Challenge accepted: from scratch to the App Store in two weeks

Challenge accepted: from scratch to the App Store in two weeksIn 14 days, Mapsiti was conceived, built and shipped to the App Store as an AI-assisted experiment. AI acted as product owner, coder and strategist; humans owned the structure, the testing and the technical integrity. Guardrails kept 18,000+ lines of SwiftUI and MapKit clean and production-ready. The experiment proved that AI plus PMO discipline can compress months into weeks — but not without lessons.By Denis Zysman — 2025-01-23The challengeIn today’s fast-moving digital landscape, time-to-market is everything. To test the limits of AI-assisted development, our team ran an experiment: develop and deploy a fully functional iOS app to the App Store in less than two weeks.The result was Mapsiti — an itinerary planning companion for iOS Maps that lets users build, save and share multi-stop trips. What would normally have taken months to scope, build and launch was delivered in 14 days. Here is how it was done, and more importantly what we learned along the way.The appApple Maps does not let you save or share an itinerary. Mapsiti is the add-on that does: create, save and share multi-stop trips, then hand off to iOS Maps for instant turn-by-turn directions.

Search and add any place, with notes and photos.

Multi-modal transport: car, walking, public transport, bicycle.

One-tap search for popular points of interest — parks, trails, museums.

The whole trip visualised on a clean, interactive map.

Reorder stops and plan the route.

Launch directions instantly in iOS Maps.

Save and share unlimited itineraries, across devices or with friends.

Seamless iCloud sync, no account needed.

Completely ad-free.

Concept: AI as a sparring partnerThe foundation of Mapsiti was shaped through deep conversations with AI models. AI simulated user interviews, stress-tested travel scenarios and challenged early assumptions. Competitor analysis exposed the gaps in existing itinerary tools and sharpened the value proposition: seamless, ad-free, multi-modal trip planning that plugs straight into iOS Maps.Before a line of code was written, AI was used to:

Review the competition on the App Store and summarise the feature gaps.

Refine the positioning around ad-free simplicity and iCloud sync with no login.

Evaluate feasibility, stress-testing technical approaches with MapKit and SwiftUI.

Run a discovery session — the structured workshop where stakeholders, designers, developers and analysts agree the vision, the goals, the technical feasibility and the roadmap.

AI works best treated as a collaborative sparring partner, not a tool. The sharper the questions, the sharper the idea. Validation is not about confirming your idea is good — it is about disproving every weakness until only a viable strategy remains.The team: four people and a lot of leverageThe team was deliberately small, with AI filling multiple roles:

One AI product owner — not only defining features, prioritising and structuring user stories, but building the app directly inside Cursor. It made constant micro-decisions while developing, shaping the vision in real time. Instead of long specification documents, requirements evolved dynamically, bending or simplifying whenever AI hit a feasibility wall. That removed most of the documentation overhead and gave the scope unusual flexibility, so progress never stalled.

One developer — to establish the architecture, enforce the rules and troubleshoot when AI got stuck or hallucinated. The developer was the final authority on technical direction and structure.

Two testers — to validate builds, run exploratory testing and give quick user-like feedback so the team could iterate fast.

AI drastically reduces role fragmentation: the AI product owner both envisions and builds. But human oversight remains critical to enforce structure and stop the drift into technical debt.Build: the stack and the standards

Cursor IDE, with heavy prompting.

Models: Claude 4 Sonnet, Claude 4.1 Opus, GPT-5, Gemini 2.5 Pro.

18,000 lines of SwiftUI and MapKit generated in two weeks.

Full localisation in six languages, translated with GPT-5.

SwiftData and CKSyncEngine were chosen — with AI’s help — to persist data and sync across devices through iCloud. These modern frameworks, released by Apple in 2023 and 2022, remove a great deal of boilerplate and abstract away the hard parts: exponential retries, disk persistence, conflict handling. The codebase came out significantly simpler and more modular, which is exactly the setup fast AI development needs.Using AI to check the latest frameworks and best practices simplifies development, extends the shelf life of the code and reduces the risk of a team’s expertise quietly going out of date.Project management

Kanban in Jira, with daily stand-ups.

Four crunches — intense late-night pushes — over the two weeks, to keep momentum.

Marketing and launch prep

App icon and visuals co-created with GPT-5, then polished into photorealistic mockups.

Store description and ASO copy refined with AI for maximum visibility.

Automated social campaigns with make.com and RSS.com to build awareness quickly.

An AI-drafted launch strategy, ads, and an assisted App Store nomination.

AI speeds up code and content alike. But without PMO discipline — a clear backlog, daily assessment, controlled crunches — the process collapses into chaos.Control: guardrails and testingThe most important lesson of the build was that AI coding without guardrails spirals into unmanageable complexity fast. To prevent it we set up a strict .cursorrules file, so the AI adhered to the project’s architecture, conventions and testing practices. A few examples:

Code quality: build, test and fix all critical errors before moving on. Refactor any function over 100 lines or class over 500. SwiftLint enforces style.

Architecture: keep feature logic local, with no cross-feature dependencies except through service interfaces. Stick to MVC or MVVM — no ad hoc logic in view controllers. Services are injected, never hard-coded API clients.

Testing and performance: code coverage enabled with a minimum threshold, performance impact measured before merging anything experimental, at least one unit test per feature branch.

Security and UX: no hardcoded API keys, secrets or credentials. HTTPS everywhere. Dark Mode and Dynamic Type from day one.

Those rules struck the right balance: AI could generate large volumes of SwiftUI quickly, but inside a frame that kept it clean, testable and production-ready.AI without guardrails is chaos. Cursor rules supplied the discipline AI lacks, turning raw generation into maintainable software.What we would tell the next team

Align early on the big functionality: core architecture must be defined upfront.

Start with user stories and wireframes, then let AI iterate on layouts and flows.

Set strong AI rules in Cursor — guardrails prevent messy, inconsistent output.

Use AI to check new standards, practices and frameworks: simpler code, longer shelf life.

Let AI take the initiative on layout, but enforce the standards.

The developer sets the structure. Humans own the foundation, AI fills in the details.

Do not die on hills. If AI stalls, escalate to the developer instead of forcing it.

Restart when lost. If AI spirals into hallucinations or a coding maze, reset with tighter constraints.

Stick to the MVP. Ship fast, then refine — profitability is only proven when $1 of marketing returns more than $1 of profit.

Scalability is a separate problemThe MVP approach was ideal for a rapid launch, but real scale takes more than AI speed. Structural foundations — robust server infrastructure, a scalable database design, a deliberate software architecture — are what carry growth, performance and reliability over time.AI generates features quickly; sustainable scaling demands investment in architecture and backend foundations. Without them, apps become fragile, hard to extend, or unable to handle real demand. The path to scale is a balance: AI for delivery speed, disciplined engineering for a core that can support what comes next.Final reflectionMapsiti went from concept to App Store reality in two weeks. But speed alone was not the victory. The real result was proving that AI, strong PMO practice and hard guardrails together deliver not just code, but strategy, design, marketing and launch readiness in parallel.

AI will not replace leadership, product managers or technical teams. But those who harness it with structure, discipline and a plan for scale will outpace those who do not.