{Building Agentic AI} Navigating Global Reach: My Workday Journey and the Power of Clean Data
I’m Gerhard, a Software Engineering Manager at Workday Dublin. My journey into technology began in primary school when I visited family who had a black and white computer with games on it. After playing just one of those games, I was hooked. This experience drew me into the world of computers and engineering, initially driven by a desire to understand how games worked and potentially build them. That interest quickly broadened into programming, architecture, design, DevOps, cloud, and the full spectrum of technology, from infrastructure and platforms all the way through to software development.
Along the way, I accidentally found myself in development leadership roles a few times. Each time, I’d decide I didn’t like it and would return to more architecture-focused positions, but I kept ending up back in leadership — it must have been a sign! Eventually, I transitioned into hybrid architecture and further leadership roles. This path ultimately led me to the role I held in 2017 just before joining Workday. At that point, that my wife, an Emergency Medicine Consultant, was offered a job in Ireland, prompting our move and my subsequent joining of Workday. We had recently married, and I remember thinking, “Great, move to Ireland? Yes, let’s go.” We moved from Cape Town, South Africa and it was quite a significant move, indeed! Nearly a decade into my journey at Workday, I’m still inspired by how our core values — like innovation — come to life every day.
From Deployment to Globalisation: My Team’s Impact
My team, Global Deployments and Adoption within Engineering Services, supports all Workday products and engineering teams. We ensure new features are easily implemented and widely adopted by customers. Initially, I focused on deployment tooling and best practices, building tools to streamline data migration and configuration. I then transitioned to the adoption side, developing frameworks to define product value, articulate business outcomes, and measure successful feature uptake. This helps Workday understand product performance and identify areas for improvement.
More recently, I moved into the globalisation space. This encompasses both internationalisation and localisation. Internationalisation is about building products in a way that allows them to be sold and function across the world in different countries, regions, cultures, and languages. This includes handling nuances like dates, currency formats, phone number verification, and address lookups. Localisation, on the other hand, is ensuring users can use Workday in their language of preference. This is a really big deal for many end-users globally, especially those who aren’t native English speakers or fluent in English as a second language. It can be incredibly challenging to perform complex work on a system if you struggle to understand the language. This is particularly true in certain markets, where English isn’t widely spoken. As Workday grows internationally, supporting local languages, laws, and requirements becomes just as critical as having great features.
The Globalisation Platform: Powering Translations
Within the globalisation organisation, we predominantly have two sides. There’s a globalisation product side, focused on program and project management, engaging with internal and external stakeholders and acquisitions to understand needs for new regions or features. This team helps define what content needs to be translated. Then, there’s our engineering side, the globalisation platform. My team’s realm starts with extracting what needs translating. We have processes to extract text from various source code systems, whether it’s the core Workday platform built on Xpresso or more traditional bespoke systems. We clean up, reformat, and refactor this extracted text.
We then have custom-built data pipelines that handle this information in different ways. This includes machine translation, connecting to third-party vendors, or sending the content to our main Translation Management team where human translators can work on it for really high-quality translations. Once these processes are complete, the translated data needs to be integrated back into the source code so systems or features can be built and shipped to users in their preferred languages.
Enabling Global Teams and Customers
My team isn’t just involved in direct translation and localisation; we also build the tools that enable these processes. This includes providing internal back-end support for Workday teams, sales and marketing, who are creating admin guides or documentation. For teams building new features, we recognise that not everyone understands internationalisation well. We are a growing company and new teams constantly form and engineers join who may not have previous experience with this. So, we invest significantly in education, setting standards, explaining expectations, and teaching best practices. A big investment for us is growing our capability to provide frameworks and tools that allow teams to self-serve. The ideal end-state is for teams to be able to assess what’s needed for a specific country or region and tell us, rather than us relying on extensive back-and-forth conversations.
Beyond this internal support, we also develop customer-facing products. Workday provides many features, functions and data, including out-of-the-box translations. However, customers also create a vast amount of their own data within their tenants, which we don’t touch; it’s theirs to control. This tenant-specific data often needs to be translated. If you consider Workday as a multinational organisation, employees in one country might create content consumed by employees on another continent. This presents unique challenges, especially since the users creating this content, like HCM administrators, are typically not linguists.
We aim to provide a hub for these users to manage translations for commonly used business objects. They can free-text translations if they have the capacity or internal linguists, or they can hook up to third-party providers or use machine translation. Looking ahead, a key focus for my team is integrating Workday’s own machine translation and Generative AI (GenAI) capabilities into this customer-facing translation hub. This is directly related to the growth of our agentic AI efforts. The goal is to offer robust translation capabilities directly within the Workday platform, reducing the reliance on external tools.
Our Technical Stack: From Traditional to Agentic AI
The technical stack we use reflects our varied responsibilities. We primarily use Python and also have Java in some settings. We interact with Xpresso, Workday’s core platform, and have react-based UIs.
As we move into the agentic AI, machine learning, and machine translation spaces, the tech stack shifts. Here, Python becomes even more prevalent, especially with tasks involving data processing and machine learning workflows. For agentic use cases, we currently use LangChain; it’s a Python library that aligns with the direction Workday is taking. We also leverage LangForge and LangGraph. These are more commercially oriented platforms that are vital for operating AI solutions at scale. They provide essential visibility into how agents are communicating, how they’re using tokens, and what their behaviour is, allowing us to gather metrics. When building new systems, especially non-deterministic ones like agents, being able to “peek under the hood” to understand what’s happening is crucial for achieving the required quality bar. These platforms also help monitor behaviour in production and identify issues.
A core principle guiding our technical approach is leveraging Workday’s existing platforms whenever possible. The power of a large company like Workday is that many teams have already built awesome platforms and capabilities. We make a conscious effort not to reinvent the wheel. For example, Workday already has platforms for storing models and executing against them. We leverage these existing capabilities. My team isn’t an ML Ops team; our focus is on utilising these platforms and technologies efficiently and understanding how to apply them.
Our work involves figuring out the best and most efficient ways to use these evolving platforms and technologies. The capabilities we build then either support other active teams with their localisation, internationalisation, deployment, or adoption needs, or they are put directly into the hands of our users to make them more efficient. For instance, translation is difficult, especially for companies operating in twenty different languages. Most people aren’t fluent in that many languages. With our style guides and available data, we can provide capabilities that allow users to get workable translations quickly, potentially removing the need to employ external translators.
The Power of Workday’s Unique, Clean Data and AI
This is where AI software engineers come in — helping to build these capabilities to perform translations directly within the Workday platform, reducing reliance on third-party tools or manual free-text entry. The key here is the ability to leverage Workday’s unique data. While off-the-shelf translation APIs exist and can be good for general translation, they lack access to Workday’s proprietary assets. We have about twenty years’ worth of Workday-related translations that are high-quality and curated by specialist linguists across various fields, areas, and countries. This data, along with Workday’s glossaries and terminologies is critical for the best output.
This means that while a public model might provide a technically correct translation from a language perspective, it will often use the wrong words or terminology because in Workday, we have specific names for things. If a user needs to find a button called “XYZ” in Workday but the translation tells them to look for “ABC”, they’re going to be stuck. Adjusting to a new system’s terminology is hard enough even in your native language. Our curated data, which includes style guides and how we explain things at Workday gives us a specialist ability to deliver a much better quality experience to users.
The importance of having clean data cannot be overstated. There’s an old adage in software engineering: “garbage in, garbage out”. No matter how good your system is, if the input data is poor, the outcome will be poor. This applies as much to new AI systems as it does to simple maths. Systems that weren’t conceived as a single source of truth pose significant challenges. You have to find the data across multiple disparate systems, extract it from different formats, load it, and process it just to get it into one place you can work with. This is time-consuming, requires varied technical skills, and can be hampered by proprietary systems that weren’t designed for data export.
Workday avoids these challenges because the data is centralised. Whether it’s customer-tenanted translations or the translations Workday provides out-of-the-box, the data resides in the same spaces within each tenant and is readily accessible. This clean, centralised data is a massive advantage. Our linguists have been curating this data for a long time; it’s a highly specialist, human-in-the-loop process that was the only way to achieve proper quality with intimate Workday knowledge. The cleanliness and specificity of our data are fundamental to building effective AI solutions for localisation.
The potential benefits of leveraging AI in this space are significant. In the past, it was a complex, multi-step process requiring planning months in advance and was slow and costly. With new AI capabilities, we can drastically reduce the time-to-market and lead time needed for translations.
The Future: AI-Driven Globalisation and Market Expansion
Workday uses a concept of language tiers. Tier 1 languages, for our priority markets, require all products and features to work seamlessly with flawless internationalisation. As you move to lower tiers, where the user base might be limited, it’s not always cost-effective to translate everything to the same quality standard. This is where machine translations become invaluable. AI is making these translations much quicker, better and more readily serviceable at a reasonable cost, making globalisation viable for more regions. This, in turn, opens up more markets for Workday.
In summary, our team’s technical work spans both traditional software engineering for our core globalisation platform and cutting-edge AI development to integrate new translation capabilities. By strategically leveraging established technologies, alongside emerging AI frameworks, and Workday’s internal platforms, we aim to build scalable, efficient, and high-quality globalisation solutions. Our unique advantage lies in working with Workday’s specific data and integrating our capabilities into the existing ecosystem to support both internal teams and external customers. In essence, Workday’s commitment to globalisation, combined with our unique, high-quality data and the emerging power of AI, positions us to deliver highly accurate and efficient translation capabilities, unlocking the full value of Workday for users worldwide. And as technology evolves, so do we — continuously pushing boundaries, refining our craft, and moving forever forward to meet the needs of a global workforce.
A brighter work day is just around the corner. ☀️ Explore career opportunities here. For more #WDAYLife content, follow us on Instagram, Facebook, Twitter, and LinkedIn.
