My lil Biz

End-to-end ownership of a live product

My lil Biz brings a freelancer's projects, clients and money together in one place, and reminds them of what matters on its own. Tagline: "One place, zero drama."

I own all of it — research, product definition, design system, interface, and building and running it with an AI coding agent. It's live at my-lil-biz.com as a responsive web app, and I run my own freelance work in it, so every feature gets tested on real projects every day.

Small business, the full set of financial chores

A typical freelancer has a few projects at a time, often not very expensive. Running them still takes the full set of tools: timelines, clients, documents, invoices, expenses, taxes and reminders.

What's on the market falls into two groups:

• Heavy project-management suites built for teams and agencies — too complex and too expensive for one person.
• Single-purpose apps — a to-do list here, an invoicing tool there — which leave freelancers juggling services and keeping the rest in their head.

The result is a constant feeling of "I forgot something": invoices never sent or paid, clients who go quiet unnoticed, a long ramp back after a break.

From research to a running product

1. Competitor research — mapped project-management, CRM and invoicing tools: what each covers, where it's too heavy or expensive for one person, which needs none of them meets.
2. Product definition and JTBD — the core user (a freelance designer), the jobs, the scope of version one. The main scenario: open Overview and see within seconds that everything is under control.
3. Structure and wireframes — sitemap, then low-fidelity wireframes in Figma for every screen, modal and state, on a lightweight wireframe system.
4. Design system — colour, type, spacing and radius tokens and a component library with variants and states.
5. Build and run — implemented with Claude Code, deployed on my own server.

Making money tasks happen without thinking about them

The heart of the product is money, and the design goal was to remove decisions, not add screens.

Invoices that fill themselves in

An invoice pulls in the project, client, rate and finished phases automatically, numbers itself, and exports to PDF in English or German — including VAT handling and the German §19 UStG small-business note. Invoices issued outside the app can be uploaded and still count toward totals.

Status that's worked out, not maintained

Overdue invoices are detected automatically; project status is calculated from its active phases; a client gets a "Long time no see" flag after 60 days without contact. The user doesn't keep the state of their business up to date — the system does.

Reminders that turn into action

"Pings" surface deadlines, overdue invoices, "time to send an invoice" and quiet clients, plus a weekly digest — and any of them becomes a to-do in one click.

Safeguards where mistakes are costly

A project with paid invoices can't be deleted. GDPR data export and account deletion are built in, and each user sees only their own data.

The assistant as a second user of the product

I built an MCP server for My lil Biz, so the workspace can be run straight from an AI assistant (Claude). That turned the AI into a second kind of user, with its own design questions:

• What the agent may do. I exposed the everyday actions — create clients and projects, add and move phases, manage to-dos, write notes, log contacts — and kept money-critical actions in the interface, where the user sees what they're confirming.
• How the user grants access. Connecting goes through OAuth with a consent screen, so the user explicitly decides what the assistant can touch.
• How intent maps to structure. A sentence like "Create a Rebrand project for Acme with a deadline on November 30" has to land in the same data model as the interface, with the same rules — a project still needs a deadline.

Design system to code, with an AI agent as pair

I build the product together with Claude Code, Anthropic's AI coding agent. I own the product, design and architecture decisions; the agent implements screens from the Figma designs, keeps the product spec up to date, reviews code and runs security checks.

The design system is the contract between us. It lives in code as tokens and a UI kit, with a live style guide page in the app, and every visual change goes from design system to UI kit to screen. Pre-commit checks guard the design system, so the agent can't quietly drift from it.

Usage analytics without looking at anyone's data

The admin panel tracks DAU / WAU / MAU, retention and time spent in the service — while the admin never sees users' own data. That's the base for deciding what to build next from behaviour, not from my own habits.

From a personal tool to a commercial product

• Paid plans with built-in billing, priced for solo freelancers, not agencies.
• More AI help — suggested next steps, smart tips and AI-generated reminders, built on the same principle as the MCP server: the AI acts, the user stays in control of money.
• A wider rollout, shaped by usage data and feedback from daily real-world use.