Real products built with AI — honest about the process, transparent about the tools.
A Garden Maintenance Planner Built with AI-Assisted Development
Tend is a garden maintenance planner — a React progressive web application that tells a gardener exactly what needs doing this month, without searching, without guesswork, and without generic advice. It was built to serve a real gap in a crowded market: almost every plant care app is built around houseplants and identification. Outdoor UK gardeners were largely ignored.
But this case study is not primarily about what Tend does. It is about how it was built — the development methodology, the discipline around AI-assisted engineering, and the iterative process that took a personal tool to a deployable, scalable product across multiple milestone releases in a matter of days.
The single most important decision made in this project was not a technical one. It was a process decision: how to work effectively with AI as a development collaborator rather than a code generator.
AI tools are only as useful as the context they operate within. An AI given a vague prompt produces generic output. An AI given a precisely maintained, always-current picture of a project — its architecture, its decisions, its constraints, its current state — produces work that integrates cleanly and advances the project meaningfully.
The framework achieves this through structured documentation living directly inside the repository — PROJECT.md, STATE.md, ROADMAP.md, granular phase plans — always authoritative, always current, always in version control alongside the code it describes.
Ideas and planning happen in a dedicated Claude.ai project space, accessible from any device — including a phone. Good thinking does not always happen at a keyboard. The planning environment makes that thinking productive and preserves it.
Active development happens in Claude Code, working directly inside the local GitHub repository. The AI reads the actual repo files, makes changes within established patterns, runs builds, and commits. This separation enforces a discipline that mirrors good engineering practice: design before implementation, specification before code, validation before commitment.
Before a line of product code was written, the competitive landscape was mapped in detail. Every major plant care app was analysed — pricing, features, user reviews, revenue estimates, and the specific complaints that drove negative ratings. That research directly shaped architectural decisions.
The finding that free tiers in competitor apps are frequently described as "useless" led to a hard constraint: Tend's free tier must be genuinely useful, and its API costs must be less than £0.001 per user per month. A product insight became an engineering constraint became a set of implementation patterns, all traceable back to primary research.
Work is organised into milestones, each scoped to a single coherent theme. Each milestone is broken into phases, each phase into numbered plans, each plan into specific tasks. This granularity makes the AI's work verifiable at every step.
Three milestones shipped across five days: Feature Flags establishing the tiering architecture, Quality & Stability covering data integrity and test scaffolding, and UI Polish & Customer Feedback covering the Plants view and plant data infrastructure. The pace was a direct product of the framework: because context was always current and tasks were always precisely specified, development sessions produced working, committed code rather than exploration and iteration.
The entire application lives in a single file until v2.0. This enforced a discipline: every component had to fit coherently, making the architecture easier to reason about and the AI's context window more useful. No backend until v2.0 meant the product could be validated before infrastructure complexity was introduced. UK English throughout established a consistent voice across code, documentation, and UI.
These constraints were not limitations to work around. They were design tools that produced better outcomes than unconstrained development would have.
Tend at v1.8 is a working, deployed, installable progressive web application with nine distinct views, a 2,004-entry plant suggestion database, a 972-entry static schedule database, AI-powered chat for premium users, and a three-tier feature flag system — all served with no CDN, no managed cloud infrastructure, and no ongoing operational cost beyond hardware already in place.
It is used daily. It solves the problem it was designed to solve. And it has a documented, sequenced path to a multi-user cloud-backed product with native mobile applications — each step planned, each dependency identified, each architectural decision already made and recorded.
What this project demonstrates is not simply the ability to build a React application. It demonstrates a disciplined approach to AI-assisted software development — one that produces consistent results at pace, keeps technical decisions traceable, and scales from a personal tool to a product others can use.
The Context Engineering Framework is the methodology. Tend is the evidence.