.png)
A few weeks ago, we flew our entire Reap Direct technical team to Hong Kong with one goal: rebuild the product from scratch. Engineers, designers, QA, and product – all in one room, for one week.
Seven days later, we shipped a completely new dashboard. A ground-up rebuild of the entire frontend – something that would typically take three to six months using traditional development.
The difference? No one on the team wrote code by hand. Every line, day and night, was generated by AI agents. Our engineers reviewed, directed, and quality-controlled – but the code itself was entirely AI-generated.
Here's how we did it, what we learned, and why we think this is the future of building software.
.png)
Why We Decided to Rebuild
Our customers use Reap Direct to manage treasury, issue cards, send payments, and run their teams. The faster and more intuitive that experience is, the less time they spend on operations and the more time they spend building their businesses.
That's the bar we hold ourselves to – and we wanted to raise it. We wanted a dashboard that was significantly faster, easier to navigate, and built to support the next wave of features and capabilities we have planned. We also wanted a frontend ready for AI-powered browsers and agents, which are quickly becoming part of how teams interact with financial tools.
So we set ourselves a challenge: rebuild the entire product in one week, using AI to move at a pace that wouldn't otherwise be possible.
How We Set Up for AI-First Development
The rebuild wasn't about handing a prompt and hoping for the best. We spent real effort designing a development environment where AI agents could be maximally effective – and where quality was enforced automatically.
Opinionated Framework Choices
We deliberately chose frameworks that reduced interpretation room for AI. Tailwind CSS instead of custom stylesheets. Shadcn as a base component library instead of building bespoke components from scratch. Standardised naming conventions with clear semantic structure.
The logic: the more opinionated and constrained the environment, the less room there is for an AI agent to make ambiguous choices. When there's one obvious way to build something, the AI builds it correctly more often – and the output is more consistent across the entire codebase.
Automated Quality Control
We built custom linting rules that validated AI-generated code against our patterns. If the output didn't conform – wrong naming convention, missing internationalisation keys, skipped QA test coverage – it got flagged and fed back into the loop automatically.
This created self-correcting feedback cycles. The AI would generate, the linter would catch issues, and the AI would regenerate – often without any human intervention. Our engineers stepped in for architectural decisions and edge cases, not for line-by-line review.
Design Workflow Changes
Designers shifted their focus entirely to UX journeys and flows rather than crafting custom components. Because we were using a base component library, there was no need to spend time on pixel-perfect component design – the system handled that. Designers could focus on what actually matters to the user: how the product feels to navigate, where friction exists, and how to eliminate it.
What the Team Actually Did for Seven Days
The week was structured as an intensive hackathon. Everyone was co-located – no async handoffs, no waiting for reviews across time zones.
Engineers set up the AI development environment, defined the architecture, wrote the prompts and constraints, and reviewed generated output. They made the high-level decisions – what to build, how to structure it, which patterns to enforce – and let the AI agents handle the volume.
Designers mapped out every user journey in the product and worked directly with engineers to validate flows in real-time as screens came together.
QA ran continuous testing against each new module, feeding issues back into the development loop immediately rather than batching them at the end.
Product prioritized ruthlessly – deciding what made it into the rebuild, what got simplified, and what got cut. Every user pain point we'd collected over the past year was on the table.
The result was a pace of development that none of us had experienced before. Entire sections of the product – treasury, card management, payments, team admin – went from architecture to working implementation within a day or two.
The Results
For Our Customers
The new Reap Direct is materially faster and easier to use:
- Improved overall performance & stability. The new dashboard loads 10x faster across the app, cards, transactions, everything. We fixed 100+ bugs for a more reliable & efficient experience across the board.
- Better treasury visibility and understanding on all your balances and fund movements at a glance.
- Easier fund movement with a streamlined single-page flow and fewer clicks to complete any transfer.
- Create cards now in one step. Spend less time managing cards.
- Simplified payment flow with a built-in document viewer. Paying your vendors got a whole lot easier.
- Balances & Statements is a new ****consolidated page replacing the old repayments view – get a clear overview, see what you owe and act on it in one place.
- Dark mode is now here with a fresh, modern design.
The full breakdown of what changed for customers here.
For Our Engineering Velocity
This is the part that matters most long-term. The rebuild didn't just produce a better product today – it created a foundation for shipping faster going forward:
- Clean, consistent codebase with enforced patterns across every module.
- AI-readable frontend that's structured for the next generation of browser-based agents and automation tools.
- Eliminated technical debt that had been slowing down feature development for months.
We can now build and ship new features on Reap Direct meaningfully faster than before.
What This Means for Our Clients
The rebuild is done, but the real payoff is what comes next. With AI tooling deeply embedded in how we develop, we can ship new features and improvements to Reap Direct significantly faster than before.
For our clients, that means a product that keeps getting better at a pace that matches how fast they move. The same approach that let us rebuild the entire dashboard in a week is now how we build everything – which means less waiting for the features and improvements that matter to your business.
Reap Direct is the business account for stablecoin-enabled businesses. Learn more or log in to see the new dashboard.
