OttoInvites: party e-vite app built in 60 hours
OttoInvites is a digital invitation platform for kids parties, built in two weeks by a single engineer. It started as an internal experiment: we needed a real product to develop and battle-test the AI development harness behind our Agentic Engineering approach.
Challenge
OttoInvites began with two problems running in parallel.
The first was methodological. We wanted to build an AI development harness, a structured environment where AI agents could write production code safely, without breaking things. But you can’t build a harness in a vacuum. We needed a real project with real users, real complexity, and real consequences for bad code. Without that, any guardrails we designed would be theoretical.
The second problem was personal. Pavel Demeshchik, datarockets’ founder, had his kids’ birthday parties coming up, and party planning had always been a source of friction. Coordinating with other parents, tracking RSVPs, remembering names of kids and their parents (and the awkward moments when you forget). Existing tools were either generic invite builders or overkill event platforms. Nothing was designed specifically for the chaos of kids parties.
So we combined both: build a real product that solves a real problem, and use it as the proving ground for our AI-first development process. One engineer. No designers. Two weeks.
Approach
Context engineering and guardrails
The core of the AI harness wasn’t about prompting tricks. It was about context engineering: giving AI agents exactly the right information to make good decisions, and constraining them so they couldn’t make bad ones.
We built a layered documentation system. AGENTS.md defines the project’s architecture, mandatory rules, and verification steps. Separate docs cover the tech stack, design guide, vision, and development approach. Every file is cross-referenced, so an agent working on a UI component automatically gets the design tokens, RTL rules, and accessibility requirements it needs.
The key insight: agents don’t need less freedom, they need better boundaries. The harness doesn’t prevent agents from being creative with solutions. It prevents them from skipping tests, ignoring the design system, or shipping code that doesn’t pass lint and type checks.
TDD as the safety net
Every code change, whether written by a human or an agent, requires tests. Unit tests cover business logic and validation. E2E tests cover user flows from creation to RSVP submission. Smoke tests verify the critical path end-to-end.
This isn’t just a best practice preference. It’s what makes autonomous AI development possible. When an agent implements a feature, the test suite acts as an automated reviewer. If a change breaks an existing flow, the agent knows immediately and fixes it before committing. The tests form the safety net that lets agents move fast without breaking things.
The AGENTS.md file enforces this at the process level: no code gets committed until lint passes, types check, and all tests are green. Agents auto-commit only after full verification.
Building a brand without a designer
One of the bigger risks was design. A kids party app needs to feel fun and trustworthy at the same time. Parents are sharing their children’s information, so the UI has to be polished enough to earn trust, while still capturing the energy of a birthday celebration.
Pavel came up with the creative direction: build the entire brand around a character. Otto, a friendly otter in a party hat who delivers your invitations. The character was created using Quiver AI, an AI vector design tool, and became the foundation for everything: the logo, the mascot, the brand personality.
From there, the design system came together: 9 party themes (Princess, Superhero, Dinosaur, Space, Safari, and more), each with its own color palette, animations, and personality. Framer Motion handles entrance animations and confetti effects. The design guide documents every token, spacing rule, and component pattern so AI agents can implement new UI without drifting from the established aesthetic.
The result is a product that looks like it had a dedicated design team. It didn’t.
Five languages, RTL included
OttoInvites supports English, Arabic, Spanish, Chinese, and Portuguese with full right-to-left layout support. Using next-intl with Next.js App Router, the entire interface, from the creation flow to the RSVP form to the host dashboard, adapts to each language natively.
Arabic RTL support isn’t a CSS afterthought. The design guide mandates logical CSS properties throughout, and every E2E test suite includes bilingual scenarios to catch layout regressions.
From harness to Paperclip
The AI harness we built for OttoInvites proved that agents could ship production code if given proper context and guardrails. But it still required a human to orchestrate the work: deciding what to build next, assigning tasks, reviewing results.
The next step was Paperclip AI, an open-source orchestration platform that coordinates multiple AI agents into functional teams. OttoInvites now runs entirely on Paperclip. Agents implement features, review each other’s code, react to user feedback, check PR comments, and even run self-evaluation: analyzing their own runs to identify improvements to the development environment, documentation, architecture, and tests.
The project went from “one engineer with AI tools” to “a team of agents running autonomously with human oversight.” The harness we built became the foundation that makes this autonomy safe.
Results
Pavel shipped OttoInvites in roughly two weeks, working solo with no other engineers or designers. 133 commits. A production SaaS running on Vercel with a PostgreSQL database on Supabase.
What shipped:
- 9 themed party templates with unique colors, animations, and character
- Real-time RSVP dashboard with Server-Sent Events for live updates
- 5 languages with full RTL support
- QR code generation for printed invites
- OTP authentication for both hosts and guests
- Photo uploads stored on Vercel Blob
- Calendar integration with Google Calendar and ICS downloads
- Comprehensive test suite covering every user flow, enforced by the AI harness
But the bigger outcome wasn’t the product. It was the methodology. The AI harness built for OttoInvites, context engineering, TDD-enforced guardrails, layered documentation, became a blueprint for how we approach every Agentic Engineering project at datarockets. And with Paperclip orchestrating the agents, OttoInvites continues to evolve autonomously, each improvement making the harness stronger for the next project.
