Azure AI Foundry
Designing seamless workflows for developers
Problem and Project Overview
Azure AI Foundry is Microsoft’s developer platform where users discover, experiment with, and deploy AI models and agents. It unifies access to models from providers like OpenAI, Meta, and DeepSeek, alongside enterprise tools for monitoring, scaling, and compliance.
Despite strong adoption, the platform’s user experience was hampered by legacy design patterns and back-end architecture. Developers described the interface as cluttered, unintuitive, and fragmented. Common workflows—onboarding, documentation lookup, and model experimentation—were unnecessarily difficult.
Developers reported that core tasks frequently broke their focus.
Documentation flow:
-
A developer working in Foundry clicked “Help”, which is placed inconsistently across the platform.
-
The system redirected them to an external documentation site in a new browser window.
-
They had to search for relevant content in a different environment.
-
Returning to Foundry required mentally reconstructing their place and task.
This context switching was frustrating and slowed progress, particularly during onboarding when users were most vulnerable to dropping off.
Experimentation friction: Without a safe, low‑friction way to try models, developers often had to write and run code just to test hypotheses, making quick comparisons and learning by exploration unnecessarily difficult.
Solutions
As a product designer on the Foundry team, I focused on two initiatives that simplified onboarding, reduced friction, and accelerated experimentation:
-
AI Chat Agent – Embedded conversational assistant that replaced disruptive documentation flows and guided users in context.
-
Playgrounds – Interactive environments for prototyping and comparing AI models across modalities.
AI Chat Agent
The Chat Agent transformed help into a seamless, inline experience and reduced decision paralysis by anticipating user needs.
Key Features
-
Surfaced documentation and tutorials directly within Foundry with consistent, contextually relevent entry points.
-
Responded to natural‑language queries like “How do I deploy Model X?” with contextual answers.
-
Executed tasks such as provisioning or deploying models on the user’s behalf.
-
Provided onboarding guidance for new developers and troubleshooting for advanced users.
-
Contextual prompts proactively suggested example questions based on page context (e.g., on a model page: “Compare this model to alternatives?”; on a project homepage: “Deploy your first template?”; in Playgrounds: “Try this prompt with another model?”).
-
Designed in alignment with Microsoft’s Copilot UX design system standards (tone, layout, interaction patterns) to ensure consistency across Microsoft surfaces.
Impact
-
Dramatically reduced reliance on external documentation.
-
Shortened time to first deployment.
-
Anticipatory prompts lowered the barrier for users who might not know what to ask, increasing discoverability of key actions.
-
Reinforced Microsoft’s long‑term vision of an agent‑first UX strategy.
Model Playgrounds
Playgrounds created a low‑friction, code‑free environment for hands‑on experimentation and model comparison.
Key Features
-
Chat Playground for model comparison, parameter tuning, and iterative testing.
-
Multimodal Playgrounds for image, audio, and video experimentation.
-
Exportable experiments, enabling smooth transition from prototype to production (e.g., open in IDE or add to a project).
-
Prominent entry points from the homepage and the persistent navigation pane.
Impact
-
One of the platform’s most‑used areas, giving developers confidence in their choices.
-
Enabled Microsoft to fulfill commitments to surface new partner models quickly in a user‑friendly, discoverable space.
Design Challenges
-
Ambiguity from partners: Model providers (e.g., Meta, DeepSeek) often released specs late. The team iterated rapidly, designing Playgrounds flexible enough to accommodate uncertain model capabilities.
-
Branding trade‑offs: Multiple iterations weighed the benefits of tying to the Copilot brand (recognition) versus a platform‑specific identity (clarity, reduced confusion).