Aitopia.

Overview

Aitopia builds AI-native productivity tools for people who actually use AI in their day. As founding engineer, I’ve owned big chunks of the frontend and product engineering — chat surfaces, agent flows, browser-extension tooling, and the bridges between them.

Problem

Most AI tools either bury intelligence in chat or sprinkle it on top of unrelated UI. Neither feels like a coherent product. The challenge was building surfaces where AI is woven through the workflow, not bolted on.

Constraints

• Cross-platform reach — web, mobile, browser extension
• Latency budget tight enough to feel instant on consumer hardware
• Small team, fast cadence, real users from day one
• Provider neutrality — never bound to a single LLM vendor

Approach

Treat AI features as first-class product surfaces, not modals. Define the interaction first, then pick the model. Keep the orchestration layer thin and provider-agnostic so we can swap behind the scenes without product churn.

Implementation

React + React Native shared product layer with platform-specific UI shells. Zustand for state, Supabase for persistence and auth, a typed orchestration layer over OpenAI and Anthropic with structured outputs. CI, telemetry, and feature flags so we can ship daily without breaking trust.

Result

The product ships continuously to a real user base, with the AI features becoming the reason people stay, not a side dish. Internally, we can move from product idea to shipped surface in days because the system stops fighting us.

Reflection

The biggest unlock was deciding early that “AI feature” isn’t a category — it’s a property of any feature. Once that shifted, scope decisions, naming, and architecture all got cleaner.