What is the watsonx Code Assistant for Z?
IBM watsonx™ Code Assistant for Z is a generative AI-assisted product built to accelerate mainframe application modernization at lower cost and with less risk than today’s alternatives.

watsonx Code Assistant for Z provides an end-to-end application developer lifecycle that includes application discovery and analysis, automated code refactoring, COBOL to Java conversion, automated test generation and code explanation.

Developers can automatically refactor selected elements of an application and continue modernizing in COBOL, or selectively transform code to Java leveraging generative AI using a highly tuned state-of-the-art large language model.

watsonx Code Assistant for Z provides the flexibility to deploy its large language model either on-premises or as-a-service.

Deb the Junior COBOL Developer

Deb is a junior COBOL developer just a few months into her new role at a large insurance company. Eager to prove herself but often unsure where to start, she’s been handed a legacy application full of dense, undocumented code. She spends hours trying to understand it, toggling between outdated internal docs and hoping someone on her team has time to help. But senior devs are swamped, and she’s hesitant to ask again.

What she really wants is to feel like she can figure it out on her own.

The original code explanation experience

The original code explanation experience was where it all began my first opportunity to shape how developers like Deb would interact with AI in watsonx Code Assistant for Z. I designed a clean, side-panel interface that allowed users to view natural language explanations alongside COBOL code, without losing their place or breaking focus. It was about making complexity feel approachable, giving Deb the clarity she needed without overwhelming her.

At the time, there was no template for how this should look or feel. Every detail had to be intentional: where the explanation would live, how much space it should take, how the information would load, and how to make the AI output feel supportive, not intimidating. I focused on tone, structure, and readability because for a newer developer like Deb, trust starts with understanding.

This panel experience became the foundation for everything that followed. It gave users like Deb a sense of control and access, helping her move forward without constantly relying on a senior developer. As the experience evolved into chat-based explanations, the early design principles clarity, support, and empowerment, continued to anchor the product.

In May 2024, these efforts culminated at Think 2024 IBM in Boston, where I had the opportunity to lead the design of a special event demo prototype for watsonx Code Assistant for Z: Code Explanation. Built specifically to showcase how our AI could explain COBOL code in a way that’s clear, trustworthy, and grounded in real business needs, the demo was featured during two and a half days of industry discussion, innovation, and thought leadership.

Designing this wasn’t just about creating a side panel, it was about opening a door for Deb, and for developers like her, to make progress, build confidence, and feel seen in a system that can often feel overwhelming. That’s the work I’m most proud of.

Empowering the Next Step: Transitioning from Side Panel to Chat

As watsonx Code Assistant for Z evolved, so did our understanding of how developers like Deb needed to interact with AI. While the original side panel brought clarity and access, it became clear that users craved even more flexibility and fluidity, especially as their questions and workflows became more complex.

This led to the transition toward a chat-based interface.

For Deb, this shift meant even greater empowerment. She was no longer just reading what the AI chose to show, but she could drive the interaction, clarify misunderstandings, and tailor the support to match her real-time thought process. It turned watsonx Code Assistant for Z from a helpful tool into a trusted coding partner.

By giving Deb a voice in the experience, the chat interface honored her curiosity, her autonomy, and her growth as a developer. It reflected the same core design values we began with clarity, support, empowerment, only now made even more dynamic, personalized, and human.

As the Lead UX Designer for watsonx Code Assistant for Z: Code Explanation, I was responsible for leading the full user experience design for the transition from a side-panel interface to a dynamic, chat-based interaction model.

While product management owned the overall business strategy, I owned and executed the UX vision: creating interaction flows, designing conversation patterns, prototyping new experiences, and ensuring that every detail from tone to structure, supported a more intuitive and empowering user journey.

My focus was on designing a system where developers like Deb could interact naturally with AI, ask questions, dive deeper into explanations, and build confidence without feeling overwhelmed. I worked closely with cross-functional teams, providing the UX leadership needed to guide the visual and interaction design while maintaining a strong user-centered perspective throughout.

This transition was about more than just adding a chat window. It was about evolving the relationship between users and AI, ensuring that as the technology became more capable, it stayed grounded in clarity, support, and human trust.

Alternate version – COBOL Paragraph Filtering

The command menu dynamically narrows to show only logical units of code (paragraphs) relevant to the file Deb is currently working in. This simplifies scanning and eliminates noise. Deb can start typing @paragraph, triggering an autocomplete list that lets her zero in on the exact code unit she wants to explain or comment on. She no longer needs to scroll through generic line numbers or memorize labels.

This change significantly lowers the cognitive load, supports faster task completion, and helps Deb build confidence in her understanding of COBOL’s structure.

Why It's a Better UX

• Clarity: @paragraph is readable and intuitive. It’s easy to infer what Deb is targeting.

• Discoverability: The type-ahead interaction and menu suggestions help guide newer users.

• Consistency: It matches how COBOL is structured and how devs talk about their work (e.g., “check paragraph 740”).

• Scalability: This pattern can expand to support other code constructs (@step, @procedure, etc.) for other languages(JCL, PLI/I).

Alternate version B – Descriptive Helper Text

In this version, each command option includes a simple, human-readable description of its functionality. When Deb sees /comment, she no longer has to guess, she’s told exactly what it does in plain language.

This helps Deb understand the difference between commands at a glance, removing guesswork and improving efficiency.

Alternate version C – Syntax + Helper Text

In this enhanced version, Deb sees not only the description but also example syntax for how to use the command, such as:
/comment <code reference> [optional input]

This gives Deb the confidence to take action without needing external documentation. It reduces trial and error and makes the interface feel smarter and more responsive to her workflow.

Alternate version – Comments + Code + Line Numbers

This version displays the code with inline AI-generated comments and preserves line numbers, helping Deb understand where the changes are occurring in the broader file. It's great for users who need orientation within the original context, and it supports referencing or sharing code with others.

Alternate version – Comments Only

This version focuses exclusively on the generated documentation, stripping away the code and showing only the inserted comments. It offers a minimal, clutter-free view that’s great for quick reviews, especially if Deb already knows the structure of the code but wants to verify what the AI produced.

Alternate version – Comments + Line Numbers (No Code)

This option shows only the comments alongside corresponding line numbers, helping Deb quickly assess which parts of the code are being annotated without rereading the logic itself. It’s ideal for high-level auditing or scenarios where visibility into structure is more important than detail.

Slash Document Experience: Cupcake

When we began exploring how to bring AI-assisted documentation into the hands of COBOL developers, we knew we had to start small and with something simple, useful, and lovable. This experience was the result of many iterations, cross-functional conversations, and hands-on testing. It became our “cupcake delivery” the first delightful slice of a much larger vision that’s still evolving(cupcake, birthday cake, and then wedding cake). The cupcake is a small, delightful first delivery that provides real user value; the birthday cake builds on it with more features and refinement; and the wedding cake is the full, end-to-end experience.

Our primary persona, Deb, is a junior COBOL developer. She’s bright, capable, and still learning. Like many new developers, she often hesitates when reading legacy code, unsure of its intent. Instead of interrupting a senior developer or combing through outdated documentation, we wanted to give her a lightweight, intuitive way to learn and move forward without ever leaving her editor.

The slash document experience gives Deb the ability to generate documentation or code comments either for the full COBOL file or for a specific paragraph via the code lens. By simply typing /document into the chat, Deb can trigger AI to generate lightweight documentation summaries across her file. She can also use the inline code lens to document a specific logical unit. From there, she chooses what to insert the documentation as a comment or she can copy it to the clipboard.

This is distinct from /explain, which produces deeper, more detailed summaries that often require more time to read and digest. Developer feedback confirmed that /document plays a critical role in supporting fast, focused workflows, which is why our WCA4Z team owns it directly (while our parent team WCA focuses more on explanation).

I led the UX design for this experience from concept through delivery, shaping every tooltip, flow, button, and fallback message to build clarity and trust. We accounted for error states, inline previews, confirmation messages, and even a progression tracker, a subtle but critical addition that lets users know how much of a large file has been documented (directly informed by real user pain points).

This experience wasn’t meant to be everything. It was our cupcake: small, thoughtful, and genuinely useful. It gave users a glimpse of what’s possible and gave our team the insight we needed to take the next step forward.

Top

Featured work

IBM z/OS Cloud Broker

EDT: Co-creator collaboration 2021 program

Zones enablement portal

State Farm & IBM EDT workshop