Garth Hinkel

CTO. Twenty-odd years building and scaling engineering teams. Using AI to stay on the tools without becoming the bottleneck.

I run technology at EvaluAgent, an AI-powered QA platform for contact centres. Since joining I've scaled the platform and the engineering function around it. We introduced agile ways of working, proper CI/CD, test automation, and automated regional deployments across AWS. Doubled daily active users, grew AI throughput by over 25x, and got ISO 27001 and SOC II certified.

Before that I spent 11 years at Business Systems, working my way from Software Development Manager to Head of Technical Services to CTO. Enterprise comms for trading floors, contact centres, and public sector. PE-backed MBO exec team. I've also done stints in sales engineering and running delivery teams, which means I've sat on both sides of the table. I know what customers actually need, not just what engineering wants to build.

Further back: five years building voice products at Storacall, and before that I was a software engineer at RealConnect in Stellenbosch, South Africa, where I'd done my engineering degree. B.Eng Electrical and Electronics, class of 2000. Moved to the UK and never quite left.

The bit I keep coming back to: my value isn't shipping code. It's understanding what we've got and keeping us pointed at what's coming. AI changed that equation. I can use it to read a codebase, explain options, evaluate architecture decisions, and stay close to the technology without pulling rank or pulling engineers off their work. More CTOs should be doing this. That's what this site is about.

Six months on: did the gains hold?
In March I wrote about what happened when we gave five teams Claude access. Honeymoon numbers are easy. Here's what it looks like once the novelty's worn off.
The stack: how our AI setup actually fits together
A map, not a tutorial. What sits where in our internal AI infrastructure, what each layer does, and how the pieces connect.
What two years of AI transformation taught me about change
Not a victory lap. What surprised me, what took longer than it should have, and why the teams I expected to adopt fastest didn't.
Transcription quality: the hidden variable in conversation AI
Everything downstream depends on the transcript. Scoring, insights, coaching. Get that layer wrong and the clever stuff on top is confidently wrong too.
MCP servers: what they actually are and why I keep building them
Plain English explanation of the Model Context Protocol, for the technical leader who keeps hearing the acronym and hasn't had time to dig in.
all posts →