Sections:

Designing an Agentic AI Platform
from the Ground Up

From undefined concept to scalable enterprise platform — leading design solo through launch, then steering a team of eight through a full platform expansion.

Section 01 — Context

A new kind of product in a familiar space

In 2024, Dialpad bet on agentic AI — not assistive features, but truly autonomous voice and chat agents capable of taking action across enterprise systems. The concept of an AI agent inside a real-time communications platform was largely undefined. No pattern existed to borrow from. The business need was clear; the design problem was not.

I was brought in as the only designer on a 50-person cross-functional task force to figure it out.

Fig. 01 — AI Agents entered a contact center with established routing, tooling, and human workflows already in place. The autonomous path was entirely new — and entirely undefined.

Section 02 — 0 → 1

Defining what an agent actually is

Before any experience could be designed, we had to answer a bigger question: what does it mean for software to act autonomously on behalf of a human — and how do we keep that human meaningfully in control?

I worked across product and engineering to define the agent's behavioral model: how it reasons, when it acts, when it defers, and how it fails gracefully. I designed the creation and management system — the workflows admins use to configure agents, define skills, attach knowledge sources, and agent guardrails. I also established the safety patterns that formed the product's trust layer, making agent behavior legible and interruptible.

Fig. 02 — The agent's own architecture: four building blocks distinct from Dialpad's existing platform capabilities.

The hardest design problems weren't visual. They were behavioral — determining how much autonomy for an agent was appropriate, and how to make that understandable and manageable.

Fig. 03 — Agent creation workflow: define identity, attach knowledge, configure skills and connectors, then test and deploy. Production screens redacted.

Ten months later, we launched to limited customers. The success of the launch led to the org tripling its investment in the agentic team and space.

Section 03 — Scaling

From sole contributor to system steward

With expansion came a fundamentally different challenge. Seven designers joined the team. The platform roadmap grew to include a visual drag-and-drop agent builder, MCP server integrations, automated testing and evaluation, analytics, and versioning — all due within eight months.

My role shifted from builder to multiplier. I structured design ownership across parallel workstreams, established shared interaction patterns to prevent fragmentation, led design reviews, and kept the system coherent under pressure. At the same time, I stayed in the work — taking direct IC ownership of the highest-leverage areas while making sure the team had clarity and support to move fast.

I also produced executive-facing demos used in board presentations, partnered with sales on customer-facing narratives, and ran customer interviews that shaped prioritization. Keeping stakeholders aligned under a compressed timeline was as much a design challenge as anything on screen.

Fig. 04 — DRI structure: managing up to stakeholders, distributing ownership across 6 parallel workstreams, while maintaining IC responsibility for the Proving Ground testing feature.

Section 04 — Design Thinking

Autonomy is only safe when it's transparent.

A consistent thread across both phases was this: as systems become more capable, the designer's job is to preserve human agency — not just enable automation. Every decision about escalation logic, override states, and transparency patterns was made with that in mind.

I apply the same principle internally. Teams move faster when trust is high. I lead through clarity, direct communication, and staying close to the work — not above it.

Section 05 — Outcomes