On AI Agents

As previously posted on the NBS blog.

As of February 2025, I'm designing AI agents that have reached over 41,000 daily users in less than a month. Here's what I've learned works best, written in plain language for founders and designers.

Quick responses matter. In the digital age, speed isn't just appreciated—it's expected. Users want instant feedback, and a delay can feel like an eternity. Show something immediately, even if it's just a loading indicator. Communicate during delays with messages like "Thinking…" or progress bars. Break long answers into digestible parts. Make it crystal clear when the AI has finished its task.

Start simple, grow naturally. The most successful AI designs evolve based on user needs. Launch with a streamlined feature set that solves the most pressing problems. Roll out advanced functionalities when users are ready. Provide subtle hints that nudge users toward discovering capabilities naturally. Let users learn by interacting—those "aha" moments foster deeper engagement.

Keep users in control. Empowerment is a cornerstone of trust. Allow users to see the entire interaction history. Signal clearly when a task is complete. Make it simple to revisit past answers or start fresh. Always include a "start over" option for those who want a clean slate.

Avoid too many settings. Overloading users with choices leads to decision paralysis. Set smart defaults that work for most users. Reveal settings progressively as users need them—not all at once.

Make your value clear. Many AI products fail because they don't communicate their benefits. Focus on outcomes users can achieve, not just the AI's capabilities. Instead of "Uses advanced NLP," say "Helps you write emails 3x faster." Show real examples that resonate with your target users. Track metrics that reflect actual user success rather than vanity metrics.

Plan your architecture for growth. Build components that can be easily upgraded as your AI capabilities evolve. Implement comprehensive logging and monitoring to catch issues early. Design with unit economics in mind—AI inference costs can scale non-linearly with user growth.

Consider the user experience at scale. As your user base expands, edge cases become common. Build systems to systematically identify and address them. Foster user communities that support one another and provide valuable feedback for product development.

Let's build something exceptional together. Our startup studio's track record of launching successful AI products, combined with your vision, can create remarkable outcomes.

Key takeaways: Launch with a single, well-executed core feature that solves a specific problem. Implement quick responses and loading states today—users appreciate being acknowledged. Track how many users actually achieve their goals rather than just usage numbers. Partner with those who have built what you're building—the right partnership can cut your time-to-market in half.