Confidential Enterprise Software
2026
AI experience design in Enterprise Workflows
Exploring how contextual, task-oriented AI agents can simplify complex enterprise workflows by moving beyond conversational chatbots toward workflow-aware guidance, actionable recommendations and human-controlled assistance.
Overview
Beyond the chatbot: designing AI that understands the task
Enterprise software often requires users to navigate interconnected systems, interpret technical information and make decisions with significant operational consequences. Introducing AI into these environments creates an opportunity to simplify the experience—but only when the assistance is relevant, contextual and trustworthy.
I explored how contextual and agentic assistance could help users move beyond finding answers toward completing tasks—with relevant guidance, actionable next steps and appropriate human control. The work progressed from competitive research and workflow analysis to opportunity mapping, user-flow design and AI-agent prototyping. The design question: How might we move from an AI that answers questions to an AI that helps users make progress within their actual workflow?
Problem Space: An answer is not the same as task completion
Enterprise users often work across dashboards, configuration interfaces, documentation and diagnostic information. Even when the required information is available, users may still need to connect the dots, decide which steps matter and determine what to do next. A conventional chatbot can make information easier to access, but it may leave the hardest part of the task to the user: translating an answer into action. This created three areas worth exploring:
01 — Context: Can assistance understand what the user is trying to accomplish and respond in relation to the current task?
02 — Actionability: Can the system guide users through a workflow instead of leaving them to interpret an answer and determine the next step independently?
03 — Trust and control: Can recommendations be explained, verified and acted upon without obscuring important decisions or giving the system unchecked authority?
The opportunity was not simply to add a conversational interface. It was to rethink how assistance fits into the user's workflow.
Competitive Research
Understanding the patterns already in the market
I began with secondary research and a competitive analysis of AI assistants and agents in enterprise infrastructure and data-resilience software. Rather than evaluating products only by their AI capabilities, I examined how those capabilities translated into user interactions. I looked at the tasks supported, the interaction patterns, deployment considerations, trust mechanisms and the limitations or opportunities suggested by each approach.
The research helped organise the landscape into three broad patterns:
These patterns revealed a useful distinction: an assistant can help users understand something, guide them through a task or proactively draw attention to something that needs their attention. The design opportunity was to explore how these capabilities could work together around user intent rather than treating chat as the default interface for every problem.
Research artifact
I created a structured competitor comparison to evaluate products against consistent criteria as shown below. The purpose of the table is to demonstrate the analytical framework—not to reproduce the full research dataset.
Workflow Analysis
Finding opportunities inside the workflow
Competitive research showed me how enterprise AI experiences were being approached across the market. The next step was to examine where assistance could create value within the user's actual journey.
I broke workflows into their constituent steps and considered the user's goal, the information required, decision points, potential errors and unnecessary effort. I then looked for opportunities to simplify the experience through interface improvements, predictable automation or contextual AI.
This distinction mattered. A clearer interface can solve some problems more effectively than an AI feature. Repetitive, predictable actions may be better handled through automation. AI becomes more relevant when users need help interpreting context, connecting information or deciding what to investigate next. I used this analysis to explore three opportunity areas:
Setup and configuration: Help users understand unfamiliar settings, identify missing information and navigate complex configuration tasks.
Troubleshooting and resolution: Help users make sense of available evidence, investigate potential causes and progress through a logical sequence of next steps.
Dashboard interpretation: Help users understand significant information, interpret patterns and identify relevant actions instead of leaving them with data alone.
Opportunity-mapping artefact
The table below illustrates the kind of reasoning used in the analysis. This is an independently written example using a fictional enterprise product, not an extract from the original internal analysis. The aim was to connect each potential AI intervention to a specific user need.
Solution Direction
From answering questions to enabling progress
The emerging design direction was a contextual, task-oriented assistant that could adapt its support to the user's goal. Instead of making users translate a generic answer into a series of actions, the proposed experience explored how an agent could help users understand their current situation, surface relevant information, recommend a next step and support the transition to action. I explored three complementary interaction modes:
These modes are complementary rather than interchangeable. A user may first need an explanation, then guided investigation and finally a recommendation. The interface should support that progression without forcing every interaction into an open-ended conversation. The concept also needed to account for enterprise constraints. Users should be able to distinguish available evidence from inferred explanations, understand the basis of a recommendation and retain control over consequential actions. The result was a design direction centred on context, guidance and actionable assistance—not conversation for its own sake.
Interaction Design & Prototyping
Translating the concept into interactions
I translated the opportunity areas into user flows, interface concepts and AI-agent prototypes. The design exploration focused on how assistance would enter a task, what contextual information it would use, how it would present guidance and how users could review the proposed next step.
I explored how conversational interactions could work alongside structured interface elements and task-oriented guidance. The intent was to make the experience feel connected to the work users were doing, rather than like a separate chatbot that required them to start over with every question. I also used Claude Code to build AI-agent prototypes to demonstrate and explore the interaction concepts. Across these explorations, I considered several design principles:
Context before conversation: Relate assistance to the user's current goal.
Evidence before recommendation: Make the basis of a suggestion understandable.
Guidance before execution: Help users understand what will happen next.
Control over consequential actions: Preserve opportunities for review and confirmation.
Honesty about uncertainty: Communicate when information is incomplete or an answer cannot be verified.
These principles informed the concepts and prototypes; they should not be interpreted as claims that every capability was implemented in a production system.
Validation & Learnings
A design concept is a starting point, not proof
The project progressed through secondary research, competitive analysis, workflow mapping, opportunity analysis, user flows, interface concepts and AI-agent prototyping. User acceptance testing was planned but did not take place following organisational and team changes. The concepts therefore remained exploratory; their usability and effectiveness were not validated with representative users.
The next step would be to test the prototypes against realistic enterprise tasks and evaluate whether users could understand the guidance, assess the evidence behind recommendations and make informed decisions. I would also examine whether the experience reduced unnecessary effort, how users responded when information was incomplete and whether they felt sufficiently informed and in control before taking consequential actions. The project reinforced three lessons for designing AI in complex environments:
Reflection
The future is not just conversational AI. It is contextual assistance. This project shaped how I think about agentic experiences in enterprise products. The opportunity is not simply to make software capable of answering more questions, but to help people navigate complex work with less uncertainty and more confidence. That requires understanding the workflow before choosing the AI interaction, making recommendations understandable and preserving meaningful human control. Thoughtful AI UX begins with a simple question:
