50% faster deployments and saved 130,000 hours in six months. That’s a lot of time to optimize within an organization. I think the value with Ericsson is that they tackled the end-to-end lifecycle, the whole value stream, not just that coding layer. They use GitLab’s platform to instrument their delivery pipeline end-to-end, you know, identify from business ideation and planning all the way through to delivery, not just code generation. If we think about how organizations like Ericsson need to evolve and improve, the biggest challenge that they’re solving for is managing complex enterprise software deployments across, you know, global environments.

So in Ericsson’s case, 300 plus global communication service providers that they’re engaging with. I think consolidation and simplification as a first path was critical to their success. Unifying into GitLab’s platform was a key part of their strategy for that system of record. And then allowed them to cut that deployment time in half and enable 10 times, you know, better testing scenarios per release. So better testing evolves into better quality outcomes for customers and better stability of the service, that you then go and offer to those customers. Which means that organizations like Ericsson and even Ericsson themselves in telecommunications industry can deliver updates to their customers in weeks rather than months, perhaps.

Henry Suryawirawan: Nice! So, I hope more organizations also embark this journey and be able to improve, you know, to the level of outcome that Ericsson just achieved, right?

[00:33:07] How Should Software Developers Evolve in the Age of AI Agents?

Henry Suryawirawan: So I wanna switch a gear a little bit about, impact to software developers, right? We all know organizations want to transform. And at the same time, software developers, right, some of us are still running, you know, with our traditional habits and mindsets. So maybe from the first thing, what would be your advice to software developers now? Knowing that, for some, you know, their identities like having a crisis. So in the past they used to write code, now agents, AI agents are writing the code for them. So maybe let’s start from here. What are some of your tips for software developers?

Andrew Haschka: Yeah, I think systems thinking is evolving to become a core skill. You know, when AI can write code, competitive advantage moves from the person who can define what the system should do. do And decompose it into the right components and evaluate whether what was built actually solves the problem. I think many practitioners, many developers writing code that I speak to, have focused on prompt engineering. I would say that really evolves into agentic orchestration fluency. You know, understanding how to give AI agents well structured intent. How to decompose those complex problems into agent executable tasks and how to critically evaluate AI output.

I think in many cases, sometimes we define a very large problem for one AI agent and what results is, the quality is maybe not what we would expect. I guess core foundations to DevOps is work in small batches. And I would say the evolution with agentic AI is the same. You know, being able to divide and decompose these tasks into smaller outcomes and then validate the quality and orchestrate those agents in a flow is becoming an increasingly valuable skill. Maybe, you know, the AI guardian or the cognitive architect or, you know, those that can design flow, is a new shift that I’m seeing in the market.

I would say maybe last point is around the thought of security and quality judgment. If AI is writing 90% of the code, you know, the human’s most important contribution is knowing whether it’s safe to ship and being able to evaluate security from security literacy point of view, testing intuition, the ability to read AI-generated code critically, being able to think about it’s all very well that we have developed this code, but hopefully we’re not then handing off to a manual process for review. We’re augmenting that with an agentic path. But we are, I guess, the director of that flow. You know, we can check in and validate whether the outcomes were correct. And I guess as we were expecting, you know, when we started this lifecycle.

Henry Suryawirawan: I think systems thinking, I think, definitely is very crucial, right, these days. And also maybe understanding about the product, the outcome that company wants to produce. I think still that it gives a high leverage, right? Because now you can think of, you know, different way of, maybe delivering the solutions, thinking about architecture and design and all that.

[00:36:26] How Is the Tech Lead Role Evolving in a Hybrid Human-AI Team?

Henry Suryawirawan: So specifically also I think, you mentioned that the tech lead’s role is gonna be more important. I think in some companies, right, they think of shrinking software development team, you know, they want to reduce the number of people. So maybe if you have some opinions on this, like what’s the… what’s a good team size these days, and what should the tech lead capability now have?

Andrew Haschka: I think team size is definitely based on the outcome, the organization, the industry, the country that one works in. I think once again, mapping out that flow of work through value stream is critical to understanding the amount of value added versus non-value added time and the amount of time that’s required to achieve the outcome. I think it has to be specific to the outcome and to the level of efficiency that is required by the organization.

I would say the tech lead role that I’m seeing in organizations is more fundamentally now about translating intent into executable direction. So when a team includes AI agents and human practitioners, that translation skill becomes more critical and more demanding. We’re managing a hybrid team these days. And in many cases, we may have agents managing and orchestrating and handing off to other agents in flow as well. So I guess the role needs to pivot to specify intent with enough precision that an agent can execute reliably, decompose complex requirements into well bounded tasks and maintain that coherent architecture across agent generated components.

So I’d say that’s actually a harder version of the existing job for organizations, not an easier one. I think we are upleveling, you know, many practitioners and roles in organizations to provide that intent, and to own that quality outcome, and to consider thinking about that agentic flow upfront rather than individual interactions. And from a GitLab point of view, we typically see a shift in the way that customers are leveraging our platform. You know, there is much more focus on sound architectural decision making upfront, you know, understanding the system design and making strategic technical choices. Having a very product-minded approach and a system-oriented approach, from a design and from a process point of view. And then mastering automated testing, observability, managing tech debt, removing some of that manual toil, and legacy technology and process that exists out there. And then orchestrating systems and guiding those agents, moving from just execution through to more direction setting capabilities within the platform.

[00:39:22] How Can Junior Developers Keep Up With the Rapid Shift in Industry Expectations?

Henry Suryawirawan: Sounds like plenty of things that a developer now needs to be responsible of. I was wondering specifically for juniors, like not necessary they have all this knowledge and capability, everything that you just mentioned just now. So how should they keep up with all these changes, especially if they wanna pursue, still pursue software development career?

Andrew Haschka: Yeah, look, I think career ladders are shifting from just delivering lines of code and shipping features and merging PRs, more to measuring judgment and impact. I think training programs are evolving to thinking about how do we architect an end-to-end workflow for agent execution, or how to evaluate AI output for security and correctness, or how to maintain accountability when AI is doing some of the implementation work. So I think junior developers are at an inflection point at the moment. You know, the more traditional path of writing lots of code or learning from mistakes or building pattern recognition is being disrupted. I think the goal is typically thinking at a higher level around outcome, systems, end-to-end flow rather than just individual coding activities. I would say that many organizations I speak to are leveraging newfound time in reducing the amount of non-value added time that their teams are having in favor of delivering more releases or optimizing stability or improving quality.

And so I think there is an evolution now of having more time to deliver more outcomes, to generate more revenue for customers. And so I think the role of a junior developer is now really critical in ensuring that they’re building that intent upfront. And they are learning along the path of how to leverage the latest features to go and deliver that value in a governed way. And I think maybe one last point on that. Organizations that have started their adoption of agentic AI with individual access from the user, you know, direct to that AI platform rather than centralizing and governing and having a prompt library or an agentic AI or flow-based catalog, are really limiting themselves and their organization. So I’d recommend that they start by setting, you know, these junior developers and practitioners up well from the start, by giving them access to those prompt libraries, and to those AI agent and flow catalogs, to give them access to the tools that they need to achieve quality outcomes for the business.

Henry Suryawirawan: Yeah, I hope organizations still give some chances to these junior developers, right? Not necessarily stop hiring all of them. So I think thanks for the tips for juniors out there. So, you need to start moving maybe your skills from, you know, being able to describe intent, being able to see end-to-end workflow. I think Andrew has mentioned a couple of times about all this. So definitely good luck with the journey.

[00:42:40] Why Do 79% of Singapore DevSecOps Practitioners Believe AI Will Create More Jobs?

Henry Suryawirawan: So one other things that you also shared with me before this recording is that there’s a survey that you did with, you know, like Asia Pacific DevSecOps practitioners, specifically in Singapore. And they believe that, like you quoted 79% of that, they believe that AI will create more jobs for engineers, not fewer. I think definitely there are a lot of debates around this, whether it’ll create more jobs for engineers or fewer jobs. So what’s your take on this, like why the survey is so optimistic?

Andrew Haschka: Yeah, absolutely. I think AI has dramatically reduced or lowered the barrier to building software so more people can now participate in software creation. Whereas previously that may have been an obstacle to more people joining in, you know, the engineering capability for organizations. I think the demand for software is essentially unlimited. Now we’re making it easier to build software which means that more software gets built, not less software. I don’t think organizations stay still. I don’t think they deliver the same number of features or the same volume software that they did last year. I think there is a constant push to deliver more at scale to help these organizations grow. I think faster delivery creates more opportunities. Your organizations can tackle more projects and serve more use cases and innovate more rapidly than they have before.

I also think new problems are evolving, that’s need solving too. You know, AI handles routine implementation, but there are still new challenges that humans will need to govern and solve for in orchestration, in architecture, in intent, in governance, that creates that demand in the market. And I think, business value is only going to compound. You know, when software delivery accelerates, organizations will invest more in digital initiatives and that’ll create more engineering work. And I think the natural evolution for most engineering leaders that I speak to is that there is a path for further growth. They do need to deliver more frequently to create more revenue for their organization. You know, they do need to reverse the number of vulnerabilities and compliance issues that they have. You know, they do need to focus on repurposing and reusing the time for higher level and higher value tasks with the teams that they’re building and hiring. And that the team will be hybrid. There’ll be a mixture of AI agents and flows, and humans working in collaboration together.

[00:45:27] Why Are Companies Reducing Staff Despite the Growing Demand for Software?

Henry Suryawirawan: Yeah, So I think what you said is true, right? Software demands won’t just disappear, right? It’s not like we have solved enough problems that can be done with software. So but specifically in the market, we can see so many companies, you know, did their layoffs right? And some of them are software development teams. What do you see this so-called, how should I say? Like the, why the difference of why companies now are actually reducing more people versus actually you can foresee so many things that developers or people can do now with agentic AI and they can serve more use cases, solve more problems. So what is actually happening? Maybe the pressure of the market or something like that. So what’s your take here?

Andrew Haschka: Look, I think many organizations that I have spoken to are potentially growing their capability. You know, in our C-suite survey from last year for the Singapore region, we found that 71% of Singapore executives believe that the optimal human-AI split should be around 50:50. Whereas the reality from the data showed that humans were handling three quarters of the work and AI one quarter. Now I think the opportunity there is really a growth opportunity to have more agentic flow with AI to leverage the human capability, but think about how we leverage that human capital and value that exists. I think in more organizations, they’re redesigning their process. They’re modernizing their software delivery practices rather than thinking that they need to reduce their head count or potentially lower the volume of staff that they have to deliver increasingly more demanding outcomes for their customers.

You know, in the customers that I have met, often there is more of a shift than a reduction. I would say that realigning people with higher value tasks and outcomes, is usually the desire and direction from most engineering leaders that I have spoken to. I would say that for organizations, accountability to the team and the outcome should be governed. And maybe the flaw that exists in some organizations is that they may see benefit in a coding phase by using an AI agent. But largely the use of that AI platform is direct to the AI platform. There’s no guardrail. Finding how that platform should be leveraged or how it may be serviced from an agent that sits in the planning phase or in the build phase or in the testing phase. And so, I think in many cases, the roles that I see in organizations will need to evolve to that level of governance capability, to define that intent, to govern the workflow, to design an architect, the end-to-end flow of work. And maybe that’s not happening in some organizations where it potentially should be.

[00:48:34] What Are the Most Common Pitfalls When Implementing Agentic Workflows?

Henry Suryawirawan: So thanks for sharing about that, right, Andrew. So for people who are now thinking of adopting more agentic workflows, so many people now are into this, right? So maybe tell us some of the pitfalls that organizations would see if they start, you know, implementing many of their workflows in the agents. Like what do you see typically some problems that they need to be aware of?

Andrew Haschka: When I speak to CTOs, engineering leaders, most of them are thinking about AI adoption. And when we ask them, what is your measure of success for this AI adoption project that you have for your organization, many of them say, oh, we would like to ensure that 100% of our staff are using AI. That is a measure of success for us. I would say that there is a gap there. And that it’s not enough to just leverage AI in a generic way. I think the result needs to be a measured approach to say, this is the task, this is the process that we are improving upon. And to my earlier point, defining that maybe code review takes three weeks for an organization with three people, you know, hundreds if not thousands of hours for some organizations. And being able to reduce that down to 30 minutes, with agentic code review flow or an agent, is typically a measure of success that can be correlated and shared with a broader leadership team and potentially even the board for quality outcomes of revenue generating software. And so very few of these leaders are thinking about AI governance. They think, not thinking about measured adoption of AI. And the organizations that will win in the next three years aren’t just the ones that adopt AI fastest. They’re the ones that will build the institutional capacity to govern AI responsibly. Governance is the competitive advantage going forward I believe, not just a compliance burden. And it goes far beyond just the use of new agentic AI features.

I would say the second thing is that interface between AI-generated software and the humans are who use it. So most leaders are focused on the production side: how do we build faster? But the consumption side is equally important. So the risk of producing technically impressive software that users don’t value is much higher in the agentic world. So really ensuring that we can focus on quality outcomes, not just faster outcomes that are not as valuable to the customers or to the business in creating more revenue.

Maybe, you know, the third point here is what happens to your engineering culture? You know, and I think you maybe touched on this in the prior question, but culture is what keeps quality high. It drives innovation, it attracts talent. If your engineers feel like they’ve been reduced to just approving AI output rather than building things, you know, the best talent will leave. The leaders who attract and retain the best people are those creating a genuine partnership model between human engineers and AI agents. You know, where humans are doing the most interesting high judgment work, not the most routine tasks. And so I think there needs to be a shift for many leaders to think about how do we achieve these outcomes more holistically for the organization in a measured way? And how do we reuse the time that we have saved and the adoption of governed AI to achieve higher quality and more frequent and more secure outcomes for our customers in order to generate new revenue streams.

[00:52:29] What Practical Steps Should Engineering Leaders Take to Govern AI Responsibly?

Henry Suryawirawan: I like specifically that you remind us about maintaining that engineering culture, right? Because culture in the end drives the quality, the output of the team, right? And specifically about governance, security, what are some of the tips that you wanna give for, you know, engineering leaders out there? Like specifically practical tips, how they should start with this governance and security aspects. Because, you know, sometimes, you know, people just consume and they don’t… They, like you mentioned, right? They forget about all these aspects and later on, you know, it becomes risky for organizations.

Andrew Haschka: Yeah. So, definitely having a measured approach is paramount. Yeah, I’ve worked with many executives across multiple countries in the APJ region. And I feel like a lot of these organizations need support and help in order to benchmark what they currently do well and where some of those gaps and maybe higher propensity tasks are for adoption and improvement with agentic AI and flow. I would say, measurement is not just about speed or the software development lifecycle and time. It’s also about ensuring that the process that is being followed is complete and valuable for today’s modern age. I think in a lot of cases, maybe lifting and shifting legacy processes into modern tools is not achieving positive outcomes. I think evaluating technology before culture is a big gap. And ensuring that we can support the evolution of culture with technology focused outcomes is key.

And I would say that most organizations with a framing on governance should have a benchmark of success for not only what their organizational standard is, but what their regulators require and what their shareholders, their stakeholders, and their board are looking for as well with regards to outcomes. Now technology shouldn’t operate in isolation within a tech team and just for, you know, the use of delivering software. It should have contributions in a measurable way to business outcomes. And drawing those lines in a measurable way is probably the biggest area of focus that I see for growth with organizational leaders this year and going into next year as well.

Henry Suryawirawan: Right. Measured approach. Make sure you have like the measurement, what you wanna achieve, right? So I think that’s a very good tips.

[00:55:13] Why Should Engineering Leaders Build an AI Strategy Before Choosing Technology?

Henry Suryawirawan: So Andrew, we have discussed a lot, right? if there’s something that you want to give to the engineering leaders out there that you think they have not been thinking about. Maybe seeing from your customers, you know, past successes and things like that. Maybe any things that you wanna convey to them?

Andrew Haschka: Yeah, absolutely. I would say it’s often much more productive to invoke change and improvement when you leverage an external vendor like GitLab to help in your path for improvement and success. It’s often recommended that making change internally is hard. And so do reach out to GitLab, maybe ask for a value stream or productivity assessment, a software delivery and supply chain security evaluation. Maybe build out that propensity matrix and return on investment view in a measured way for adoption of agentic AI and agentic flow-based governance. And really ensure that you build your strategy before you choose to adopt the technology so that you can have a measured approach, to have further investment and growth of your team going forward. So yeah, we’d love to collaborate with, you know, everyone who is listening to this session. You know, do reach out and, you know, I think the journey that we can build together is only gonna be more successful through collaboration.

Henry Suryawirawan: Yeah, so, sometimes definitely changing internally is hard, right? Especially when you have a few people not able to align with each other. So yeah, sometimes having vendors can definitely help, especially if you don’t have that expertise. So I think if you are interested in GitLab, make sure you check them out. So I think they have quite a very solid, I would say, consolidated platform, not just for your, you know, typical CI/CD aspects, right? But they have something that goes beyond that at the moment, right? So I think do check them out.

[00:57:15] 3 Tech Lead Wisdom

Henry Suryawirawan: So Andrew, as we wrap up our conversation, I only have one last question for you. I call this the three technical leadership wisdom. Just think of it like advice you wanna give the listeners to close our conversation. Maybe if you can share today, that would be great.

Andrew Haschka: Absolutely. We often get these requests from engineering leaders and CTOs and CIOs as well. And I would typically start with measure the whole system, not just the parts you can see. You know, the AI paradox is a systems problem. Leaders who instrument their entire value stream and act on what it tells them will consistently outpace those who optimize individual stages in isolation.

The second point is governance as a competitive advantage. In the agentic era, organizations that move fastest sustainably are those that have invested in the accountability from infrastructure to govern agents, recover from failures gracefully, and maintain stakeholder trust at speed.

And the third one is protect that culture. It’s what makes great engineering possible. AI changes how software is built, but it doesn’t change why the best engineers come to work. So you need to give your people the highest judgment work, the design, the architecture, the critical evaluation, and they will build things that AI cannot.

Henry Suryawirawan: Lovely. So again, like emphasizing the culture, again, I really think it’s still crucial, right? Not just using AI to actually, you know, just remove all developers that you have in the company. So, Andrew, if people would love to connect with you, ask you more questions, maybe beyond this recording, any place where they can find you online?

Andrew Haschka: Yeah, of course! You can definitely connect with me on LinkedIn. More than happy to field any inquiries. And I’m really keen to collaborate with everyone on their journey going forward.

Henry Suryawirawan: Right. Thanks again for your time today. So thanks for sharing all those insights that I think are really important for those leaders out there who are embarking their AI transformation journey.

Andrew Haschka: Fantastic. Thank you so much.

– End –