8 minutes to read With insights from... Andrija Ljubojevic Principal Software Engineering Consultant andrija.ljubojevic@zuehlke.com AI-augmented software engineering promises to improve productivity, boost efficiencies, enhance code quality, and more. It's no surprise then that development teams have jumped on the opportunity. GenAI tools, harnessing decades of collective knowledge, are already changing the way software engineers work.But to realise the full potential of AI in software engineering, GenAI needs to be adopted in a structured, compliant, and measurable way. Navigating this complexity and introducing GenAI into a software development setting at scale is not easy. It's something we’ve been exploring, discussing, and implementing with clients and service providers over the past year. In doing so, we’ve discovered some key learnings about how to harness the combined strengths of humans and machines in software development. Here's what our research, scaled experiments, and hands-on experience shows... GenAI for software development: promise and pitfalls Any LLM GenAI tool worth its salt can produce a whole lot of content, fast. But is that content useful? That’s the billion-dollar question in software development. Pumping out tons of code quickly can be incredibly useful. Producing lots of bad code is the polar opposite.GitHub Copilot, for example, suggests its users have seen 55% efficiency gains, while other major studies put that figure at around 30%. BAIN research suggests a fairly vague 15-40% time saving in enterprise code generation. And Google, whose Gemini tools are now rolling out more broadly, says over 25% of its own code is now AI-generated. (Of course, writing code is only one part of the lifecycle too. More on that later).The problem? These claims are open to interpretation. What is a 55% efficiency gain? If you’re more efficient in generating lines and lines of code, but that code isn’t efficient itself, what good is it? If you gain raw output speed, what do you lose in security, accuracy, and general oversight? And does better code quality and coding efficiency translate into a better software product – and greater value to your users?There are also notable differences between the greenfield cases – which most of these publicly available stats apply to – and refactoring use cases where AI has made less of an impact to date.Ultimately, you’ll need to determine what value creation looks like in the context of your own software engineering processes. This is a group effort, so engaging teams from the outset is key to progress. Here’s what to expect... Key considerations for bringing GenAI into software engineering Some of the key issues you need to navigate when scaling GenAI adoption across software teams include: Diverse stakeholder groups & emotions Diverse stakeholder groups & emotions GenAI is a very triggering topic for people. Some individuals will be mistrustful and skeptical about the value of GenAI output. Others will be concerned about what all this means for their professional future (see ‘Will AI replace software engineers?’). Then there are the passionate advocates with sky-high expectations. The picture can look very different from one group or stakeholder to the next. A nuanced, people-first approach is essential for building and maintaining trust. Organisational resistance Organisational resistance As with any new technology or process, GenAI adoption might be impacted or slowed by organisational inertia. Stakeholders are often wary of high-promise solutions. If you don’t deploy GenAI in a planned, purposeful way and it misses the mark, the wider organisation will likely double down on its hesitant stance. Running before you can walk Running before you can walk If you deploy a tool without a plan, and its abilities are too broad, you won’t be able to monitor its success or steer it towards more targeted uses. At the end of the day, GenAI is just another tool, and the usual rules around tool adoption apply: it needs to deliver the relevant and measurable results that will motivate people to use it more and more. Managing these issues requires a structured deployment process that delivers results you can measure. With governance to ensure you’re staying true to your policies and values in a rapidly evolving AI landscape.‘Tech adoption requires change, and change is about people. Many engineers will be skeptical about how GenAI tools can help them. However, our projects consistently demonstrate that, with hands-on experience and expert guidance, engineers increasingly recognise the value these tools bring to their work’. A structured approach for deploying GenAI in software teams Our findings show that the best results come from starting small to build success stories the organisation can grow and build upon.Together, our client work, a long heritage in software engineering, and our collaborations with Microsoft-owned GitHub Copilot suggest that GenAI adoption benefits from an approach that:Starts with a pilot programme and a phased approachManages expectations and addresses concerns up frontEstablishes a governance framework and reinforces accountabilityGathers qualitative and quantitative insights to enable steering and monitoringFocuses on skills development and continuous improvementFosters knowledge exchange across company structuresTaking a structured approach like this means you’ll be able to deploy GenAI solutions as part of a framework that can scale sustainably. 5 essential steps in your GenAI pilot programme In our experience, the following five steps are key for starting your GenAI pilot and measuring the impact within your software engineering teams. Graphic: A cyclical framework for piloting GenAI adoption, progressing through five key stages from initiation to continuous improvement. 1. Kickoff 1. Kickoff Defining your goal: The initial phase is about aligning expectations, defining the pilot’s aim, and identifying the participating team. This includes clearly outlining your goals, resources, and constraints. And addressing the essential question: what would success look like in your environment or context? 2. Discovery 2. Discovery Making it tangible and relevant: This phase assesses the maturity of your technical environment and software development processes to ensure the right rollout and use cases for your organisation. This is where you’ll identify potential pain points, alongside highlighting bespoke onboarding needs. The goal here is to make sure the pilot and onboarding are as relevant to everyday use cases as possible. That’s key to engaging people and ensuring their ongoing participation and motivation. This is where you also define the metrics you'll monitor throughout the pilot phase to inform how you’ll continue to move forward. 3. Onboarding 3. Onboarding Training your people: Sessions are hosted to train people on how to combine and apply their skills with tools relevant to their work. Here the emphasis is on the potential of tools within existing workflows. And to pinpoint the limitations and risks associated with their usage. It’s essential to reinforce engineers’ accountability for the code they create or commit. 4. Pilot programme 4. Pilot programme Gathering and shaping hands-on experience: A successful pilot programme is tight in scope and frequently measured. It involves one team working on one project, collecting insights to help you gauge impact and success. The aim is to accelerate adoption through positive, tailored use, not by proffering a vague ‘game-changer.’ Real experience comes from doing and shaping. 5. Iteration and improvement 5. Iteration and improvement Analysis and capturing learnings: Here, your collected data is analysed for useful, actionable insights. Objectively reporting on successes and failures (and clearly outlining why things went the way they did) can help develop a roadmap for scaling up GenAI use in the business. In that sense, this final phase is also the first step of the next iteration. Ultimately, this framework prevents GenAI from being used as a blunt instrument. Instead, it’s about sharpening your approach by finding answers to your hypotheses – Will GenAI improve efficiency? Does it increase team satisfaction? How does it influence the quality of deliverables? How does it impact team dynamics and communication? What use-cases really benefit from it and why? – so you focus on a select few value-driving use cases. ‘Design your pilot programme in a way that accelerates the pathway from first exposure to GenAI tools and the delivery of value your software engineers recognise and value. Growing your teams’ self-confidence with these tools will help foster a mindset shift in the way engineers use these tools to help solve problems’. How to embed GenAI in end-to-end software delivery Of course, GenAI adoption in software engineering is only one piece of the puzzle, prompting a wider evaluation of how GenAI can improve the effectiveness and efficiency of the entire software development lifecycle – from design and requirements engineering, to testing, monitoring, quality assurance, and operations. Why? Well, to use a sports analogy, integrating GenAI in software engineering tooling can give you a ‘better bat’, accelerating the software engineering lifecycle and increasing efficiency. But integrating GenAI into software engineering processes can change the game, profoundly altering the way software is developed as disciplines change or converge with the help of AI. And fully integrating LLMs into every stage of the development lifecycle produces a new game altogether. Here at Zühlke, our Cybernetic Delivery Method (CDM) is a structured approach for achieving exactly this. Our aim is to help businesses:increase productivity;manage the entire digital solution delivery lifecycle more effectively and efficiently; andmake the essential shift from doing things right to doing the right things. We’ll share more on this in the coming weeks and months, so watch this space. Taking a human-centred approach to GenAI adoption Codifying your approach to enterprise GenAI adoption within a clearly defined framework is key. But you also need supporting systems that ensure you can improve productivity in wider, more long-term ways.For one thing, enabling a pathway from discovery to value is going to need business-wide upskilling. Gartner suggests that 80% of the engineering workforce will need to upskill in areas like natural language prompt engineering, GenAI RAG skills, natural language algorithms, and data analytics.What’s more, the role of a software engineer is becoming more interdisciplinary. Not necessarily in the sense of needing broad, generalist ‘T-shaped’ skills, but in requiring the ability to deeply understand and solve problems within a specific domain or industry. This interdisciplinary approach means integrating knowledge from various fields to address domain-specific challenges effectively.For these reasons, GenAI adoption requires a holistic, people-first approach. It’s not just about tech and tooling, and your strategy needs to include everything from talent acquisition and development to communication and change management.Ultimately, embedding GenAI in the software lifecycle is about evolution not revolution. You need to take it step-by-step – safely, securely, and in a structured way. It’s the ‘long tail’ of these small steps and solutions that will drive the big wins. You might also like... Will AI replace software engineers? AI software development: what business leaders should know What does AI mean in the context of security? Reap the rewards of GenAI across the entire software lifecycle As product innovation complexity and market volatility grows, we’re helping businesses harness GenAI to extend the benefits of automation across multidisciplinary tasks and teams to create better digital products, faster. Contact the team to learn more. Contact the team