We have never been a company that tries to force engineers to write code quicker. There are only so many times you can hit a key in a day. We have been heavily investing in AI for the last 18 months and like to think that we are very much on top of the current situation. We have almost every model running across a variety of business operations and are constantly tinkering. 

The SDLC is out latest obsession. We asked our engineers to have some fun with AI and see what they could do to increase throughput. We realised we were pretty good at it and since then we've delivered some impressive solutions internally and for our clients. They were fun to build too.

We have leant into an entirely new development process - called Spec Driven Development - and we are seeing some really impressive results. We believe it is the 

 to set up an SDLC  for the next decade. It is not vibe coding, and it relies on multiple quality gates manned by real human beings :D

The spec is the asset we maintain. 3 inputs stay live at all times.

- A specification that describes the system as it actually is now, not as someone described it eighteen months ago. It is updated and monitored to ensure it always represents real life.

- An approved design library, pulled straight out of Figma via MCP. A clear, up to date and definitive set of designs.

- The API spec, either an OpenAPI schema or the live APIs themselves. 

The more detailed, and precise the "spec" is the more successful the Agentic delivery pipeline will be. 

The above then gets consumed by a delivery pipeline that’s human owned, and Agent powered.  There are loosely 4 steps. Engineering led spec authoring to describe a feature. Agents generate boilerplate, typed models, UI scaffolds and unit test suites, monitored, owned by an engineer. CodeRabbit takes a first review pass, then two engineers review before anything merges to main, and every place AI touches the code is flagged in the PR. QA, automation and regression testing runs and final sign off is given by an engineer.

A clear Definition of Ready sits at  the start, and a clear Definition of Done at end provide critical quality gates.

 autonomous and we don't pretend it is. Someone is accountable at every gate and Catalyst is ultimately responsible to mange the pipeline the same way we manage a human pipeline, and to ensure production assets operate with enterprise level quality at huge scale within local and international retail. 

Each pipeline is built a custom, based on the project, the client, the inputs and what the end goal is. Engineers are now more focused on input quality, managing the team and optimising the delivery pipeline for speed and quality. The better this Human-AI team is, the more value delivered. 

- Each feature costs less than the last one did. Improve the pipeline and every future build inherits it, so the gains stack up over time instead of resetting on each project.

- We don't have technology discussions anymore. React Native versus native. Flutter v React. Node Golang. Code is an output now. If the right answer changes in three years we change with it, and it isn't a complete rebuild. We can now build the best front end for the customer, and the business logic is maintained in one place via the specification assets.

- You see results sooner and can be more dynamic and responsive with roadamps. 

- And the documentation is true, relevant and a real representation of the production landscape. That sounds like a small thing but it enables new engineers to get productive fast. The knowledge sits with the business, not a key man.

When we build from scratch, this is what we use. For existing projects, we are actively moving toward Agentic Spec Driven Development. 

It is early stages, but starting now means you learn the lessons, gain the experience and compound the delivery before anyone else. 

We believe this is going to be a strategic advantage for our clients, and ourselves. 

Overview.pdf

How Agents are used in development

We have an incredibly exciting partnership with Pick n Pay, helping them bring their online vision to reality. We recently caught up with Tech Central around our role in delivering Penny to millions of customers via the asap! app. 

Penny is an AI-powered conversational shopping assistant launched by Pick n Pay inside its asap! on-demand grocery delivery app. Built using Google’s Gemini models, it allows users to build custom carts using text prompts, voice notes, or photos (like handwritten lists or fridge contents) rather than manual searching.

https://techcentral.co.za/meet-the-sa-software-house-behind-pick-n-pays-penny/283664/

Blog

How Agents are used in development

Catalyst is in the news!