The projects nobody had a product for kept landing on my desk.
Ten years in banking and asset management. Fourteen years at PwC consulting across cybersecurity, digital forensics, risk, regulatory and financial services. Mostly deals and regulatory compliance, where the answer did not exist yet and somebody was waiting to audit the one you built.
I build enterprise systems for sensitive, regulated data.
Work out what is allowed. That means sitting with regulators and lawyers. Work out what can be built. Then build it so it gives the same answer every time. It has to, before the deal closes. The other side will check it, and so will the government.
It started with small teams I pulled together, from inside PwC and out, for problems no product could handle.
Scale was the other half. Hundreds of millions of files, spread across countries and the people who held them, inside the window before a deal closed or a deadline landed. Cloud was the only way to reach that, so we engineered the systems ourselves: architecture, data model, pipelines. Where the platforms could not carry what we put through them, I worked directly with Google and AWS engineers to change how their services behaved.
The U.S. Treasury asked to see one of them run, live. It was built under a CFIUS national security agreement. They said they had not seen technology layered that way, or cover that much.
I founded Octosight to build the tool my teams and clients always needed.I created Runtime to build it with agents.
Octosight protects large volumes of sensitive data with AI. I co-founded it in 2023, and my former team left their jobs to come and build it. What changed was moving the work onto coding agents to go faster. By late 2025, with Claude Code and Codex, the speed was real. Knowing what we had was not.
I started building the Runtime components so my teams could go faster.
Delivering one of those took years of learning how to run the people who build them. Architects, engineers, pen testers, product owners. Different instincts, different ways of being managed, all of it pointed at the date a multibillion dollar deal had to close.
That work now involves a fleet of agents with their own personality quirks. The job has not changed. Get the best out of each one, and keep all of it pointed at the same date. The way I have always known whether it is going well is to ask, and to be shown. Agents do not answer that way on their own. So the harness asks for me, and it keeps the evidence behind every answer.
The first Runtime capability started as frustration. Hours a day going into work that should have been one repeatable task, and no tool that would do it. So I built one. Then another. Before long there was enough there to make a harness out of them, and enough to be worth releasing.
None of this was learned alone. The documentation and the courses, hackathons, and the people who post their problem and their fix instead of keeping it. Everyone is working this out at the same time, in the open. Runtime goes out as open source because that is how I got here.
MeerkatOps is where I put the tools I build for my teams to work better with agents, whether that is running a startup or building for an enterprise. Three of them so far: building the product, the security operations around it, and the compliance evidence underneath.
I volunteer as a firefighter, and I do woodworking and mill lumber with a chainsaw with my kids. The rest of my life happens on a screen. This is where they learn the value of building things with their hands and being a useful human.