Founding designer: shaping the product and the AI together

COROS AI is an ontological coaching platform that helps professionals navigate moods, repair relationships, and take action when they’re stuck, built on the thesis that the transformation of human beings is the transformation of our language.

I joined COROS AI as the founding designer after a 2-month apprenticeship, drawn to their core mission: we’ve been working with machines for so long that we’ve forgotten how to work with each other. As someone entering the AI era through human-centered design, I saw an opportunity to shape how AI could restore human connection rather than replace it. The role went well beyond traditional UX/UI:

• Product design: three flagship features (Dimensions, Influences, Personality Slider)
• UX design: end-to-end UX/UI (onboarding, chat interface, settings)
• AI: prompt engineering, memory algorithm design, RAG optimization
• User research: user interviews, user testing
• Visual identity: logo, brand, and design system

The product itself is a conversation: users bring a real moment, and COROS coaches through it, on desktop and mobile.

Early users of COROS AI shared a consistent frustration: the AI was coaching too quickly without truly knowing them yet. While the insights were often thoughtful, guidance felt rushed, overly directive, and misaligned with users’ emotional readiness or personal context.

The problem: the AI coached before understanding users, so guidance felt rushed, generic, and misaligned with emotional readiness.

Before designing a solution, I analyzed existing AI coaching and companion products (Pi, Headspace, Claude, and others) to understand how they establish user context.

• Insight: most AI coaching and conversational AIs gather context through an onboarding.
• Caveat: this context is treated as static. It rarely updates as users’ situations evolve.

The solution wasn’t just introducing an onboarding flow; it was designing how the AI gathers and uses context to coach effectively. I created three core features that let users define their context upfront, then prompt engineered the AI to reference this information throughout coaching conversations.

Dimensions lets users declare which areas of life matter most to them: Work, Family, Self, Health, Belonging, Meaning, World.

How it works: users select their focus areas during onboarding. Throughout coaching conversations, the AI references these dimensions to anchor guidance in what’s personally relevant, while also identifying which dimension a current struggle relates to, helping users see how challenges map across their life and track growth in specific areas over time.

Influences lets users declare the thinkers, belief systems, or frameworks that shape their worldview, for example Brené Brown, Rumi, Stoicism, or Islamic values.

How it works: users select their influences during onboarding. The AI references them sparingly and strategically, only when a specific quote or teaching would significantly deepen a key coaching point.

The Personality Slider lets users control the AI’s coaching intensity by choosing among Supportive (patient, gentle, calm), Balanced (grounded, curious, discerning), and Provocative (candid, bold, perturbing) modes.

How it works: users select their preferred mode, which switches the entire AI prompt architecture:

• Supportive: “I hear you. How are you doing as you bring this up? What’s happening at work?”
• Balanced: “What really matters to you here?” or “I’m hearing a mood of overwhelm – does that feel right?”
• Provocative: “Are you going to do it or not?” or “If you don’t want to do anything about it, why are you here?”

The original design: a three-position slider spanning the three coaching modes.

Iteration 1: cutting supportive mode

I initially designed three modes (Supportive, Balanced, Provocative), but through testing and stakeholder alignment, we discovered that a coddling “supportive” approach contradicted the ontological coaching framework, which requires challenging limiting beliefs to drive growth.

Iteration 2: the slider wasn’t a slider

With Supportive removed, only two modes remained, but the control still looked like a slider, sliding between just two end states:

Alpha user testing revealed a confusion:

Result: the slider was redesigned as a Personality Toggle.

The shipped onboarding, end to end: welcome, a light warm-up, then the three context-gathering features (dimensions, influences, personality) working together as one system.

Alongside the flagship features, I built out the component language the product ships with: settings surfaces plus a full set of buttons, inputs, and action buttons specified across every size, intent, and state.

Settings surfaces

Component examples

Every component is specified as a full matrix: variants, intents, sizes, and interaction states, so engineering can build from a single source of truth.

For COROS, I led the brand and logo design end-to-end to translate a philosophical product vision into a coherent, premium visual identity. I shaped the logo direction, color system, and brand language to reflect COROS’s core stance: seriousness, care, and long-term commitment.

Brand guide

These decisions were collected into a complete brand guide documenting the logo system, colour, and typography.

I also designed the company’s investment pitch deck, extending the brand into fundraising materials.

Working at COROS was my first experience designing in a fast-moving startup, and it taught me how to operate under constraints I hadn’t faced before.

I learned to design alongside engineers, not in isolation. The product evolved quickly, often while being built. I had to involve engineers early, understanding what was hard, what was impossible, and where I could push. This meant fewer beautiful-but-unbuildable ideas and more solutions that actually shipped.

I learned to let go of perfection. We didn’t have time or resources to solve every edge case or polish every detail. I had to prioritize ruthlessly, focus on the core problem, define a realistic MVP, and make peace with compromises. Some of my favorite ideas didn’t make it. That was hard, but necessary.

I learned to move forward without all the answers. Requirements shifted. Direction changed. I didn’t always have complete clarity before making decisions. Instead of waiting, I proposed directions, tested quickly, and adjusted based on what I learned. I got more comfortable with ambiguity and faster at recovering when I was wrong.

I learned to stay grounded when everything else was shifting. What kept me anchored were the users. Their words, their confusion, their relief when something finally clicked. When debates got abstract or timelines got tight, I came back to what they actually said. That kept the work honest.

COROS taught me how to design under pressure, and how to stay focused on what matters when everything around you is moving fast.