DevOps AI Search-First UI

The current trend to chatbot everything is a step backwards toward the 80’s command-line experience. User’s have mostly learned to use gestures, voice, and text. The chatbot experience reverts the user to a text-only or text-priority behavior. Along with an outmoded experience, users have had a plethora of failed  chatbot experiences from misinformation, to failed responses to completely missing the point of a request. This has been the experience from phonebots, chat windows, and voice assistants. Add to these inconveniences the fact that AI agents can introduce bias and spread biased information based on training data, while users often remain unaware of data storage practices and struggle with overwhelming information blocks that can’t be refined or scoped to their specific needs.

Forward looking interfaces need to break the sidewindow chatbot experience. When users already have a large window in front of them, forcing them into two separate interaction points—even if they are semi-linked—creates confusion and extra effort. Prompt engineering can be complicated, requiring heavy cognitive load before the agent is actually helpful. Users may need contextual onboarding and auto-formatting assistance for proper prompt configuration and help understanding why prompts failed to reduce the cognitive burden of prompt engineering. One further aid, users should be able to manually refine data scope and control information density rather than being forced to accept whatever the AI returns.

Users have been trained by search, assistive features, voice, and gestures to accomplish tasks naturally. AI interfaces need built-in flexibility to support how users want to work, incorporating the future evolution toward more voice interaction, natural gestures, and natural language processing—potentially even enabling smaller, contextual displays rather than forcing users into large, overwhelming interface windows. For example, if a user wants to circle something or remove an element, a quick gesture should enable that process.

In the early days, search radically changed how we interacted with the web. Prior to that, people had to hunt through directory structures or memorize exact addresses. Google transformed the web from a navigation nightmare into an instant discovery tool. This project applied that same transformation to enterprise operations, asking: how do we help operators achieve their goals more promptly? Guiding users from the abstraction of multiple navigation points to a specific query is the goal of flattening the information structure.

Along with this nested navigation paradigm came the composite behavior of adding a chat window on top of an already complex UI without meaningful integration between the interfaces. This means that the UI goes from deeply nested nav to also having a persistent overlay that the user must work around or within to accomplish their needs. This defeats the benefit of a smarter workflow. It further compartmentalizes the UI.

What the users need is for the interface to mold to their needs rather than forcing them to conform to the UI. Through stakeholder interviews, we discovered that operators managing customer environments already know what they need to accomplish when they come to DevOps. They’re primarily problem-hunting and rarely need to focus on healthy environments. When we demonstrated the search-first behavior, the operators immediately explained that a natural language query field matched exactly how they think about their work. They would typically enter queries like: ‘show me all [customer name] environments with high CPU usage’ or ‘environments in distressed state with DR enabled.’ For broader monitoring needs, queries like ‘system overview’ or ‘all environments status’ would provide comprehensive dashboards. With this they could compound the filters to define the information they need now. Then the UI presents the user with exactly what they need, how they need it, and when. 

Leaping from an overloaded navigational system to an economic, lean search-first design was a large undertaking that would require a vast amount of time and planning. Therefore, the stakeholders and I needed to devise a phased rollout plan focusing on the most critical aspect of this new UI—natural language search. Keep in mind that, throughout all phases, the existing navigational structure would remain fully functional, ensuring operational continuity and providing familiar fallbacks for users during the transition.

Natural language search needed to be phase one because the system needed to begin collecting user interaction data and learning query patterns. It had to work alongside the current APIs and UI without disrupting existing workflows, while gradually becoming the primary interaction method.

Phase two would introduce agents using a Model Context Protocol (MCP) approach that would further guide system training and assist users. Individual agents would handle specific tasks while a central controller would distribute requirements and coordinate responses.

Phase three would leverage agentic insights to build more intelligent data structures and responses.

Phase four would implement dynamic page generation based on user queries and context.

Finally, assuming successful adoption, phase five would systematically remove less important static pages as users migrated to the search-first workflow.