I kinda have a weird job.
I used to do a lot of cloud security. Now, most days, I’m securing AI agents, building them to do security work, and building the benchmarks that make sure they’re both secure and effective at their jobs - then leveling up other teams to do the same.
It’s less of a leap than it sounds: an agent is really just a workload that acts with severe amounts of non-determinism, so many of the principles - and the engineering - are the same.
After a decade of cloud conversations, I prefer to build platforms that give developers the autonomy they need while keeping security teams happy. This approach has proven useful enough to scale to some pretty large cloud deployments. I’ve built CMDB and CIEM platforms, earned two patents that solved scaled cloud problems, and helped organizations secure environments spanning 25k+ AWS accounts and 13M+ assets.
The environments I work on tend to break vendors. When off-the-shelf tools couldn’t handle analyzing millions of assets or traversing IAM policies across thousands of accounts, we built our own stuff. Now we’re doing it again for agents (and hopefully doing it well enough). This requires a sometimes exhausting blend of distributed systems, cloud expertise, adversarial thinking and threat modeling, and a willingness to sometimes complain loudly.
I’ve spent time on both sides of the table - building security products at vendors as a principal engineer, and implementing internally around uncommon complexity. I hope this two-sided view helps me understand what actually works in the real world.
Outside the day job, I help organize fwd:cloudsec North America and sit on the CVE AI working group.
Finally, security involves a lot of coordination between security teams, engineering teams, the agents who just want to ship code, and the humans who get exhausted by all of that. I spend my time writing software, designing architectures, and occasionally explaining why “just open port 22 to 0.0.0.0/0” isn’t a great idea.