Leena AI
From Chatbot to Agentic Colleague
I led the interface evolution behind Leena AI's assistant, from a rule-based FAQ bot to a full agentic AI colleague, across five product generations: rule-based chatbot, NLP-based chatbot, generative assistant, and finally agentic assistant.
Role
Lead Product Designer
Timeline
Shipped to production across every generation
team
1 Designers & 1 PM
Scope
Mobile, Agentic AI, Conversational UX
Final Results
Reduced time to value by 48%
2x adoption with each iteration
1.3x user retention over previous experience
One design system scaled across 5 departments
Why this mattered
Two problems arrived together.
The first was trust: once the assistant started proposing autonomous actions, people didn't trust it enough to let it act.
The second was scope: the product had grown from a handful of FAQ use cases to fifteen, spanning HR, IT, Procurement, Sales and Finance, so one misread request now cost far more than it used to.
The problem, by stage
Stage | What was broken |
|---|---|
Rule-based chatbot | Decision trees, dead ends, no discoverability |
NLP-based chatbot | Couldn't hold context, users lost their place mid-workflow |
Generative assistant | Hallucination risk, inconsistent tone, unclear visibility into answers |
Agentic assistant | No visibility into what the assistant did, users afraid of losing control |
I inherited Chatterton, a rule-based chat integration, and led the interface through each generation that followed: two chatbot versions, a generative assistant and an agentic assistant.
What research surfaced
24 interviews, 1,200+ conversation logs and 85 usability sessions produced four personas and four recurring themes:
Discoverability. People didn't know what the assistant could do.
Scale. Context got lost once a request ran past a couple of turns.
Transparency. Confidence in an answer meant nothing without a reason behind it.
Control. The more the assistant could do on its own, the more people wanted a say before it acted.
I also mapped real conversations against what the user did and what the assistant did at the moment things broke.
Five breakdowns kept recurring: an active workflow being interrupted, unclear status after a workflow finished, new input interrupting a workflow, new input disrupting one already running, and users who couldn't answer what they were asked.
The decision that shaped everything after
Once the assistant could act and not just answer, conversation design stopped being enough. The interface had to account for itself: what it did, and why, every time. So I changed what I was designing for. I stopped asking how the assistant should talk and started asking what it owes the user after it acts. The next five changes follow from that.
What I changed, and why ↘
Guided chatbot
I replaced dead-end intent matching with guided prompts, fallback flows and structured forms. The product no longer depended on users getting every phrasing right.
A persistent workspace
Chat couldn't carry fifteen use cases on its own. I made a modular workspace the anchor and left chat as one entry point among several, instead of stretching chat to cover everything.
A system that scaled across departments
I built it modular enough that patterns proven in HR could extend to IT, Procurement, Sales and Finance without a rebuild each time.
Calibrated trust for the generative assistant
Scaffolded AI actions, fallback actions and lightweight feedback let people build trust turn by turn, not all at once.
Accountability for the agentic assistant
Action previews, per-action confidence indicators, step-by-step logs, and human-in-the-loop confirmation before anything irreversible happened.
Testing before trusting the design
Every major decision went through usability testing and A/B experiments instead of shipping on instinct. Testing confirmed three things: better discoverability cut repeated fallback conversations, structured repeat flows improved task completion over open-ended chat, and visible reasoning increased how often people let the assistant act on its own.
Results
Time to value fell 48% with Gen AI adoption.
Adoption doubled across HR, IT, Procurement, Sales and Finance.
User retention rose about 1.5x over the previous experience.
One design system scaled across five departments with no rebuild.
Conversations dropped 72% once guided flows replaced dead-end intent matching.
Adoption of autonomous AI grew about 16% month on month.
Looking back
As the assistant's autonomy grew, the real design problem moved from conversation design to accountability design. Confidence indicators and action previews weren't polish. They were what let people hand real decisions to the assistant instead of double-checking everything it did.
