Soph's Plant Kitchen
Building a minimum testable product with AI in the loop — proving speed and craft can coexist.
Sophie Macfie, the creator behind Soph’s Plant Kitchen, came to us with a clear belief: that a protein-first, plant-based lifestyle can empower people to feel confident in their diets.
Her mission is rooted in making plant-based living accessible to all through trusted recipes and an engaged community.
But while she had cultivated this following on social platforms, she faced limitations — immediacy over depth, diluted messaging, and limited monetisation. Her ambition was to scale into a sustainable business on a platform she owned, one that people wouldn’t just use, but pay for, without sacrificing the trust and quality her audience valued.
For us, this partnership and build was a chance to align with a creator whose values reflect our own, while also testing a bigger hypothesis: can AI meaningfully accelerate product development without compromising the craft that defines us? And would this mean eliminating ourselves in the process?
The impact
Following launch, the product demonstrated strong signals of market fit:
84% month one retention, stabilising at 61% by month three
131% growth in the first three months of launch
* Retention significantly exceeds typical benchmarks for paid wellness and subscription apps, where three-month retention averages around 30–35%.
Whilst Soph’s audience arrives highly qualified, these results are indicative of how effectively the product delivers ongoing value and meets their needs robustly.
Users are returning and building the product into their routines, and translating its content into meaningful changes in their daily lives. The experience brings structure, personalisation, and actionability to Soph’s expertise for users.
From discovery to final product shipping, the ustwo team delivered, moving fast, efficiently, and with the insight you would expect from their expertise.
The challenge
The project had two concurrent pathways:
- The business challenge: create an independent, paid product that could cut through the noise of the wellness market.
- The craft challenge: deliver a Minimum Testable Product (MTP) while stress-testing AI across strategy, product, delivery, design, engineering, and QA.
Many wellness products fail because they are either too generic or too complex. For Soph’s Plant Kitchen, success depended on clarity of content and helping people understand protein and fibre in ways that are approachable, usable, and inspiring.
That meant designing features that delivered utility, turning engagement into habit, and building a monetisation model that respected Soph’s community.
AI added a second layer of uncertainty. Tools promised speed and efficiency, but their limits were still untested in this context. Cut corners, and the product could feel flimsy, eroding the very trust Soph had built.
Our process
We had just nine weeks. After a quick bootstrap to align on scope and ways of working, the team moved into a six-week build and wrapped with a stabilisation sprint focused on infrastructure and QA. Progress relied on trust and transparency, both internally and with the client, as well as constant cross-discipline collaboration.
At the heart of the build were four key features:
- Protein calculator using medically reviewed formulas to set personalised daily targets
- Recipe library with filters, nutrition scoring, and a Protein Boost toggle
- Meal planner, intentionally manual-first to keep trust and usability high
- Membership and paywall, integrating Stripe subscriptions and CRM triggers
The stack was pragmatic but modern, ensuring AI was in the loop, not the headline; Vercel for hosting, Strapi for content, Supabase for authentication and database, Stripe for payments, Plausible for analytics, and Customer.io for CRM. This setup balanced speed to market with enough headroom for future growth.
We used v0 to turn Figma screens into working code scaffolds, Cursor to refine and extend those foundations, custom GPTs trained on research and brand to normalise recipe content and act as a design sounding board, and NotebookLM as a living memory of decisions, experiments and transcripts.
Throughout, we documented what truly accelerated us, what only looked finished, and where we needed to slow down deliberately to preserve architecture, accessibility, and design nuance at the standard we are known for. By the end, we’d delivered both an MTP in beta and we had laid the foundations of an AI-enabled Product Ops method grounded in now tested delivery.
Why it matters
This was about using AI pragmatically to build a product with a balance of speed and standards. The tools compressed cycles — getting from concept to prototype faster, keeping backlogs tidier, and formatting content at scale — but the work only became credible when paired with deliberate and expert human oversight.
The breakthrough came from wiring multiple tools into a coherent chain, with documentation and review at every stage, and knowing when to slow down to raise fidelity. For clients, that balance translates into confidence, we can deliver quickly without sacrificing polish or trust.
Looking forward, we see this work as the foundation for a robust, repeatable AI ProductOps process: prompts, QA gates, and workflow checklists that future teams can adopt without reinventing the wheel. Human expertise is still the fundamental component of our builds, and is present every step of the way.
