Research

How to build with AI agents — and trust the result.

We study what makes agent-assisted engineering reliable and secure, and turn the answers into open tools and standards anyone can use.

Research areas

Reliable agent engineering

Orchestration, decomposition and guardrails that keep AI agents on-task through long, multi-step work — without stopping early or drifting from the spec.

Security of AI-assisted delivery

How agent-written code fails, how to catch it, and how to make secure-by-design the default: IAM, AppSec and authorized red-teaming of what agents produce.

Standards & evaluation

What 'good' means for agent output, and how to measure it: normative playbooks, quality gates and tests that prove behavior instead of vibes.

Developer experience

Making the right way the easy way — tooling that installs standards, keeps them current, and gets out of the developer's way.

Our approach

In the open

Findings and tools are published, not hoarded. Reproducibility is the point.

Grounded in practice

We research by building and shipping — the lab and the products feed each other.

Security-first

If it isn't safe to run, it isn't done. Threat modeling and testing are part of the method.

Follow the research.

We write up what we learn on the blog, and the tools land on GitHub.