A mobile-first AI agent workspace that turns walls of text into actionable UI — approval cards, inline charts, and live multi-step workflows, all on a 6-inch screen.
🔗 Live Demo → agentic.aurimas.io · ← Back to aurimas.io · LinkedIn
Available for hire — Fractional AI Engineer, AI Product Manager, or AI Architect. I design and ship AI-native products end-to-end, from strategy to deployed prototypes. Get in touch →
Agentic Mobile is a functional MVP built to demonstrate expertise in AI-native UX patterns, agentic system design, and rapid prototyping for B2B mobile users.
The core thesis: B2B decision-makers need to unblock workflows from their phone — but today's mobile AI tools just dump text. This prototype makes the case for Generative UI (rendering approval cards, charts, and live workflow progress directly in a conversational feed) as the right pattern for mobile-first B2B.
An end-to-end MVP built with Next.js, OpenAI, and Supabase, deployed on Netlify.
| Pain Point | Impact |
|---|---|
| 📱 Desktop UIs crammed onto mobile | Too many taps, complex navigation, high abandonment |
| 💬 AI chatbots return walls of text | Unreadable on a 6-inch screen, zero actionability |
| ⏳ Workflows stall without approvals | Hours of delay because the decision-maker isn't at their laptop |
| 🔀 Context-switching between apps | Slack → CRM → Sheets → Email — productivity collapse |
The insight: Mobile B2B users don't need information. They need decisions unblocked in a tap or two.
Instead of text replies, the AI renders the right interface for each task:
| User Intent | Traditional Chatbot | Agentic Mobile |
|---|---|---|
| "Approve the Q3 budget" | Returns a paragraph of text | Renders a native Approve / Reject card |
| "Show pipeline velocity" | Dumps a data table | Generates a mini-chart inline in the feed |
| "Run the daily standup report" | "Here's a summary…" | Executes a multi-step agentic workflow with live progress |
| Opens the app cold | Blank screen, cursor blinking | Contextual zero-state suggesting pending tasks |
- 🎯 Generative UI — Every agent response is a renderable component, not a string
- 🎙️ Voice-First Input — Designed for users on the move (roadmap)
- 🧠 Contextual Zero-State — Proactive suggestions based on time and workflow state
This is a prototype, so the numbers below are the metrics the product is built to improve — not measured results.
| Metric | Why it matters |
|---|---|
| ⚡ Decision velocity | Time-to-decision on off-policy quote approvals — hours down to minutes |
| 📲 Mobile engagement | Agent runs per sales leader from their phone, between meetings |
| 🎯 Zero-state engagement | Share of sessions that act on a proactive suggestion |
| 🔁 Retention (D7 / D30) | Voice vs. text adoption over time |
- Bet 1: B2B users prefer generated UI components over text on mobile — explored via the agentic workflow + approval-card pattern
- Bet 2: No-code agents can be managed from a phone — explored via full agent CRUD (create, configure, pause, delete)
- Bet 3: Speed-to-insight beats depth-of-insight on mobile — explored via concise, card-based responses
| Layer | Technology | Why |
|---|---|---|
| Frontend | Next.js 13 · React · TypeScript | Fast initial load, type-safe rapid iteration |
| Styling | Tailwind CSS | Pixel-perfect mobile control, faster than component libs |
| AI | OpenAI API (GPT-4o-mini) | Best cost/quality ratio; streaming-ready |
| Database | Supabase (Postgres) | Instant setup, RLS, real-time — zero backend boilerplate |
| Deploy | Netlify | Git-push deploys via @netlify/plugin-nextjs, zero DevOps |
- Client-side state in this MVP for speed → production would persist to Supabase with optimistic updates
- GPT-4o-mini over GPT-4o → UX pattern matters more than model power at prototype stage
- No auth → scoped out to focus on core agentic UX; production would use Supabase Auth + RLS
| Problem | B2B mobile workflows rarely get AI agent treatment — sales, field ops, and service teams work on mobile but AI tools are desktop-first |
| Approach | Agent-driven mobile MVP for workflow approvals, real-time reporting, and multi-agent orchestration |
| Role | Concept, prototype, user flow design, full-stack implementation |
| What this proves | End-to-end ownership — product strategy, PRD, design, and full-stack implementation in one working prototype |
git clone https://github.com/aurimas13/web_application.git
cd web_application
npm install
npm run devCreate .env.local:
OPENAI_API_KEY=your-openai-api-key
NEXT_PUBLIC_SUPABASE_URL=your-supabase-url
NEXT_PUBLIC_SUPABASE_ANON_KEY=your-supabase-anon-keySupabase is optional — without it, chat still streams; message logging is simply skipped.
Open http://localhost:3000.
MIT