Case study codi

Creating a cutting-edge AI build agent for experimentation

Technologies

The tech stack we used to build this project

Kubernetes

OpenAI

LangChain

Anthropic

TypeScript

Docker

Project summary

TopFlight created Codi, our own AI build agent for experimentation. Drawing on hundreds of real-world A/B tests, we turned specialist engineering expertise into a product that helps product, CRO and development teams build working variations from prompts, ready for review and export.

Contact

The problem

Good hypotheses were stuck in backlogs. Every experiment competed for scarce engineering time, slowing learning and delaying potential revenue gains. Teams had the testing platforms and ideas, but lacked the capacity to get enough experiments built.

The challenge

Build an AI product that understands real websites and the engineering behind reliable experiments. Codi needed to handle dynamic pages, changing states and complex interactions, while producing editable code that fits existing testing platforms and approval processes.

The services

We shaped the product around delivery bottlenecks, refining workflows through internal use and customer pilots. Live-page context, agent orchestration and TopFlight’s experimentation SDK are combined to generate and refine variations, backed by browser-based checks and repeatable benchmarks that assess behaviour, build quality, speed and AI costs. We delivered a desktop application, a browser sandbox and an API for partner integrations.

Related case studies

Empowering engineering excellence in innovation and experimentation case study

Centre of excellence

Warranty & insurance

Empowering engineering excellence in innovation and experimentation

Over a four-year partnership, we've brought rapid engineering delivery into Domestic and General to allow them to innovate and experiment at scale across all their core journeys

From a stalled website to direct bookings in six weeks case study

Speed to market

Hospitality and real estate

From a stalled website to direct bookings in six weeks

We launched a bespoke booking platform for Vacationer Dubai in six weeks, now generating 20-30+ direct bookings and approximately £60,000 in revenue per month, and keep improving it through Statsig experiments.

Establishing engineering excellence in experimentation case study

Centre of excellence

Digital experimentation agency

Establishing engineering excellence in experimentation

We’ve been partnered with Creative CX for over two years, accelerating their experimentation delivery across multiple enterprise clients. By bringing rapid A/B testing development and solid engineering standards, we’ve helped scale their experimentation output and improve test quality across squads.

Speed to market

Education

Optimising education processes using AI and MVP experiments

We helped Learning DNA turn manual assessment and reporting services into digital products, scaling it to serve 300+ schools and become one of the UK’s fastest-growing EdTech start-ups.

Building a dental optimisation hub to maximise revenue with integrated reporting intelligence case study

Integration

Dental healthcare

Building a dental optimisation hub to maximise revenue with integrated reporting intelligence

We built a bespoke operations hub for Alexandra Dental, a group of 40+ Irish clinics, bringing together reporting, appointment capacity, call activity and patient follow-up in one live platform.

Travel

Practical AI for complex safari operations

We built a bespoke AI hub for Yellow Zebra Safaris, combining an itinerary checker with a CRM-connected customer assistant and saving hundreds of hours of manual work.

Creating a cutting-edge AI build agent for experimentation case study

Product

AI and technology