Atomicwork, 2024
Designing the Atom Assistant Controls: Creating a Unified Hub for Agentic AI Configuration
team
Product Manager
Design manager
Frontend devs
Backend devs
Marketing team
AI team
My role
Qualitative Research
Conceptualization
Design
Dev handoff
UX writing
Duration
2 months
------o background and problem
This project aims to fix the issue of not having a single, easy-to-find place in the Atom (Assistant) UI for administrators to manage different settings. These settings control how Atom (Assistant) works, learns, and interacts, making it easier to customize and control the assistant.
------o but, What is atomicwork?
Atomicwork offers an Agentic Service Management solution that combines AI agents, a modern service platform, and automation engines to simplify enterprise support and service operations.
------o User want one unified experience for assistant control
What did it just do?”, “Why did it suggest this?”, or “Can I turn this off?” Without clear boundaries, AI can feel unpredictable or even intrusive. Teams also face friction when AI behaves differently across roles or workspaces. Admins want governance, while end users need flexibility.
Start here, settings page
AI controls solve this by giving teams the ability to fine-tune how the assistant works, define when it steps in, and decide who gets to manage those rules. Done right, this builds trust, reduces hesitation, and helps AI features land smoothly across different types of users.
An “Assistant” tab inside Settings (under organization) All org admins will be able to visit view and configure all tabs within this page
FOCUS AREA
1. Clear and flexible controls for users
People want to understand what the assistant is doing — and be able to change that. Giving them controls like toggles, logs, and modes helps build trust and lets them use the assistant in a way that fits their workflow.
2. Role-based access for better management
Not everyone in a company should control everything. With AI, it's important that admins set the rules (like what’s allowed across the org), while individual users can customize how they interact with the assistant.
Problem: “Who decides how the assistant works?”
Why it matters: Without clear governance, teams get messy access and behavior.
Problem: “Can I turn this off?”
Why it matters: Not all users want constant AI involvement.
Problem: “Why did it suggest this?”
Why it matters: Users skip AI help if it feels random.
Problem: “What did the assistant just do?”
Why it matters: Sudden actions confuse users and reduce trust.
Problem: “This AI is too aggressive / too passive.”
Why it matters: If it's too much or too little, users ignore it.
idea 1: Assistant settings into 4 major categories
The initial idea was to have a knowledge hub at one place and bring in skills(automated workflow templates around assistant) in assistant settings
Users would prefer having everything in one place, assuming it might make it easier to manage both knowledge and workflows for the assistant.
idea 2: Assistant setting and knowledge in data table format
organizes assistant settings and knowledge into a clear data table, to view, sort, and manage all items in one place.
Users would be overwhelmed by the table format since it offers little context, making it hard to understand and manage the assistant effectively.
idea 3: cONNECTED APPS FOLLOWED BY KNOWLEDGE SOURCES
linking multiple apps and data sources (like SharePoint, Confluence, Slack, Azure AD) to sync and centralize organizational knowledge, so the AI assistant can pull accurate info from all relevant places seamlessly.
Users would get overwhelmed by too many synced sources without clear organization, making it hard to find accurate and relevant answers quickly.
idea 4: 1 PAGE FOR assistant settings ✅
For v1, this would be the best solution giving enough control to customize assistant
and get familiar with new concept.
It offers just enough control and clarity for users to customize the assistant and get comfortable with how it works.
------o Assistant branding
Personalize Atom’s name, tone, and behavior to match your org—while IT keeps full control over what’s shared and how
------o small talk
Add utterances, variations and static responses for every small talk pair.
- Search across primary message of small talk and variations.
-After ingestion, querying tenant should prioritize small talk.
- If same utterance exists across small talk, Verified Answer, Service Items etc —> small talk gets prioritized.
------o Acronyms
Do you have any common acronyms that your office uses? Let’s make sure our assistant is included too! Define common abbreviations used by your employees.
How can you add these Acronyms - 3 ways
Manually from UI - create, update, delete
Detected acronyms during ingestions
Automatically add with both Key: Value pairs
Supported file types / syncs- manually uploaded pdfs, .docx, .ppt, [.csv, .xlsx] not in scope, confluence sync, sharepoint sync, notion sync (same as .docx)
------o REPLY TEMPLATES
Positive, Negative and Fallback message configuration
------o cONTENT RESTRICTION
Hardcoded set of topics. Can choose to Allow or Restrict on all tenants, default for these topics are ‘disabled’ state
------o trusted knowledge
Admins in the platform can feed Atom with enterprise knowledge like processes, troubleshooting steps, how-to's etc so Atom can answer employees IT related questions. However, a lot of this data is related to third-party tools used by the company and is available publicly. Feeding Atom with this data out of the box has several benefits
SIGNING OFF!
KEY TAKEAWAYS
This project was jam-packed with a lot of work, exploration and discussion with stakeholders. It feels almost impossible to boil down my learning, but if I had to,
here they are!
WHAT THIS PROJECT TAUGHT ME
Designing AI controls made me think beyond UI — about roles, responsibility, and how much control users actually need. I learned how clarity in copy, structure, and access can shape trust. This wasn’t just about building a feature; it was about making sure the assistant felt reliable and manageable, both now and as it evolves.
Learnings That’ll Stick
I realised design is as much about questions as it is about visuals. Being part of real discussions, trade-offs, and reviews helped me grow fast. Even as an intern, I saw how clear thinking and ownership can move ideas forward. It taught me to work smarter, communicate better, and stay flexible through iterations.
Made with <3 and caffeine xD