Kirubha Kittusamy
For agents
Principal Product Designer & Builder

I design AI-nativeproducts & systems people & AI agents.

15+ years· Capital One · Google · Walmart · UX / product design · Shipped AI / ML products since 2018
Agentic systems· commerce · developer tools · complex enterprise B2B platforms and mobile apps
Design ↔ code· Design systems · React · APIs · GitHub · AI evaluations and release workflows
GitHub · Last 30 days318+ contributions
Working Sandbox demoBuyer’s AgentIndependent0 → 1

Agentic Commerce - Autonomous Buying

Repeated household purchases led me to explore emerging commerce agents.

My questionHow much purchasing authority should an AI agent actually have?
What I exploredShopify agentic commerce, Muse connector eligibility, UCP, ACP, Google AP2, Visa TAP, and direct retailer APIs.
Designed, built & verifiedA memory-first grocery agent with household purchasing rules. A successful controlled autonomous purchase: WooCommerce demo store & Stripe. A cart-ready checkout experience (no payment executed): Shopify UCP.
💡Production autonomous purchasing requires retailer and payment access.
Read case study →Interactive prototype ↗
Buyer's Agent mobile interface
Buyer's Agent purchase decision interface
Buyer's Agent product decision detail

How Buyer’s Agent works

RememberHousehold preferences + confirmed purchase history
Authorize
AUTHORIZED · ASK · BLOCK
ExecuteBuy within rules · Verify · Record
Capital One SoftwareContractAgentic AIEnterprise cloud platform

Context-Aware Agentic Investigations

Slingshot had to help data teams understand what the system knew, what it inferred, what it was doing next—and when a person needed to step back in.

ProblemEnterprise investigations are multi-step, evidence-heavy and expensive when the system’s reasoning is opaque.
My focusDesign around agent state, provenance, next action and human control—not a generic chat interface.
EvidenceWorkflow models, product-state design, AI-assisted investigation flows and production enterprise constraints.
SignalPrincipal-level value comes from shaping behavior and operating model, not only screens.
ImpactMade multi-step investigations easier to follow by surfacing agent state, evidence and points for human intervention.
Read case study →Read article - Capital One ↗
Slingshot enterprise AI investigation interface
ContextLogs · signals · system history · user intent
Agent reasoningHypotheses · next action · supporting evidence
Human controlReview · redirect · verify · recover
OutcomeRoot cause, explanation and next best action
Agentic AIDesign systems

AI-Ready Design Systems

Agentic systems repeatedly surface the same design problems: what the agent is doing, what it is allowed to do, when a person should intervene, and how the experience recovers when confidence or execution breaks down.

ProblemAgent experiences are often designed one screen at a time, creating inconsistent states, permissions and recovery behavior.
My decisionModel agent behavior as reusable system primitives: states, authority, approval, provenance, failure and recovery.
EvidenceShared component contracts, state patterns, permission patterns and interaction guidance that can be used by designers and agents.
Why it mattersThe system turns repeated AI interaction decisions into reusable product infrastructure rather than isolated UI solutions.
Read case study →
Agent state primitives

State is visible.

IdlePlanningActingWaitingBlockedComplete
Authority patterns

Permission is explicit.

ActAskApproveEscalateBlock
Provenance

Evidence stays connected.

Show what source supports the agent’s action.
Recovery + guardrails

Failure is designed.

Retry, redirect, human takeover and understandable boundaries.
More work
2026Soon in Play StoreTournament OSIndependent product · Mobile · 0→1 · Field tested↗2024Assistive AIGoogle / AndroidAI-Assisted Developer Experience · Developer Tools↗2024Recommendation AIWalmartMerchant AI Recommendations · Enterprise commerce↗2022AutomationAmazon PANDAProcess automation · FinTech · Enterprise↗2022Chatbot / ConversationPepsiCoOmnichannel order management · Commerce↗2020Predictive AIOXY — Spatial MLPredictive ML · Geospatial intelligence · Enterprise↗2020Design systemOXY — Design SystemReusable product system · Enterprise platform↗
Interested in roles across AI-native products, agentic systems, developer tools, enterprise platforms and high-trust product experiences.

I'm always up for a good conversation.

Let’s talk →

Previous portfolios: 2000 · 2001 · 2024 · 2025

2026 © Kirubha Kittusamy