Part 1 of 3 in our series AI Agents in Design.

What changes for designers when users stop clicking through screens and simply say what they want? Kumar Gaurav is Head of Product and Design at Verloop.io, an AI-powered customer experience platform, where AI agents now handle 92% of tier-1 support queries. Before that he designed at companies including American Express, Meesho, Byju’s and Toptal. He talks about designing for intent, where humans must stay in the loop, and why he thinks designs will soon keep learning after launch.

“We’ve moved from designing for clicks to designing for clarity of intent.”

How do AI agents shift UX design from interface-driven to intent-driven systems, and what does this mean for designers and product leaders?

The biggest shift I’ve seen with AI agents is that we’ve moved from designing for clicks to designing for clarity of intent. I recently worked on an AI voice assistant that could handle multi-step tasks like updating a customer’s address and rescheduling a delivery in a single interaction. Traditionally, that would mean navigating three different screens or calls. Now, the user simply states what they need, and the system orchestrates everything in the background.

For designers, this changes everything. We’re no longer obsessing over every button placement — we’re architecting conversation flows that understand context, know when to act autonomously, and know when to pause for confirmation. At Verloop, this meant designing for ambiguity itself. How do you handle it when a customer says “I want to cancel” but doesn’t specify what? Do you list all their active services, or do you ask clarifying questions? Getting that clarity of intent from users is incredibly nuanced, because people communicate messily, with incomplete thoughts and implied context.

The real design challenge becomes building systems that can gracefully handle uncertainty while maintaining user trust. We’re essentially designing the personality and decision-making logic of these agents, not just their interface. It’s like being a conversation designer, system architect and behavioural psychologist all at once — and honestly, it’s an area we’re still refining every day as AI capabilities evolve.

How can AI agents be embedded in product ecosystems to enhance user experience and also directly influence revenue and scalability?

Here’s the thing: AI agents aren’t just improving user experience, they’re fundamentally reshaping business economics. At Verloop, we automated 92% of tier-1 support queries, which freed up human teams to focus on complex cases requiring empathy and judgment. But the real win wasn’t just cost savings. Our human agents could now spend quality time on high-value customer retention conversations instead of password resets.

The revenue impact becomes even clearer when you look at lead qualification. I’ve worked on systems where AI agents could qualify and score leads in real time during conversations, immediately routing hot prospects to sales while nurturing others with personalised follow-ups. Instead of leads sitting in queues for days, sales teams were engaging with pre-warmed, context-rich prospects within minutes. That acceleration alone drove a 40% increase in conversion rates.

But here’s what most people miss: the real strategic value is in the compound effects. When AI handles routine tasks efficiently, it creates a data flywheel. Every interaction teaches the system more about user behaviour, which improves personalisation, which increases engagement, which generates more revenue per user. You’re not just scaling operations, you’re scaling intelligence. The agents become smarter about upselling, retention timing, and even predicting churn before it happens. That’s where the real scalability happens: when your AI agents start generating insights that drive strategic business decisions.

Which design frameworks or principles keep AI agents aligned with user needs and values, and where is human intervention necessary?

The framework I always come back to is what I call “human-in-the-loop design ethics”: building checkpoints where human judgment is non-negotiable. When you’re deploying agents in sensitive domains like finance or health, I stick to three core principles: explainability, progressive disclosure, and safety triggers.

In a recent generative AI automation for a bank’s chat support, we made sure the agent could point customers to precise policy documents using our RAG-based knowledge centre. Account balance queries started with quick summaries, expanding into transaction history only when customers asked for more detail. And any high-risk action like a fund transfer required explicit human confirmation with secure OTP flows.

At Verloop, this became really tangible when our AI was handling 92% of customer queries automatically. We had to build in human intervention points around sensitive issues: frustrated customers dealing with billing problems, or culturally sensitive interactions across different markets in India and Southeast Asia. Our AI might technically resolve queries, but we needed humans to catch emotional nuance and make sure customers didn’t feel dismissed or misunderstood.

“Even the smartest AI benefits from humans in the loop — not because AI can’t do the task, but because judgment, brand voice and cultural context are still uniquely human strengths.”

The most successful implementations aren’t where AI replaces humans entirely, but where we create seamless handoff systems that leverage both AI efficiency and human empathy.

As agents become more autonomous, what capabilities and mindsets should product and design teams cultivate, and how does it change the workflow across teams?

I think the biggest shift we need to make is moving from being creators to being orchestrators and quality curators. Design teams need to get really good at prompt engineering and iterative refinement — it’s like learning a new design language, where you’re having conversations with AI rather than just pushing pixels. The mindset shift is huge too: instead of “how do I make this?” it becomes “how do I guide this AI to explore possibilities I might not have considered?”

What’s been a game changer for me personally is being able to ask AI to code exact interactions and just play with them until I’m satisfied, with no back and forth with developers needed. I can literally say “make this button have a bouncy spring animation when clicked” and iterate on the feel in real time. Same with user flows: I can generate working prototypes of complex multi-step processes and test them immediately. It makes designing interactions really smooth, because there’s no friction between idea and execution.

What I’m seeing work really well is when teams develop a hybrid workflow where AI handles the heavy lifting of exploration, code generation and variation, while humans focus on strategic thinking, user empathy and the final creative judgment call. It’s actually making cross-functional collaboration stronger, because designers can rapidly prototype concepts to share with product managers and engineers, and we’re having richer conversations earlier, with working examples rather than static mock-ups.

How might AI agents or companions reshape the process from inception to deployment in the next 3–5 years?

Honestly, I think we’re looking at a complete transformation of the design-to-deployment pipeline. In the next three to five years, I see AI companions becoming like having a full creative team in your pocket. Imagine an AI that knows your design system inside out, understands your brand voice, and can instantly generate not just mock-ups but working code, user research insights and even A/B test variations, all from a single conversation.

The inception phase is going to be wild. Instead of starting with blank canvases, we’ll be having strategic conversations with AI that can pull market research, analyse competitor patterns and suggest user journey frameworks before we even open Figma. Then during development, these agents will be continuously optimising in the background: tweaking micro-interactions based on user behaviour, suggesting accessibility improvements, even updating designs automatically when your design tokens change.

What excites me most is the idea of “living designs”, where your AI companion is constantly learning from user interactions and suggesting refinements after launch. We’re moving from a linear handoff model to something more like a creative partner that never stops iterating. Deployment becomes less about “shipping and forgetting” and more about continuous evolution guided by both human intuition and AI insights. It’s going to compress timelines dramatically while actually improving quality — which honestly feels like magic from where we sit today.

Next in the series: Patrick Henz on AI agents and ethical decision-making.